net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/atomic.h>
  21 #include <linux/bpf_verifier.h>
  22 #include <linux/module.h>
  23 #include <linux/types.h>
  24 #include <linux/mm.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/socket.h>
  27 #include <linux/sock_diag.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/if_packet.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/gfp.h>
  34 #include <net/inet_common.h>
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/netlink.h>
  38 #include <linux/skbuff.h>
  39 #include <linux/skmsg.h>
  40 #include <net/sock.h>
  41 #include <net/flow_dissector.h>
  42 #include <linux/errno.h>
  43 #include <linux/timer.h>
  44 #include <linux/uaccess.h>
  45 #include <asm/unaligned.h>
  46 #include <linux/filter.h>
  47 #include <linux/ratelimit.h>
  48 #include <linux/seccomp.h>
  49 #include <linux/if_vlan.h>
  50 #include <linux/bpf.h>
  51 #include <linux/btf.h>
  52 #include <net/sch_generic.h>
  53 #include <net/cls_cgroup.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/dst.h>
  56 #include <net/sock_reuseport.h>
  57 #include <net/busy_poll.h>
  58 #include <net/tcp.h>
  59 #include <net/xfrm.h>
  60 #include <net/udp.h>
  61 #include <linux/bpf_trace.h>
  62 #include <net/xdp_sock.h>
  63 #include <linux/inetdevice.h>
  64 #include <net/inet_hashtables.h>
  65 #include <net/inet6_hashtables.h>
  66 #include <net/ip_fib.h>
  67 #include <net/nexthop.h>
  68 #include <net/flow.h>
  69 #include <net/arp.h>
  70 #include <net/ipv6.h>
  71 #include <net/net_namespace.h>
  72 #include <linux/seg6_local.h>
  73 #include <net/seg6.h>
  74 #include <net/seg6_local.h>
  75 #include <net/lwtunnel.h>
  76 #include <net/ipv6_stubs.h>
  77 #include <net/bpf_sk_storage.h>
  78 #include <net/transp_v6.h>
  79 #include <linux/btf_ids.h>
  80 #include <net/tls.h>
  81 #include <net/xdp.h>
  82 #include <net/mptcp.h>
  83 #include <net/netfilter/nf_conntrack_bpf.h>
  84
  85 static const struct bpf_func_proto *
  86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  87
  88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  89 {
  90         if (in_compat_syscall()) {
  91                 struct compat_sock_fprog f32;
  92
  93                 if (len != sizeof(f32))
  94                         return -EINVAL;
  95                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  96                         return -EFAULT;
  97                 memset(dst, 0, sizeof(*dst));
  98                 dst->len = f32.len;
  99                 dst->filter = compat_ptr(f32.filter);
 100         } else {
 101                 if (len != sizeof(*dst))
 102                         return -EINVAL;
 103                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 104                         return -EFAULT;
 105         }
 106
 107         return 0;
 108 }
 109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 110
 111 /**
 112  *      sk_filter_trim_cap - run a packet through a socket filter
 113  *      @sk: sock associated with &sk_buff
 114  *      @skb: buffer to filter
 115  *      @cap: limit on how short the eBPF program may trim the packet
 116  *
 117  * Run the eBPF program and then cut skb->data to correct size returned by
 118  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 119  * than pkt_len we keep whole skb->data. This is the socket level
 120  * wrapper to bpf_prog_run. It returns 0 if the packet should
 121  * be accepted or -EPERM if the packet should be tossed.
 122  *
 123  */
 124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 125 {
 126         int err;
 127         struct sk_filter *filter;
 128
 129         /*
 130          * If the skb was allocated from pfmemalloc reserves, only
 131          * allow SOCK_MEMALLOC sockets to use it as this socket is
 132          * helping free memory
 133          */
 134         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 135                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 136                 return -ENOMEM;
 137         }
 138         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 139         if (err)
 140                 return err;
 141
 142         err = security_sock_rcv_skb(sk, skb);
 143         if (err)
 144                 return err;
 145
 146         rcu_read_lock();
 147         filter = rcu_dereference(sk->sk_filter);
 148         if (filter) {
 149                 struct sock *save_sk = skb->sk;
 150                 unsigned int pkt_len;
 151
 152                 skb->sk = sk;
 153                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 154                 skb->sk = save_sk;
 155                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 156         }
 157         rcu_read_unlock();
 158
 159         return err;
 160 }
 161 EXPORT_SYMBOL(sk_filter_trim_cap);
 162
 163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 164 {
 165         return skb_get_poff(skb);
 166 }
 167
 168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 169 {
 170         struct nlattr *nla;
 171
 172         if (skb_is_nonlinear(skb))
 173                 return 0;
 174
 175         if (skb->len < sizeof(struct nlattr))
 176                 return 0;
 177
 178         if (a > skb->len - sizeof(struct nlattr))
 179                 return 0;
 180
 181         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 182         if (nla)
 183                 return (void *) nla - (void *) skb->data;
 184
 185         return 0;
 186 }
 187
 188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 189 {
 190         struct nlattr *nla;
 191
 192         if (skb_is_nonlinear(skb))
 193                 return 0;
 194
 195         if (skb->len < sizeof(struct nlattr))
 196                 return 0;
 197
 198         if (a > skb->len - sizeof(struct nlattr))
 199                 return 0;
 200
 201         nla = (struct nlattr *) &skb->data[a];
 202         if (nla->nla_len > skb->len - a)
 203                 return 0;
 204
 205         nla = nla_find_nested(nla, x);
 206         if (nla)
 207                 return (void *) nla - (void *) skb->data;
 208
 209         return 0;
 210 }
 211
 212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 213            data, int, headlen, int, offset)
 214 {
 215         u8 tmp, *ptr;
 216         const int len = sizeof(tmp);
 217
 218         if (offset >= 0) {
 219                 if (headlen - offset >= len)
 220                         return *(u8 *)(data + offset);
 221                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 222                         return tmp;
 223         } else {
 224                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 225                 if (likely(ptr))
 226                         return *(u8 *)ptr;
 227         }
 228
 229         return -EFAULT;
 230 }
 231
 232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 233            int, offset)
 234 {
 235         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 236                                          offset);
 237 }
 238
 239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 240            data, int, headlen, int, offset)
 241 {
 242         __be16 tmp, *ptr;
 243         const int len = sizeof(tmp);
 244
 245         if (offset >= 0) {
 246                 if (headlen - offset >= len)
 247                         return get_unaligned_be16(data + offset);
 248                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 249                         return be16_to_cpu(tmp);
 250         } else {
 251                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 252                 if (likely(ptr))
 253                         return get_unaligned_be16(ptr);
 254         }
 255
 256         return -EFAULT;
 257 }
 258
 259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 260            int, offset)
 261 {
 262         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 263                                           offset);
 264 }
 265
 266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 267            data, int, headlen, int, offset)
 268 {
 269         __be32 tmp, *ptr;
 270         const int len = sizeof(tmp);
 271
 272         if (likely(offset >= 0)) {
 273                 if (headlen - offset >= len)
 274                         return get_unaligned_be32(data + offset);
 275                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 276                         return be32_to_cpu(tmp);
 277         } else {
 278                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 279                 if (likely(ptr))
 280                         return get_unaligned_be32(ptr);
 281         }
 282
 283         return -EFAULT;
 284 }
 285
 286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 287            int, offset)
 288 {
 289         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 290                                           offset);
 291 }
 292
 293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 294                               struct bpf_insn *insn_buf)
 295 {
 296         struct bpf_insn *insn = insn_buf;
 297
 298         switch (skb_field) {
 299         case SKF_AD_MARK:
 300                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 301
 302                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 303                                       offsetof(struct sk_buff, mark));
 304                 break;
 305
 306         case SKF_AD_PKTTYPE:
 307                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
 308                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 309 #ifdef __BIG_ENDIAN_BITFIELD
 310                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 311 #endif
 312                 break;
 313
 314         case SKF_AD_QUEUE:
 315                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 316
 317                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 318                                       offsetof(struct sk_buff, queue_mapping));
 319                 break;
 320
 321         case SKF_AD_VLAN_TAG:
 322                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 323
 324                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 325                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 326                                       offsetof(struct sk_buff, vlan_tci));
 327                 break;
 328         case SKF_AD_VLAN_TAG_PRESENT:
 329                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
 330                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 331                                       offsetof(struct sk_buff, vlan_all));
 332                 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
 333                 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
 334                 break;
 335         }
 336
 337         return insn - insn_buf;
 338 }
 339
 340 static bool convert_bpf_extensions(struct sock_filter *fp,
 341                                    struct bpf_insn **insnp)
 342 {
 343         struct bpf_insn *insn = *insnp;
 344         u32 cnt;
 345
 346         switch (fp->k) {
 347         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 348                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 349
 350                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 351                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 352                                       offsetof(struct sk_buff, protocol));
 353                 /* A = ntohs(A) [emitting a nop or swap16] */
 354                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 355                 break;
 356
 357         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 358                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 359                 insn += cnt - 1;
 360                 break;
 361
 362         case SKF_AD_OFF + SKF_AD_IFINDEX:
 363         case SKF_AD_OFF + SKF_AD_HATYPE:
 364                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 365                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 366
 367                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 368                                       BPF_REG_TMP, BPF_REG_CTX,
 369                                       offsetof(struct sk_buff, dev));
 370                 /* if (tmp != 0) goto pc + 1 */
 371                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 372                 *insn++ = BPF_EXIT_INSN();
 373                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 374                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 375                                             offsetof(struct net_device, ifindex));
 376                 else
 377                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 378                                             offsetof(struct net_device, type));
 379                 break;
 380
 381         case SKF_AD_OFF + SKF_AD_MARK:
 382                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 383                 insn += cnt - 1;
 384                 break;
 385
 386         case SKF_AD_OFF + SKF_AD_RXHASH:
 387                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 388
 389                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 390                                     offsetof(struct sk_buff, hash));
 391                 break;
 392
 393         case SKF_AD_OFF + SKF_AD_QUEUE:
 394                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 395                 insn += cnt - 1;
 396                 break;
 397
 398         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 399                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 400                                          BPF_REG_A, BPF_REG_CTX, insn);
 401                 insn += cnt - 1;
 402                 break;
 403
 404         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 405                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 406                                          BPF_REG_A, BPF_REG_CTX, insn);
 407                 insn += cnt - 1;
 408                 break;
 409
 410         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 411                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 412
 413                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 414                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 415                                       offsetof(struct sk_buff, vlan_proto));
 416                 /* A = ntohs(A) [emitting a nop or swap16] */
 417                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 418                 break;
 419
 420         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 421         case SKF_AD_OFF + SKF_AD_NLATTR:
 422         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 423         case SKF_AD_OFF + SKF_AD_CPU:
 424         case SKF_AD_OFF + SKF_AD_RANDOM:
 425                 /* arg1 = CTX */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 427                 /* arg2 = A */
 428                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 429                 /* arg3 = X */
 430                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 431                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 432                 switch (fp->k) {
 433                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 434                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 435                         break;
 436                 case SKF_AD_OFF + SKF_AD_NLATTR:
 437                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 438                         break;
 439                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 440                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 441                         break;
 442                 case SKF_AD_OFF + SKF_AD_CPU:
 443                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 444                         break;
 445                 case SKF_AD_OFF + SKF_AD_RANDOM:
 446                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 447                         bpf_user_rnd_init_once();
 448                         break;
 449                 }
 450                 break;
 451
 452         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 453                 /* A ^= X */
 454                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 455                 break;
 456
 457         default:
 458                 /* This is just a dummy call to avoid letting the compiler
 459                  * evict __bpf_call_base() as an optimization. Placed here
 460                  * where no-one bothers.
 461                  */
 462                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 463                 return false;
 464         }
 465
 466         *insnp = insn;
 467         return true;
 468 }
 469
 470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 471 {
 472         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 473         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 474         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 475                       BPF_SIZE(fp->code) == BPF_W;
 476         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 477         const int ip_align = NET_IP_ALIGN;
 478         struct bpf_insn *insn = *insnp;
 479         int offset = fp->k;
 480
 481         if (!indirect &&
 482             ((unaligned_ok && offset >= 0) ||
 483              (!unaligned_ok && offset >= 0 &&
 484               offset + ip_align >= 0 &&
 485               offset + ip_align % size == 0))) {
 486                 bool ldx_off_ok = offset <= S16_MAX;
 487
 488                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 489                 if (offset)
 490                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 491                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 492                                       size, 2 + endian + (!ldx_off_ok * 2));
 493                 if (ldx_off_ok) {
 494                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 495                                               BPF_REG_D, offset);
 496                 } else {
 497                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 498                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 499                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 500                                               BPF_REG_TMP, 0);
 501                 }
 502                 if (endian)
 503                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 504                 *insn++ = BPF_JMP_A(8);
 505         }
 506
 507         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 508         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 509         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 510         if (!indirect) {
 511                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 512         } else {
 513                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 514                 if (fp->k)
 515                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 516         }
 517
 518         switch (BPF_SIZE(fp->code)) {
 519         case BPF_B:
 520                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 521                 break;
 522         case BPF_H:
 523                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 524                 break;
 525         case BPF_W:
 526                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 527                 break;
 528         default:
 529                 return false;
 530         }
 531
 532         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 533         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 534         *insn   = BPF_EXIT_INSN();
 535
 536         *insnp = insn;
 537         return true;
 538 }
 539
 540 /**
 541  *      bpf_convert_filter - convert filter program
 542  *      @prog: the user passed filter program
 543  *      @len: the length of the user passed filter program
 544  *      @new_prog: allocated 'struct bpf_prog' or NULL
 545  *      @new_len: pointer to store length of converted program
 546  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 547  *
 548  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 549  * style extended BPF (eBPF).
 550  * Conversion workflow:
 551  *
 552  * 1) First pass for calculating the new program length:
 553  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 554  *
 555  * 2) 2nd pass to remap in two passes: 1st pass finds new
 556  *    jump offsets, 2nd pass remapping:
 557  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 558  */
 559 static int bpf_convert_filter(struct sock_filter *prog, int len,
 560                               struct bpf_prog *new_prog, int *new_len,
 561                               bool *seen_ld_abs)
 562 {
 563         int new_flen = 0, pass = 0, target, i, stack_off;
 564         struct bpf_insn *new_insn, *first_insn = NULL;
 565         struct sock_filter *fp;
 566         int *addrs = NULL;
 567         u8 bpf_src;
 568
 569         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 570         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 571
 572         if (len <= 0 || len > BPF_MAXINSNS)
 573                 return -EINVAL;
 574
 575         if (new_prog) {
 576                 first_insn = new_prog->insnsi;
 577                 addrs = kcalloc(len, sizeof(*addrs),
 578                                 GFP_KERNEL | __GFP_NOWARN);
 579                 if (!addrs)
 580                         return -ENOMEM;
 581         }
 582
 583 do_pass:
 584         new_insn = first_insn;
 585         fp = prog;
 586
 587         /* Classic BPF related prologue emission. */
 588         if (new_prog) {
 589                 /* Classic BPF expects A and X to be reset first. These need
 590                  * to be guaranteed to be the first two instructions.
 591                  */
 592                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 593                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 594
 595                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 596                  * In eBPF case it's done by the compiler, here we need to
 597                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 598                  */
 599                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 600                 if (*seen_ld_abs) {
 601                         /* For packet access in classic BPF, cache skb->data
 602                          * in callee-saved BPF R8 and skb->len - skb->data_len
 603                          * (headlen) in BPF R9. Since classic BPF is read-only
 604                          * on CTX, we only need to cache it once.
 605                          */
 606                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 607                                                   BPF_REG_D, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, data));
 609                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 610                                                   offsetof(struct sk_buff, len));
 611                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 612                                                   offsetof(struct sk_buff, data_len));
 613                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 614                 }
 615         } else {
 616                 new_insn += 3;
 617         }
 618
 619         for (i = 0; i < len; fp++, i++) {
 620                 struct bpf_insn tmp_insns[32] = { };
 621                 struct bpf_insn *insn = tmp_insns;
 622
 623                 if (addrs)
 624                         addrs[i] = new_insn - first_insn;
 625
 626                 switch (fp->code) {
 627                 /* All arithmetic insns and skb loads map as-is. */
 628                 case BPF_ALU | BPF_ADD | BPF_X:
 629                 case BPF_ALU | BPF_ADD | BPF_K:
 630                 case BPF_ALU | BPF_SUB | BPF_X:
 631                 case BPF_ALU | BPF_SUB | BPF_K:
 632                 case BPF_ALU | BPF_AND | BPF_X:
 633                 case BPF_ALU | BPF_AND | BPF_K:
 634                 case BPF_ALU | BPF_OR | BPF_X:
 635                 case BPF_ALU | BPF_OR | BPF_K:
 636                 case BPF_ALU | BPF_LSH | BPF_X:
 637                 case BPF_ALU | BPF_LSH | BPF_K:
 638                 case BPF_ALU | BPF_RSH | BPF_X:
 639                 case BPF_ALU | BPF_RSH | BPF_K:
 640                 case BPF_ALU | BPF_XOR | BPF_X:
 641                 case BPF_ALU | BPF_XOR | BPF_K:
 642                 case BPF_ALU | BPF_MUL | BPF_X:
 643                 case BPF_ALU | BPF_MUL | BPF_K:
 644                 case BPF_ALU | BPF_DIV | BPF_X:
 645                 case BPF_ALU | BPF_DIV | BPF_K:
 646                 case BPF_ALU | BPF_MOD | BPF_X:
 647                 case BPF_ALU | BPF_MOD | BPF_K:
 648                 case BPF_ALU | BPF_NEG:
 649                 case BPF_LD | BPF_ABS | BPF_W:
 650                 case BPF_LD | BPF_ABS | BPF_H:
 651                 case BPF_LD | BPF_ABS | BPF_B:
 652                 case BPF_LD | BPF_IND | BPF_W:
 653                 case BPF_LD | BPF_IND | BPF_H:
 654                 case BPF_LD | BPF_IND | BPF_B:
 655                         /* Check for overloaded BPF extension and
 656                          * directly convert it if found, otherwise
 657                          * just move on with mapping.
 658                          */
 659                         if (BPF_CLASS(fp->code) == BPF_LD &&
 660                             BPF_MODE(fp->code) == BPF_ABS &&
 661                             convert_bpf_extensions(fp, &insn))
 662                                 break;
 663                         if (BPF_CLASS(fp->code) == BPF_LD &&
 664                             convert_bpf_ld_abs(fp, &insn)) {
 665                                 *seen_ld_abs = true;
 666                                 break;
 667                         }
 668
 669                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 670                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 671                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 672                                 /* Error with exception code on div/mod by 0.
 673                                  * For cBPF programs, this was always return 0.
 674                                  */
 675                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 676                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 677                                 *insn++ = BPF_EXIT_INSN();
 678                         }
 679
 680                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 681                         break;
 682
 683                 /* Jump transformation cannot use BPF block macros
 684                  * everywhere as offset calculation and target updates
 685                  * require a bit more work than the rest, i.e. jump
 686                  * opcodes map as-is, but offsets need adjustment.
 687                  */
 688
 689 #define BPF_EMIT_JMP                                                    \
 690         do {                                                            \
 691                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 692                 s32 off;                                                \
 693                                                                         \
 694                 if (target >= len || target < 0)                        \
 695                         goto err;                                       \
 696                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 697                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 698                 off -= insn - tmp_insns;                                \
 699                 /* Reject anything not fitting into insn->off. */       \
 700                 if (off < off_min || off > off_max)                     \
 701                         goto err;                                       \
 702                 insn->off = off;                                        \
 703         } while (0)
 704
 705                 case BPF_JMP | BPF_JA:
 706                         target = i + fp->k + 1;
 707                         insn->code = fp->code;
 708                         BPF_EMIT_JMP;
 709                         break;
 710
 711                 case BPF_JMP | BPF_JEQ | BPF_K:
 712                 case BPF_JMP | BPF_JEQ | BPF_X:
 713                 case BPF_JMP | BPF_JSET | BPF_K:
 714                 case BPF_JMP | BPF_JSET | BPF_X:
 715                 case BPF_JMP | BPF_JGT | BPF_K:
 716                 case BPF_JMP | BPF_JGT | BPF_X:
 717                 case BPF_JMP | BPF_JGE | BPF_K:
 718                 case BPF_JMP | BPF_JGE | BPF_X:
 719                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 720                                 /* BPF immediates are signed, zero extend
 721                                  * immediate into tmp register and use it
 722                                  * in compare insn.
 723                                  */
 724                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 725
 726                                 insn->dst_reg = BPF_REG_A;
 727                                 insn->src_reg = BPF_REG_TMP;
 728                                 bpf_src = BPF_X;
 729                         } else {
 730                                 insn->dst_reg = BPF_REG_A;
 731                                 insn->imm = fp->k;
 732                                 bpf_src = BPF_SRC(fp->code);
 733                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 734                         }
 735
 736                         /* Common case where 'jump_false' is next insn. */
 737                         if (fp->jf == 0) {
 738                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 739                                 target = i + fp->jt + 1;
 740                                 BPF_EMIT_JMP;
 741                                 break;
 742                         }
 743
 744                         /* Convert some jumps when 'jump_true' is next insn. */
 745                         if (fp->jt == 0) {
 746                                 switch (BPF_OP(fp->code)) {
 747                                 case BPF_JEQ:
 748                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 749                                         break;
 750                                 case BPF_JGT:
 751                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 752                                         break;
 753                                 case BPF_JGE:
 754                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 755                                         break;
 756                                 default:
 757                                         goto jmp_rest;
 758                                 }
 759
 760                                 target = i + fp->jf + 1;
 761                                 BPF_EMIT_JMP;
 762                                 break;
 763                         }
 764 jmp_rest:
 765                         /* Other jumps are mapped into two insns: Jxx and JA. */
 766                         target = i + fp->jt + 1;
 767                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 768                         BPF_EMIT_JMP;
 769                         insn++;
 770
 771                         insn->code = BPF_JMP | BPF_JA;
 772                         target = i + fp->jf + 1;
 773                         BPF_EMIT_JMP;
 774                         break;
 775
 776                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 777                 case BPF_LDX | BPF_MSH | BPF_B: {
 778                         struct sock_filter tmp = {
 779                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 780                                 .k      = fp->k,
 781                         };
 782
 783                         *seen_ld_abs = true;
 784
 785                         /* X = A */
 786                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 787                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 788                         convert_bpf_ld_abs(&tmp, &insn);
 789                         insn++;
 790                         /* A &= 0xf */
 791                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 792                         /* A <<= 2 */
 793                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 794                         /* tmp = X */
 795                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 796                         /* X = A */
 797                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 798                         /* A = tmp */
 799                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 800                         break;
 801                 }
 802                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 803                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 804                  */
 805                 case BPF_RET | BPF_A:
 806                 case BPF_RET | BPF_K:
 807                         if (BPF_RVAL(fp->code) == BPF_K)
 808                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 809                                                         0, fp->k);
 810                         *insn = BPF_EXIT_INSN();
 811                         break;
 812
 813                 /* Store to stack. */
 814                 case BPF_ST:
 815                 case BPF_STX:
 816                         stack_off = fp->k * 4  + 4;
 817                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 818                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 819                                             -stack_off);
 820                         /* check_load_and_stores() verifies that classic BPF can
 821                          * load from stack only after write, so tracking
 822                          * stack_depth for ST|STX insns is enough
 823                          */
 824                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 825                                 new_prog->aux->stack_depth = stack_off;
 826                         break;
 827
 828                 /* Load from stack. */
 829                 case BPF_LD | BPF_MEM:
 830                 case BPF_LDX | BPF_MEM:
 831                         stack_off = fp->k * 4  + 4;
 832                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 833                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 834                                             -stack_off);
 835                         break;
 836
 837                 /* A = K or X = K */
 838                 case BPF_LD | BPF_IMM:
 839                 case BPF_LDX | BPF_IMM:
 840                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 841                                               BPF_REG_A : BPF_REG_X, fp->k);
 842                         break;
 843
 844                 /* X = A */
 845                 case BPF_MISC | BPF_TAX:
 846                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 847                         break;
 848
 849                 /* A = X */
 850                 case BPF_MISC | BPF_TXA:
 851                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 852                         break;
 853
 854                 /* A = skb->len or X = skb->len */
 855                 case BPF_LD | BPF_W | BPF_LEN:
 856                 case BPF_LDX | BPF_W | BPF_LEN:
 857                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 858                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 859                                             offsetof(struct sk_buff, len));
 860                         break;
 861
 862                 /* Access seccomp_data fields. */
 863                 case BPF_LDX | BPF_ABS | BPF_W:
 864                         /* A = *(u32 *) (ctx + K) */
 865                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 866                         break;
 867
 868                 /* Unknown instruction. */
 869                 default:
 870                         goto err;
 871                 }
 872
 873                 insn++;
 874                 if (new_prog)
 875                         memcpy(new_insn, tmp_insns,
 876                                sizeof(*insn) * (insn - tmp_insns));
 877                 new_insn += insn - tmp_insns;
 878         }
 879
 880         if (!new_prog) {
 881                 /* Only calculating new length. */
 882                 *new_len = new_insn - first_insn;
 883                 if (*seen_ld_abs)
 884                         *new_len += 4; /* Prologue bits. */
 885                 return 0;
 886         }
 887
 888         pass++;
 889         if (new_flen != new_insn - first_insn) {
 890                 new_flen = new_insn - first_insn;
 891                 if (pass > 2)
 892                         goto err;
 893                 goto do_pass;
 894         }
 895
 896         kfree(addrs);
 897         BUG_ON(*new_len != new_flen);
 898         return 0;
 899 err:
 900         kfree(addrs);
 901         return -EINVAL;
 902 }
 903
 904 /* Security:
 905  *
 906  * As we dont want to clear mem[] array for each packet going through
 907  * __bpf_prog_run(), we check that filter loaded by user never try to read
 908  * a cell if not previously written, and we check all branches to be sure
 909  * a malicious user doesn't try to abuse us.
 910  */
 911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 912 {
 913         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 914         int pc, ret = 0;
 915
 916         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 917
 918         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 919         if (!masks)
 920                 return -ENOMEM;
 921
 922         memset(masks, 0xff, flen * sizeof(*masks));
 923
 924         for (pc = 0; pc < flen; pc++) {
 925                 memvalid &= masks[pc];
 926
 927                 switch (filter[pc].code) {
 928                 case BPF_ST:
 929                 case BPF_STX:
 930                         memvalid |= (1 << filter[pc].k);
 931                         break;
 932                 case BPF_LD | BPF_MEM:
 933                 case BPF_LDX | BPF_MEM:
 934                         if (!(memvalid & (1 << filter[pc].k))) {
 935                                 ret = -EINVAL;
 936                                 goto error;
 937                         }
 938                         break;
 939                 case BPF_JMP | BPF_JA:
 940                         /* A jump must set masks on target */
 941                         masks[pc + 1 + filter[pc].k] &= memvalid;
 942                         memvalid = ~0;
 943                         break;
 944                 case BPF_JMP | BPF_JEQ | BPF_K:
 945                 case BPF_JMP | BPF_JEQ | BPF_X:
 946                 case BPF_JMP | BPF_JGE | BPF_K:
 947                 case BPF_JMP | BPF_JGE | BPF_X:
 948                 case BPF_JMP | BPF_JGT | BPF_K:
 949                 case BPF_JMP | BPF_JGT | BPF_X:
 950                 case BPF_JMP | BPF_JSET | BPF_K:
 951                 case BPF_JMP | BPF_JSET | BPF_X:
 952                         /* A jump must set masks on targets */
 953                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 954                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 955                         memvalid = ~0;
 956                         break;
 957                 }
 958         }
 959 error:
 960         kfree(masks);
 961         return ret;
 962 }
 963
 964 static bool chk_code_allowed(u16 code_to_probe)
 965 {
 966         static const bool codes[] = {
 967                 /* 32 bit ALU operations */
 968                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 969                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 970                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 971                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 972                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 973                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 974                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 975                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 976                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 977                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 978                 [BPF_ALU | BPF_AND | BPF_K] = true,
 979                 [BPF_ALU | BPF_AND | BPF_X] = true,
 980                 [BPF_ALU | BPF_OR | BPF_K] = true,
 981                 [BPF_ALU | BPF_OR | BPF_X] = true,
 982                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 983                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 984                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 985                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 986                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 987                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 988                 [BPF_ALU | BPF_NEG] = true,
 989                 /* Load instructions */
 990                 [BPF_LD | BPF_W | BPF_ABS] = true,
 991                 [BPF_LD | BPF_H | BPF_ABS] = true,
 992                 [BPF_LD | BPF_B | BPF_ABS] = true,
 993                 [BPF_LD | BPF_W | BPF_LEN] = true,
 994                 [BPF_LD | BPF_W | BPF_IND] = true,
 995                 [BPF_LD | BPF_H | BPF_IND] = true,
 996                 [BPF_LD | BPF_B | BPF_IND] = true,
 997                 [BPF_LD | BPF_IMM] = true,
 998                 [BPF_LD | BPF_MEM] = true,
 999                 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000                 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001                 [BPF_LDX | BPF_IMM] = true,
1002                 [BPF_LDX | BPF_MEM] = true,
1003                 /* Store instructions */
1004                 [BPF_ST] = true,
1005                 [BPF_STX] = true,
1006                 /* Misc instructions */
1007                 [BPF_MISC | BPF_TAX] = true,
1008                 [BPF_MISC | BPF_TXA] = true,
1009                 /* Return instructions */
1010                 [BPF_RET | BPF_K] = true,
1011                 [BPF_RET | BPF_A] = true,
1012                 /* Jump instructions */
1013                 [BPF_JMP | BPF_JA] = true,
1014                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1022         };
1023
1024         if (code_to_probe >= ARRAY_SIZE(codes))
1025                 return false;
1026
1027         return codes[code_to_probe];
1028 }
1029
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031                                 unsigned int flen)
1032 {
1033         if (filter == NULL)
1034                 return false;
1035         if (flen == 0 || flen > BPF_MAXINSNS)
1036                 return false;
1037
1038         return true;
1039 }
1040
1041 /**
1042  *      bpf_check_classic - verify socket filter code
1043  *      @filter: filter to verify
1044  *      @flen: length of filter
1045  *
1046  * Check the user's filter code. If we let some ugly
1047  * filter code slip through kaboom! The filter must contain
1048  * no references or jumps that are out of range, no illegal
1049  * instructions, and must end with a RET instruction.
1050  *
1051  * All jumps are forward as they are not signed.
1052  *
1053  * Returns 0 if the rule set is legal or -EINVAL if not.
1054  */
1055 static int bpf_check_classic(const struct sock_filter *filter,
1056                              unsigned int flen)
1057 {
1058         bool anc_found;
1059         int pc;
1060
1061         /* Check the filter code now */
1062         for (pc = 0; pc < flen; pc++) {
1063                 const struct sock_filter *ftest = &filter[pc];
1064
1065                 /* May we actually operate on this code? */
1066                 if (!chk_code_allowed(ftest->code))
1067                         return -EINVAL;
1068
1069                 /* Some instructions need special checks */
1070                 switch (ftest->code) {
1071                 case BPF_ALU | BPF_DIV | BPF_K:
1072                 case BPF_ALU | BPF_MOD | BPF_K:
1073                         /* Check for division by zero */
1074                         if (ftest->k == 0)
1075                                 return -EINVAL;
1076                         break;
1077                 case BPF_ALU | BPF_LSH | BPF_K:
1078                 case BPF_ALU | BPF_RSH | BPF_K:
1079                         if (ftest->k >= 32)
1080                                 return -EINVAL;
1081                         break;
1082                 case BPF_LD | BPF_MEM:
1083                 case BPF_LDX | BPF_MEM:
1084                 case BPF_ST:
1085                 case BPF_STX:
1086                         /* Check for invalid memory addresses */
1087                         if (ftest->k >= BPF_MEMWORDS)
1088                                 return -EINVAL;
1089                         break;
1090                 case BPF_JMP | BPF_JA:
1091                         /* Note, the large ftest->k might cause loops.
1092                          * Compare this with conditional jumps below,
1093                          * where offsets are limited. --ANK (981016)
1094                          */
1095                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1096                                 return -EINVAL;
1097                         break;
1098                 case BPF_JMP | BPF_JEQ | BPF_K:
1099                 case BPF_JMP | BPF_JEQ | BPF_X:
1100                 case BPF_JMP | BPF_JGE | BPF_K:
1101                 case BPF_JMP | BPF_JGE | BPF_X:
1102                 case BPF_JMP | BPF_JGT | BPF_K:
1103                 case BPF_JMP | BPF_JGT | BPF_X:
1104                 case BPF_JMP | BPF_JSET | BPF_K:
1105                 case BPF_JMP | BPF_JSET | BPF_X:
1106                         /* Both conditionals must be safe */
1107                         if (pc + ftest->jt + 1 >= flen ||
1108                             pc + ftest->jf + 1 >= flen)
1109                                 return -EINVAL;
1110                         break;
1111                 case BPF_LD | BPF_W | BPF_ABS:
1112                 case BPF_LD | BPF_H | BPF_ABS:
1113                 case BPF_LD | BPF_B | BPF_ABS:
1114                         anc_found = false;
1115                         if (bpf_anc_helper(ftest) & BPF_ANC)
1116                                 anc_found = true;
1117                         /* Ancillary operation unknown or unsupported */
1118                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119                                 return -EINVAL;
1120                 }
1121         }
1122
1123         /* Last instruction must be a RET code */
1124         switch (filter[flen - 1].code) {
1125         case BPF_RET | BPF_K:
1126         case BPF_RET | BPF_A:
1127                 return check_load_and_stores(filter, flen);
1128         }
1129
1130         return -EINVAL;
1131 }
1132
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134                                       const struct sock_fprog *fprog)
1135 {
1136         unsigned int fsize = bpf_classic_proglen(fprog);
1137         struct sock_fprog_kern *fkprog;
1138
1139         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140         if (!fp->orig_prog)
1141                 return -ENOMEM;
1142
1143         fkprog = fp->orig_prog;
1144         fkprog->len = fprog->len;
1145
1146         fkprog->filter = kmemdup(fp->insns, fsize,
1147                                  GFP_KERNEL | __GFP_NOWARN);
1148         if (!fkprog->filter) {
1149                 kfree(fp->orig_prog);
1150                 return -ENOMEM;
1151         }
1152
1153         return 0;
1154 }
1155
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1157 {
1158         struct sock_fprog_kern *fprog = fp->orig_prog;
1159
1160         if (fprog) {
1161                 kfree(fprog->filter);
1162                 kfree(fprog);
1163         }
1164 }
1165
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1167 {
1168         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1169                 bpf_prog_put(prog);
1170         } else {
1171                 bpf_release_orig_filter(prog);
1172                 bpf_prog_free(prog);
1173         }
1174 }
1175
1176 static void __sk_filter_release(struct sk_filter *fp)
1177 {
1178         __bpf_prog_release(fp->prog);
1179         kfree(fp);
1180 }
1181
1182 /**
1183  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1184  *      @rcu: rcu_head that contains the sk_filter to free
1185  */
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1187 {
1188         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1189
1190         __sk_filter_release(fp);
1191 }
1192
1193 /**
1194  *      sk_filter_release - release a socket filter
1195  *      @fp: filter to remove
1196  *
1197  *      Remove a filter from a socket and release its resources.
1198  */
1199 static void sk_filter_release(struct sk_filter *fp)
1200 {
1201         if (refcount_dec_and_test(&fp->refcnt))
1202                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1203 }
1204
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1206 {
1207         u32 filter_size = bpf_prog_size(fp->prog->len);
1208
1209         atomic_sub(filter_size, &sk->sk_omem_alloc);
1210         sk_filter_release(fp);
1211 }
1212
1213 /* try to charge the socket memory if there is space available
1214  * return true on success
1215  */
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1217 {
1218         u32 filter_size = bpf_prog_size(fp->prog->len);
1219         int optmem_max = READ_ONCE(sysctl_optmem_max);
1220
1221         /* same check as in sock_kmalloc() */
1222         if (filter_size <= optmem_max &&
1223             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224                 atomic_add(filter_size, &sk->sk_omem_alloc);
1225                 return true;
1226         }
1227         return false;
1228 }
1229
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1231 {
1232         if (!refcount_inc_not_zero(&fp->refcnt))
1233                 return false;
1234
1235         if (!__sk_filter_charge(sk, fp)) {
1236                 sk_filter_release(fp);
1237                 return false;
1238         }
1239         return true;
1240 }
1241
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1243 {
1244         struct sock_filter *old_prog;
1245         struct bpf_prog *old_fp;
1246         int err, new_len, old_len = fp->len;
1247         bool seen_ld_abs = false;
1248
1249         /* We are free to overwrite insns et al right here as it won't be used at
1250          * this point in time anymore internally after the migration to the eBPF
1251          * instruction representation.
1252          */
1253         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254                      sizeof(struct bpf_insn));
1255
1256         /* Conversion cannot happen on overlapping memory areas,
1257          * so we need to keep the user BPF around until the 2nd
1258          * pass. At this time, the user BPF is stored in fp->insns.
1259          */
1260         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261                            GFP_KERNEL | __GFP_NOWARN);
1262         if (!old_prog) {
1263                 err = -ENOMEM;
1264                 goto out_err;
1265         }
1266
1267         /* 1st pass: calculate the new program length. */
1268         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269                                  &seen_ld_abs);
1270         if (err)
1271                 goto out_err_free;
1272
1273         /* Expand fp for appending the new filter representation. */
1274         old_fp = fp;
1275         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1276         if (!fp) {
1277                 /* The old_fp is still around in case we couldn't
1278                  * allocate new memory, so uncharge on that one.
1279                  */
1280                 fp = old_fp;
1281                 err = -ENOMEM;
1282                 goto out_err_free;
1283         }
1284
1285         fp->len = new_len;
1286
1287         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1289                                  &seen_ld_abs);
1290         if (err)
1291                 /* 2nd bpf_convert_filter() can fail only if it fails
1292                  * to allocate memory, remapping must succeed. Note,
1293                  * that at this time old_fp has already been released
1294                  * by krealloc().
1295                  */
1296                 goto out_err_free;
1297
1298         fp = bpf_prog_select_runtime(fp, &err);
1299         if (err)
1300                 goto out_err_free;
1301
1302         kfree(old_prog);
1303         return fp;
1304
1305 out_err_free:
1306         kfree(old_prog);
1307 out_err:
1308         __bpf_prog_release(fp);
1309         return ERR_PTR(err);
1310 }
1311
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313                                            bpf_aux_classic_check_t trans)
1314 {
1315         int err;
1316
1317         fp->bpf_func = NULL;
1318         fp->jited = 0;
1319
1320         err = bpf_check_classic(fp->insns, fp->len);
1321         if (err) {
1322                 __bpf_prog_release(fp);
1323                 return ERR_PTR(err);
1324         }
1325
1326         /* There might be additional checks and transformations
1327          * needed on classic filters, f.e. in case of seccomp.
1328          */
1329         if (trans) {
1330                 err = trans(fp->insns, fp->len);
1331                 if (err) {
1332                         __bpf_prog_release(fp);
1333                         return ERR_PTR(err);
1334                 }
1335         }
1336
1337         /* Probe if we can JIT compile the filter and if so, do
1338          * the compilation of the filter.
1339          */
1340         bpf_jit_compile(fp);
1341
1342         /* JIT compiler couldn't process this filter, so do the eBPF translation
1343          * for the optimized interpreter.
1344          */
1345         if (!fp->jited)
1346                 fp = bpf_migrate_filter(fp);
1347
1348         return fp;
1349 }
1350
1351 /**
1352  *      bpf_prog_create - create an unattached filter
1353  *      @pfp: the unattached filter that is created
1354  *      @fprog: the filter program
1355  *
1356  * Create a filter independent of any socket. We first run some
1357  * sanity checks on it to make sure it does not explode on us later.
1358  * If an error occurs or there is insufficient memory for the filter
1359  * a negative errno code is returned. On success the return is zero.
1360  */
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1362 {
1363         unsigned int fsize = bpf_classic_proglen(fprog);
1364         struct bpf_prog *fp;
1365
1366         /* Make sure new filter is there and in the right amounts. */
1367         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1368                 return -EINVAL;
1369
1370         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371         if (!fp)
1372                 return -ENOMEM;
1373
1374         memcpy(fp->insns, fprog->filter, fsize);
1375
1376         fp->len = fprog->len;
1377         /* Since unattached filters are not copied back to user
1378          * space through sk_get_filter(), we do not need to hold
1379          * a copy here, and can spare us the work.
1380          */
1381         fp->orig_prog = NULL;
1382
1383         /* bpf_prepare_filter() already takes care of freeing
1384          * memory in case something goes wrong.
1385          */
1386         fp = bpf_prepare_filter(fp, NULL);
1387         if (IS_ERR(fp))
1388                 return PTR_ERR(fp);
1389
1390         *pfp = fp;
1391         return 0;
1392 }
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1394
1395 /**
1396  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1397  *      @pfp: the unattached filter that is created
1398  *      @fprog: the filter program
1399  *      @trans: post-classic verifier transformation handler
1400  *      @save_orig: save classic BPF program
1401  *
1402  * This function effectively does the same as bpf_prog_create(), only
1403  * that it builds up its insns buffer from user space provided buffer.
1404  * It also allows for passing a bpf_aux_classic_check_t handler.
1405  */
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407                               bpf_aux_classic_check_t trans, bool save_orig)
1408 {
1409         unsigned int fsize = bpf_classic_proglen(fprog);
1410         struct bpf_prog *fp;
1411         int err;
1412
1413         /* Make sure new filter is there and in the right amounts. */
1414         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1415                 return -EINVAL;
1416
1417         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418         if (!fp)
1419                 return -ENOMEM;
1420
1421         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422                 __bpf_prog_free(fp);
1423                 return -EFAULT;
1424         }
1425
1426         fp->len = fprog->len;
1427         fp->orig_prog = NULL;
1428
1429         if (save_orig) {
1430                 err = bpf_prog_store_orig_filter(fp, fprog);
1431                 if (err) {
1432                         __bpf_prog_free(fp);
1433                         return -ENOMEM;
1434                 }
1435         }
1436
1437         /* bpf_prepare_filter() already takes care of freeing
1438          * memory in case something goes wrong.
1439          */
1440         fp = bpf_prepare_filter(fp, trans);
1441         if (IS_ERR(fp))
1442                 return PTR_ERR(fp);
1443
1444         *pfp = fp;
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1448
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1450 {
1451         __bpf_prog_release(fp);
1452 }
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1454
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1456 {
1457         struct sk_filter *fp, *old_fp;
1458
1459         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1460         if (!fp)
1461                 return -ENOMEM;
1462
1463         fp->prog = prog;
1464
1465         if (!__sk_filter_charge(sk, fp)) {
1466                 kfree(fp);
1467                 return -ENOMEM;
1468         }
1469         refcount_set(&fp->refcnt, 1);
1470
1471         old_fp = rcu_dereference_protected(sk->sk_filter,
1472                                            lockdep_sock_is_held(sk));
1473         rcu_assign_pointer(sk->sk_filter, fp);
1474
1475         if (old_fp)
1476                 sk_filter_uncharge(sk, old_fp);
1477
1478         return 0;
1479 }
1480
1481 static
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1483 {
1484         unsigned int fsize = bpf_classic_proglen(fprog);
1485         struct bpf_prog *prog;
1486         int err;
1487
1488         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489                 return ERR_PTR(-EPERM);
1490
1491         /* Make sure new filter is there and in the right amounts. */
1492         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493                 return ERR_PTR(-EINVAL);
1494
1495         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1496         if (!prog)
1497                 return ERR_PTR(-ENOMEM);
1498
1499         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500                 __bpf_prog_free(prog);
1501                 return ERR_PTR(-EFAULT);
1502         }
1503
1504         prog->len = fprog->len;
1505
1506         err = bpf_prog_store_orig_filter(prog, fprog);
1507         if (err) {
1508                 __bpf_prog_free(prog);
1509                 return ERR_PTR(-ENOMEM);
1510         }
1511
1512         /* bpf_prepare_filter() already takes care of freeing
1513          * memory in case something goes wrong.
1514          */
1515         return bpf_prepare_filter(prog, NULL);
1516 }
1517
1518 /**
1519  *      sk_attach_filter - attach a socket filter
1520  *      @fprog: the filter program
1521  *      @sk: the socket to use
1522  *
1523  * Attach the user's filter code. We first run some sanity checks on
1524  * it to make sure it does not explode on us later. If an error
1525  * occurs or there is insufficient memory for the filter a negative
1526  * errno code is returned. On success the return is zero.
1527  */
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1529 {
1530         struct bpf_prog *prog = __get_filter(fprog, sk);
1531         int err;
1532
1533         if (IS_ERR(prog))
1534                 return PTR_ERR(prog);
1535
1536         err = __sk_attach_prog(prog, sk);
1537         if (err < 0) {
1538                 __bpf_prog_release(prog);
1539                 return err;
1540         }
1541
1542         return 0;
1543 }
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1545
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1547 {
1548         struct bpf_prog *prog = __get_filter(fprog, sk);
1549         int err;
1550
1551         if (IS_ERR(prog))
1552                 return PTR_ERR(prog);
1553
1554         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1555                 err = -ENOMEM;
1556         else
1557                 err = reuseport_attach_prog(sk, prog);
1558
1559         if (err)
1560                 __bpf_prog_release(prog);
1561
1562         return err;
1563 }
1564
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1566 {
1567         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568                 return ERR_PTR(-EPERM);
1569
1570         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1571 }
1572
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1574 {
1575         struct bpf_prog *prog = __get_bpf(ufd, sk);
1576         int err;
1577
1578         if (IS_ERR(prog))
1579                 return PTR_ERR(prog);
1580
1581         err = __sk_attach_prog(prog, sk);
1582         if (err < 0) {
1583                 bpf_prog_put(prog);
1584                 return err;
1585         }
1586
1587         return 0;
1588 }
1589
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1591 {
1592         struct bpf_prog *prog;
1593         int err;
1594
1595         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1596                 return -EPERM;
1597
1598         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599         if (PTR_ERR(prog) == -EINVAL)
1600                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1601         if (IS_ERR(prog))
1602                 return PTR_ERR(prog);
1603
1604         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606                  * bpf prog (e.g. sockmap).  It depends on the
1607                  * limitation imposed by bpf_prog_load().
1608                  * Hence, sysctl_optmem_max is not checked.
1609                  */
1610                 if ((sk->sk_type != SOCK_STREAM &&
1611                      sk->sk_type != SOCK_DGRAM) ||
1612                     (sk->sk_protocol != IPPROTO_UDP &&
1613                      sk->sk_protocol != IPPROTO_TCP) ||
1614                     (sk->sk_family != AF_INET &&
1615                      sk->sk_family != AF_INET6)) {
1616                         err = -ENOTSUPP;
1617                         goto err_prog_put;
1618                 }
1619         } else {
1620                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1622                         err = -ENOMEM;
1623                         goto err_prog_put;
1624                 }
1625         }
1626
1627         err = reuseport_attach_prog(sk, prog);
1628 err_prog_put:
1629         if (err)
1630                 bpf_prog_put(prog);
1631
1632         return err;
1633 }
1634
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 {
1637         if (!prog)
1638                 return;
1639
1640         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1641                 bpf_prog_put(prog);
1642         else
1643                 bpf_prog_destroy(prog);
1644 }
1645
1646 struct bpf_scratchpad {
1647         union {
1648                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649                 u8     buff[MAX_BPF_STACK];
1650         };
1651 };
1652
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1654
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656                                           unsigned int write_len)
1657 {
1658         return skb_ensure_writable(skb, write_len);
1659 }
1660
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662                                         unsigned int write_len)
1663 {
1664         int err = __bpf_try_make_writable(skb, write_len);
1665
1666         bpf_compute_data_pointers(skb);
1667         return err;
1668 }
1669
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1671 {
1672         return bpf_try_make_writable(skb, skb_headlen(skb));
1673 }
1674
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1676 {
1677         if (skb_at_tc_ingress(skb))
1678                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1679 }
1680
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1682 {
1683         if (skb_at_tc_ingress(skb))
1684                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1685 }
1686
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688            const void *, from, u32, len, u64, flags)
1689 {
1690         void *ptr;
1691
1692         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1693                 return -EINVAL;
1694         if (unlikely(offset > INT_MAX))
1695                 return -EFAULT;
1696         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1697                 return -EFAULT;
1698
1699         ptr = skb->data + offset;
1700         if (flags & BPF_F_RECOMPUTE_CSUM)
1701                 __skb_postpull_rcsum(skb, ptr, len, offset);
1702
1703         memcpy(ptr, from, len);
1704
1705         if (flags & BPF_F_RECOMPUTE_CSUM)
1706                 __skb_postpush_rcsum(skb, ptr, len, offset);
1707         if (flags & BPF_F_INVALIDATE_HASH)
1708                 skb_clear_hash(skb);
1709
1710         return 0;
1711 }
1712
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714         .func           = bpf_skb_store_bytes,
1715         .gpl_only       = false,
1716         .ret_type       = RET_INTEGER,
1717         .arg1_type      = ARG_PTR_TO_CTX,
1718         .arg2_type      = ARG_ANYTHING,
1719         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1720         .arg4_type      = ARG_CONST_SIZE,
1721         .arg5_type      = ARG_ANYTHING,
1722 };
1723
1724 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from,
1725                           u32 len, u64 flags)
1726 {
1727         return ____bpf_skb_store_bytes(skb, offset, from, len, flags);
1728 }
1729
1730 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1731            void *, to, u32, len)
1732 {
1733         void *ptr;
1734
1735         if (unlikely(offset > INT_MAX))
1736                 goto err_clear;
1737
1738         ptr = skb_header_pointer(skb, offset, len, to);
1739         if (unlikely(!ptr))
1740                 goto err_clear;
1741         if (ptr != to)
1742                 memcpy(to, ptr, len);
1743
1744         return 0;
1745 err_clear:
1746         memset(to, 0, len);
1747         return -EFAULT;
1748 }
1749
1750 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1751         .func           = bpf_skb_load_bytes,
1752         .gpl_only       = false,
1753         .ret_type       = RET_INTEGER,
1754         .arg1_type      = ARG_PTR_TO_CTX,
1755         .arg2_type      = ARG_ANYTHING,
1756         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1757         .arg4_type      = ARG_CONST_SIZE,
1758 };
1759
1760 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len)
1761 {
1762         return ____bpf_skb_load_bytes(skb, offset, to, len);
1763 }
1764
1765 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1766            const struct bpf_flow_dissector *, ctx, u32, offset,
1767            void *, to, u32, len)
1768 {
1769         void *ptr;
1770
1771         if (unlikely(offset > 0xffff))
1772                 goto err_clear;
1773
1774         if (unlikely(!ctx->skb))
1775                 goto err_clear;
1776
1777         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1778         if (unlikely(!ptr))
1779                 goto err_clear;
1780         if (ptr != to)
1781                 memcpy(to, ptr, len);
1782
1783         return 0;
1784 err_clear:
1785         memset(to, 0, len);
1786         return -EFAULT;
1787 }
1788
1789 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1790         .func           = bpf_flow_dissector_load_bytes,
1791         .gpl_only       = false,
1792         .ret_type       = RET_INTEGER,
1793         .arg1_type      = ARG_PTR_TO_CTX,
1794         .arg2_type      = ARG_ANYTHING,
1795         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1796         .arg4_type      = ARG_CONST_SIZE,
1797 };
1798
1799 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1800            u32, offset, void *, to, u32, len, u32, start_header)
1801 {
1802         u8 *end = skb_tail_pointer(skb);
1803         u8 *start, *ptr;
1804
1805         if (unlikely(offset > 0xffff))
1806                 goto err_clear;
1807
1808         switch (start_header) {
1809         case BPF_HDR_START_MAC:
1810                 if (unlikely(!skb_mac_header_was_set(skb)))
1811                         goto err_clear;
1812                 start = skb_mac_header(skb);
1813                 break;
1814         case BPF_HDR_START_NET:
1815                 start = skb_network_header(skb);
1816                 break;
1817         default:
1818                 goto err_clear;
1819         }
1820
1821         ptr = start + offset;
1822
1823         if (likely(ptr + len <= end)) {
1824                 memcpy(to, ptr, len);
1825                 return 0;
1826         }
1827
1828 err_clear:
1829         memset(to, 0, len);
1830         return -EFAULT;
1831 }
1832
1833 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1834         .func           = bpf_skb_load_bytes_relative,
1835         .gpl_only       = false,
1836         .ret_type       = RET_INTEGER,
1837         .arg1_type      = ARG_PTR_TO_CTX,
1838         .arg2_type      = ARG_ANYTHING,
1839         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1840         .arg4_type      = ARG_CONST_SIZE,
1841         .arg5_type      = ARG_ANYTHING,
1842 };
1843
1844 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1845 {
1846         /* Idea is the following: should the needed direct read/write
1847          * test fail during runtime, we can pull in more data and redo
1848          * again, since implicitly, we invalidate previous checks here.
1849          *
1850          * Or, since we know how much we need to make read/writeable,
1851          * this can be done once at the program beginning for direct
1852          * access case. By this we overcome limitations of only current
1853          * headroom being accessible.
1854          */
1855         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1856 }
1857
1858 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1859         .func           = bpf_skb_pull_data,
1860         .gpl_only       = false,
1861         .ret_type       = RET_INTEGER,
1862         .arg1_type      = ARG_PTR_TO_CTX,
1863         .arg2_type      = ARG_ANYTHING,
1864 };
1865
1866 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1867 {
1868         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1869 }
1870
1871 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1872         .func           = bpf_sk_fullsock,
1873         .gpl_only       = false,
1874         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1875         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1876 };
1877
1878 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1879                                            unsigned int write_len)
1880 {
1881         return __bpf_try_make_writable(skb, write_len);
1882 }
1883
1884 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1885 {
1886         /* Idea is the following: should the needed direct read/write
1887          * test fail during runtime, we can pull in more data and redo
1888          * again, since implicitly, we invalidate previous checks here.
1889          *
1890          * Or, since we know how much we need to make read/writeable,
1891          * this can be done once at the program beginning for direct
1892          * access case. By this we overcome limitations of only current
1893          * headroom being accessible.
1894          */
1895         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1896 }
1897
1898 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1899         .func           = sk_skb_pull_data,
1900         .gpl_only       = false,
1901         .ret_type       = RET_INTEGER,
1902         .arg1_type      = ARG_PTR_TO_CTX,
1903         .arg2_type      = ARG_ANYTHING,
1904 };
1905
1906 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1907            u64, from, u64, to, u64, flags)
1908 {
1909         __sum16 *ptr;
1910
1911         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1912                 return -EINVAL;
1913         if (unlikely(offset > 0xffff || offset & 1))
1914                 return -EFAULT;
1915         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1916                 return -EFAULT;
1917
1918         ptr = (__sum16 *)(skb->data + offset);
1919         switch (flags & BPF_F_HDR_FIELD_MASK) {
1920         case 0:
1921                 if (unlikely(from != 0))
1922                         return -EINVAL;
1923
1924                 csum_replace_by_diff(ptr, to);
1925                 break;
1926         case 2:
1927                 csum_replace2(ptr, from, to);
1928                 break;
1929         case 4:
1930                 csum_replace4(ptr, from, to);
1931                 break;
1932         default:
1933                 return -EINVAL;
1934         }
1935
1936         return 0;
1937 }
1938
1939 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1940         .func           = bpf_l3_csum_replace,
1941         .gpl_only       = false,
1942         .ret_type       = RET_INTEGER,
1943         .arg1_type      = ARG_PTR_TO_CTX,
1944         .arg2_type      = ARG_ANYTHING,
1945         .arg3_type      = ARG_ANYTHING,
1946         .arg4_type      = ARG_ANYTHING,
1947         .arg5_type      = ARG_ANYTHING,
1948 };
1949
1950 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1951            u64, from, u64, to, u64, flags)
1952 {
1953         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1954         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1955         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1956         __sum16 *ptr;
1957
1958         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1959                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1960                 return -EINVAL;
1961         if (unlikely(offset > 0xffff || offset & 1))
1962                 return -EFAULT;
1963         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1964                 return -EFAULT;
1965
1966         ptr = (__sum16 *)(skb->data + offset);
1967         if (is_mmzero && !do_mforce && !*ptr)
1968                 return 0;
1969
1970         switch (flags & BPF_F_HDR_FIELD_MASK) {
1971         case 0:
1972                 if (unlikely(from != 0))
1973                         return -EINVAL;
1974
1975                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1976                 break;
1977         case 2:
1978                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1979                 break;
1980         case 4:
1981                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1982                 break;
1983         default:
1984                 return -EINVAL;
1985         }
1986
1987         if (is_mmzero && !*ptr)
1988                 *ptr = CSUM_MANGLED_0;
1989         return 0;
1990 }
1991
1992 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1993         .func           = bpf_l4_csum_replace,
1994         .gpl_only       = false,
1995         .ret_type       = RET_INTEGER,
1996         .arg1_type      = ARG_PTR_TO_CTX,
1997         .arg2_type      = ARG_ANYTHING,
1998         .arg3_type      = ARG_ANYTHING,
1999         .arg4_type      = ARG_ANYTHING,
2000         .arg5_type      = ARG_ANYTHING,
2001 };
2002
2003 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
2004            __be32 *, to, u32, to_size, __wsum, seed)
2005 {
2006         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
2007         u32 diff_size = from_size + to_size;
2008         int i, j = 0;
2009
2010         /* This is quite flexible, some examples:
2011          *
2012          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2013          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2014          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2015          *
2016          * Even for diffing, from_size and to_size don't need to be equal.
2017          */
2018         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2019                      diff_size > sizeof(sp->diff)))
2020                 return -EINVAL;
2021
2022         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2023                 sp->diff[j] = ~from[i];
2024         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2025                 sp->diff[j] = to[i];
2026
2027         return csum_partial(sp->diff, diff_size, seed);
2028 }
2029
2030 static const struct bpf_func_proto bpf_csum_diff_proto = {
2031         .func           = bpf_csum_diff,
2032         .gpl_only       = false,
2033         .pkt_access     = true,
2034         .ret_type       = RET_INTEGER,
2035         .arg1_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2036         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2037         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2038         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2039         .arg5_type      = ARG_ANYTHING,
2040 };
2041
2042 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2043 {
2044         /* The interface is to be used in combination with bpf_csum_diff()
2045          * for direct packet writes. csum rotation for alignment as well
2046          * as emulating csum_sub() can be done from the eBPF program.
2047          */
2048         if (skb->ip_summed == CHECKSUM_COMPLETE)
2049                 return (skb->csum = csum_add(skb->csum, csum));
2050
2051         return -ENOTSUPP;
2052 }
2053
2054 static const struct bpf_func_proto bpf_csum_update_proto = {
2055         .func           = bpf_csum_update,
2056         .gpl_only       = false,
2057         .ret_type       = RET_INTEGER,
2058         .arg1_type      = ARG_PTR_TO_CTX,
2059         .arg2_type      = ARG_ANYTHING,
2060 };
2061
2062 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2063 {
2064         /* The interface is to be used in combination with bpf_skb_adjust_room()
2065          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2066          * is passed as flags, for example.
2067          */
2068         switch (level) {
2069         case BPF_CSUM_LEVEL_INC:
2070                 __skb_incr_checksum_unnecessary(skb);
2071                 break;
2072         case BPF_CSUM_LEVEL_DEC:
2073                 __skb_decr_checksum_unnecessary(skb);
2074                 break;
2075         case BPF_CSUM_LEVEL_RESET:
2076                 __skb_reset_checksum_unnecessary(skb);
2077                 break;
2078         case BPF_CSUM_LEVEL_QUERY:
2079                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2080                        skb->csum_level : -EACCES;
2081         default:
2082                 return -EINVAL;
2083         }
2084
2085         return 0;
2086 }
2087
2088 static const struct bpf_func_proto bpf_csum_level_proto = {
2089         .func           = bpf_csum_level,
2090         .gpl_only       = false,
2091         .ret_type       = RET_INTEGER,
2092         .arg1_type      = ARG_PTR_TO_CTX,
2093         .arg2_type      = ARG_ANYTHING,
2094 };
2095
2096 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2097 {
2098         return dev_forward_skb_nomtu(dev, skb);
2099 }
2100
2101 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2102                                       struct sk_buff *skb)
2103 {
2104         int ret = ____dev_forward_skb(dev, skb, false);
2105
2106         if (likely(!ret)) {
2107                 skb->dev = dev;
2108                 ret = netif_rx(skb);
2109         }
2110
2111         return ret;
2112 }
2113
2114 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2115 {
2116         int ret;
2117
2118         if (dev_xmit_recursion()) {
2119                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2120                 kfree_skb(skb);
2121                 return -ENETDOWN;
2122         }
2123
2124         skb->dev = dev;
2125         skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb));
2126         skb_clear_tstamp(skb);
2127
2128         dev_xmit_recursion_inc();
2129         ret = dev_queue_xmit(skb);
2130         dev_xmit_recursion_dec();
2131
2132         return ret;
2133 }
2134
2135 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2136                                  u32 flags)
2137 {
2138         unsigned int mlen = skb_network_offset(skb);
2139
2140         if (unlikely(skb->len <= mlen)) {
2141                 kfree_skb(skb);
2142                 return -ERANGE;
2143         }
2144
2145         if (mlen) {
2146                 __skb_pull(skb, mlen);
2147
2148                 /* At ingress, the mac header has already been pulled once.
2149                  * At egress, skb_pospull_rcsum has to be done in case that
2150                  * the skb is originated from ingress (i.e. a forwarded skb)
2151                  * to ensure that rcsum starts at net header.
2152                  */
2153                 if (!skb_at_tc_ingress(skb))
2154                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2155         }
2156         skb_pop_mac_header(skb);
2157         skb_reset_mac_len(skb);
2158         return flags & BPF_F_INGRESS ?
2159                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2160 }
2161
2162 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2163                                  u32 flags)
2164 {
2165         /* Verify that a link layer header is carried */
2166         if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2167                 kfree_skb(skb);
2168                 return -ERANGE;
2169         }
2170
2171         bpf_push_mac_rcsum(skb);
2172         return flags & BPF_F_INGRESS ?
2173                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2174 }
2175
2176 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2177                           u32 flags)
2178 {
2179         if (dev_is_mac_header_xmit(dev))
2180                 return __bpf_redirect_common(skb, dev, flags);
2181         else
2182                 return __bpf_redirect_no_mac(skb, dev, flags);
2183 }
2184
2185 #if IS_ENABLED(CONFIG_IPV6)
2186 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2187                             struct net_device *dev, struct bpf_nh_params *nh)
2188 {
2189         u32 hh_len = LL_RESERVED_SPACE(dev);
2190         const struct in6_addr *nexthop;
2191         struct dst_entry *dst = NULL;
2192         struct neighbour *neigh;
2193
2194         if (dev_xmit_recursion()) {
2195                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2196                 goto out_drop;
2197         }
2198
2199         skb->dev = dev;
2200         skb_clear_tstamp(skb);
2201
2202         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2203                 skb = skb_expand_head(skb, hh_len);
2204                 if (!skb)
2205                         return -ENOMEM;
2206         }
2207
2208         rcu_read_lock();
2209         if (!nh) {
2210                 dst = skb_dst(skb);
2211                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2212                                       &ipv6_hdr(skb)->daddr);
2213         } else {
2214                 nexthop = &nh->ipv6_nh;
2215         }
2216         neigh = ip_neigh_gw6(dev, nexthop);
2217         if (likely(!IS_ERR(neigh))) {
2218                 int ret;
2219
2220                 sock_confirm_neigh(skb, neigh);
2221                 local_bh_disable();
2222                 dev_xmit_recursion_inc();
2223                 ret = neigh_output(neigh, skb, false);
2224                 dev_xmit_recursion_dec();
2225                 local_bh_enable();
2226                 rcu_read_unlock();
2227                 return ret;
2228         }
2229         rcu_read_unlock_bh();
2230         if (dst)
2231                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2232 out_drop:
2233         kfree_skb(skb);
2234         return -ENETDOWN;
2235 }
2236
2237 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2238                                    struct bpf_nh_params *nh)
2239 {
2240         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2241         struct net *net = dev_net(dev);
2242         int err, ret = NET_XMIT_DROP;
2243
2244         if (!nh) {
2245                 struct dst_entry *dst;
2246                 struct flowi6 fl6 = {
2247                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2248                         .flowi6_mark  = skb->mark,
2249                         .flowlabel    = ip6_flowinfo(ip6h),
2250                         .flowi6_oif   = dev->ifindex,
2251                         .flowi6_proto = ip6h->nexthdr,
2252                         .daddr        = ip6h->daddr,
2253                         .saddr        = ip6h->saddr,
2254                 };
2255
2256                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2257                 if (IS_ERR(dst))
2258                         goto out_drop;
2259
2260                 skb_dst_set(skb, dst);
2261         } else if (nh->nh_family != AF_INET6) {
2262                 goto out_drop;
2263         }
2264
2265         err = bpf_out_neigh_v6(net, skb, dev, nh);
2266         if (unlikely(net_xmit_eval(err)))
2267                 dev->stats.tx_errors++;
2268         else
2269                 ret = NET_XMIT_SUCCESS;
2270         goto out_xmit;
2271 out_drop:
2272         dev->stats.tx_errors++;
2273         kfree_skb(skb);
2274 out_xmit:
2275         return ret;
2276 }
2277 #else
2278 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2279                                    struct bpf_nh_params *nh)
2280 {
2281         kfree_skb(skb);
2282         return NET_XMIT_DROP;
2283 }
2284 #endif /* CONFIG_IPV6 */
2285
2286 #if IS_ENABLED(CONFIG_INET)
2287 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2288                             struct net_device *dev, struct bpf_nh_params *nh)
2289 {
2290         u32 hh_len = LL_RESERVED_SPACE(dev);
2291         struct neighbour *neigh;
2292         bool is_v6gw = false;
2293
2294         if (dev_xmit_recursion()) {
2295                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2296                 goto out_drop;
2297         }
2298
2299         skb->dev = dev;
2300         skb_clear_tstamp(skb);
2301
2302         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2303                 skb = skb_expand_head(skb, hh_len);
2304                 if (!skb)
2305                         return -ENOMEM;
2306         }
2307
2308         rcu_read_lock();
2309         if (!nh) {
2310                 struct dst_entry *dst = skb_dst(skb);
2311                 struct rtable *rt = container_of(dst, struct rtable, dst);
2312
2313                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2314         } else if (nh->nh_family == AF_INET6) {
2315                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2316                 is_v6gw = true;
2317         } else if (nh->nh_family == AF_INET) {
2318                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2319         } else {
2320                 rcu_read_unlock();
2321                 goto out_drop;
2322         }
2323
2324         if (likely(!IS_ERR(neigh))) {
2325                 int ret;
2326
2327                 sock_confirm_neigh(skb, neigh);
2328                 local_bh_disable();
2329                 dev_xmit_recursion_inc();
2330                 ret = neigh_output(neigh, skb, is_v6gw);
2331                 dev_xmit_recursion_dec();
2332                 local_bh_enable();
2333                 rcu_read_unlock();
2334                 return ret;
2335         }
2336         rcu_read_unlock();
2337 out_drop:
2338         kfree_skb(skb);
2339         return -ENETDOWN;
2340 }
2341
2342 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2343                                    struct bpf_nh_params *nh)
2344 {
2345         const struct iphdr *ip4h = ip_hdr(skb);
2346         struct net *net = dev_net(dev);
2347         int err, ret = NET_XMIT_DROP;
2348
2349         if (!nh) {
2350                 struct flowi4 fl4 = {
2351                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2352                         .flowi4_mark  = skb->mark,
2353                         .flowi4_tos   = RT_TOS(ip4h->tos),
2354                         .flowi4_oif   = dev->ifindex,
2355                         .flowi4_proto = ip4h->protocol,
2356                         .daddr        = ip4h->daddr,
2357                         .saddr        = ip4h->saddr,
2358                 };
2359                 struct rtable *rt;
2360
2361                 rt = ip_route_output_flow(net, &fl4, NULL);
2362                 if (IS_ERR(rt))
2363                         goto out_drop;
2364                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2365                         ip_rt_put(rt);
2366                         goto out_drop;
2367                 }
2368
2369                 skb_dst_set(skb, &rt->dst);
2370         }
2371
2372         err = bpf_out_neigh_v4(net, skb, dev, nh);
2373         if (unlikely(net_xmit_eval(err)))
2374                 dev->stats.tx_errors++;
2375         else
2376                 ret = NET_XMIT_SUCCESS;
2377         goto out_xmit;
2378 out_drop:
2379         dev->stats.tx_errors++;
2380         kfree_skb(skb);
2381 out_xmit:
2382         return ret;
2383 }
2384 #else
2385 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2386                                    struct bpf_nh_params *nh)
2387 {
2388         kfree_skb(skb);
2389         return NET_XMIT_DROP;
2390 }
2391 #endif /* CONFIG_INET */
2392
2393 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2394                                 struct bpf_nh_params *nh)
2395 {
2396         struct ethhdr *ethh = eth_hdr(skb);
2397
2398         if (unlikely(skb->mac_header >= skb->network_header))
2399                 goto out;
2400         bpf_push_mac_rcsum(skb);
2401         if (is_multicast_ether_addr(ethh->h_dest))
2402                 goto out;
2403
2404         skb_pull(skb, sizeof(*ethh));
2405         skb_unset_mac_header(skb);
2406         skb_reset_network_header(skb);
2407
2408         if (skb->protocol == htons(ETH_P_IP))
2409                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2410         else if (skb->protocol == htons(ETH_P_IPV6))
2411                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2412 out:
2413         kfree_skb(skb);
2414         return -ENOTSUPP;
2415 }
2416
2417 /* Internal, non-exposed redirect flags. */
2418 enum {
2419         BPF_F_NEIGH     = (1ULL << 1),
2420         BPF_F_PEER      = (1ULL << 2),
2421         BPF_F_NEXTHOP   = (1ULL << 3),
2422 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2423 };
2424
2425 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2426 {
2427         struct net_device *dev;
2428         struct sk_buff *clone;
2429         int ret;
2430
2431         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2432                 return -EINVAL;
2433
2434         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2435         if (unlikely(!dev))
2436                 return -EINVAL;
2437
2438         clone = skb_clone(skb, GFP_ATOMIC);
2439         if (unlikely(!clone))
2440                 return -ENOMEM;
2441
2442         /* For direct write, we need to keep the invariant that the skbs
2443          * we're dealing with need to be uncloned. Should uncloning fail
2444          * here, we need to free the just generated clone to unclone once
2445          * again.
2446          */
2447         ret = bpf_try_make_head_writable(skb);
2448         if (unlikely(ret)) {
2449                 kfree_skb(clone);
2450                 return -ENOMEM;
2451         }
2452
2453         return __bpf_redirect(clone, dev, flags);
2454 }
2455
2456 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2457         .func           = bpf_clone_redirect,
2458         .gpl_only       = false,
2459         .ret_type       = RET_INTEGER,
2460         .arg1_type      = ARG_PTR_TO_CTX,
2461         .arg2_type      = ARG_ANYTHING,
2462         .arg3_type      = ARG_ANYTHING,
2463 };
2464
2465 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2466 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2467
2468 int skb_do_redirect(struct sk_buff *skb)
2469 {
2470         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2471         struct net *net = dev_net(skb->dev);
2472         struct net_device *dev;
2473         u32 flags = ri->flags;
2474
2475         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2476         ri->tgt_index = 0;
2477         ri->flags = 0;
2478         if (unlikely(!dev))
2479                 goto out_drop;
2480         if (flags & BPF_F_PEER) {
2481                 const struct net_device_ops *ops = dev->netdev_ops;
2482
2483                 if (unlikely(!ops->ndo_get_peer_dev ||
2484                              !skb_at_tc_ingress(skb)))
2485                         goto out_drop;
2486                 dev = ops->ndo_get_peer_dev(dev);
2487                 if (unlikely(!dev ||
2488                              !(dev->flags & IFF_UP) ||
2489                              net_eq(net, dev_net(dev))))
2490                         goto out_drop;
2491                 skb->dev = dev;
2492                 dev_sw_netstats_rx_add(dev, skb->len);
2493                 return -EAGAIN;
2494         }
2495         return flags & BPF_F_NEIGH ?
2496                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2497                                     &ri->nh : NULL) :
2498                __bpf_redirect(skb, dev, flags);
2499 out_drop:
2500         kfree_skb(skb);
2501         return -EINVAL;
2502 }
2503
2504 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2505 {
2506         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2507
2508         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2509                 return TC_ACT_SHOT;
2510
2511         ri->flags = flags;
2512         ri->tgt_index = ifindex;
2513
2514         return TC_ACT_REDIRECT;
2515 }
2516
2517 static const struct bpf_func_proto bpf_redirect_proto = {
2518         .func           = bpf_redirect,
2519         .gpl_only       = false,
2520         .ret_type       = RET_INTEGER,
2521         .arg1_type      = ARG_ANYTHING,
2522         .arg2_type      = ARG_ANYTHING,
2523 };
2524
2525 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2526 {
2527         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2528
2529         if (unlikely(flags))
2530                 return TC_ACT_SHOT;
2531
2532         ri->flags = BPF_F_PEER;
2533         ri->tgt_index = ifindex;
2534
2535         return TC_ACT_REDIRECT;
2536 }
2537
2538 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2539         .func           = bpf_redirect_peer,
2540         .gpl_only       = false,
2541         .ret_type       = RET_INTEGER,
2542         .arg1_type      = ARG_ANYTHING,
2543         .arg2_type      = ARG_ANYTHING,
2544 };
2545
2546 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2547            int, plen, u64, flags)
2548 {
2549         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2550
2551         if (unlikely((plen && plen < sizeof(*params)) || flags))
2552                 return TC_ACT_SHOT;
2553
2554         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2555         ri->tgt_index = ifindex;
2556
2557         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2558         if (plen)
2559                 memcpy(&ri->nh, params, sizeof(ri->nh));
2560
2561         return TC_ACT_REDIRECT;
2562 }
2563
2564 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2565         .func           = bpf_redirect_neigh,
2566         .gpl_only       = false,
2567         .ret_type       = RET_INTEGER,
2568         .arg1_type      = ARG_ANYTHING,
2569         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2570         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2571         .arg4_type      = ARG_ANYTHING,
2572 };
2573
2574 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2575 {
2576         msg->apply_bytes = bytes;
2577         return 0;
2578 }
2579
2580 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2581         .func           = bpf_msg_apply_bytes,
2582         .gpl_only       = false,
2583         .ret_type       = RET_INTEGER,
2584         .arg1_type      = ARG_PTR_TO_CTX,
2585         .arg2_type      = ARG_ANYTHING,
2586 };
2587
2588 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2589 {
2590         msg->cork_bytes = bytes;
2591         return 0;
2592 }
2593
2594 static void sk_msg_reset_curr(struct sk_msg *msg)
2595 {
2596         u32 i = msg->sg.start;
2597         u32 len = 0;
2598
2599         do {
2600                 len += sk_msg_elem(msg, i)->length;
2601                 sk_msg_iter_var_next(i);
2602                 if (len >= msg->sg.size)
2603                         break;
2604         } while (i != msg->sg.end);
2605
2606         msg->sg.curr = i;
2607         msg->sg.copybreak = 0;
2608 }
2609
2610 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2611         .func           = bpf_msg_cork_bytes,
2612         .gpl_only       = false,
2613         .ret_type       = RET_INTEGER,
2614         .arg1_type      = ARG_PTR_TO_CTX,
2615         .arg2_type      = ARG_ANYTHING,
2616 };
2617
2618 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2619            u32, end, u64, flags)
2620 {
2621         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2622         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2623         struct scatterlist *sge;
2624         u8 *raw, *to, *from;
2625         struct page *page;
2626
2627         if (unlikely(flags || end <= start))
2628                 return -EINVAL;
2629
2630         /* First find the starting scatterlist element */
2631         i = msg->sg.start;
2632         do {
2633                 offset += len;
2634                 len = sk_msg_elem(msg, i)->length;
2635                 if (start < offset + len)
2636                         break;
2637                 sk_msg_iter_var_next(i);
2638         } while (i != msg->sg.end);
2639
2640         if (unlikely(start >= offset + len))
2641                 return -EINVAL;
2642
2643         first_sge = i;
2644         /* The start may point into the sg element so we need to also
2645          * account for the headroom.
2646          */
2647         bytes_sg_total = start - offset + bytes;
2648         if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2649                 goto out;
2650
2651         /* At this point we need to linearize multiple scatterlist
2652          * elements or a single shared page. Either way we need to
2653          * copy into a linear buffer exclusively owned by BPF. Then
2654          * place the buffer in the scatterlist and fixup the original
2655          * entries by removing the entries now in the linear buffer
2656          * and shifting the remaining entries. For now we do not try
2657          * to copy partial entries to avoid complexity of running out
2658          * of sg_entry slots. The downside is reading a single byte
2659          * will copy the entire sg entry.
2660          */
2661         do {
2662                 copy += sk_msg_elem(msg, i)->length;
2663                 sk_msg_iter_var_next(i);
2664                 if (bytes_sg_total <= copy)
2665                         break;
2666         } while (i != msg->sg.end);
2667         last_sge = i;
2668
2669         if (unlikely(bytes_sg_total > copy))
2670                 return -EINVAL;
2671
2672         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2673                            get_order(copy));
2674         if (unlikely(!page))
2675                 return -ENOMEM;
2676
2677         raw = page_address(page);
2678         i = first_sge;
2679         do {
2680                 sge = sk_msg_elem(msg, i);
2681                 from = sg_virt(sge);
2682                 len = sge->length;
2683                 to = raw + poffset;
2684
2685                 memcpy(to, from, len);
2686                 poffset += len;
2687                 sge->length = 0;
2688                 put_page(sg_page(sge));
2689
2690                 sk_msg_iter_var_next(i);
2691         } while (i != last_sge);
2692
2693         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2694
2695         /* To repair sg ring we need to shift entries. If we only
2696          * had a single entry though we can just replace it and
2697          * be done. Otherwise walk the ring and shift the entries.
2698          */
2699         WARN_ON_ONCE(last_sge == first_sge);
2700         shift = last_sge > first_sge ?
2701                 last_sge - first_sge - 1 :
2702                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2703         if (!shift)
2704                 goto out;
2705
2706         i = first_sge;
2707         sk_msg_iter_var_next(i);
2708         do {
2709                 u32 move_from;
2710
2711                 if (i + shift >= NR_MSG_FRAG_IDS)
2712                         move_from = i + shift - NR_MSG_FRAG_IDS;
2713                 else
2714                         move_from = i + shift;
2715                 if (move_from == msg->sg.end)
2716                         break;
2717
2718                 msg->sg.data[i] = msg->sg.data[move_from];
2719                 msg->sg.data[move_from].length = 0;
2720                 msg->sg.data[move_from].page_link = 0;
2721                 msg->sg.data[move_from].offset = 0;
2722                 sk_msg_iter_var_next(i);
2723         } while (1);
2724
2725         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2726                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2727                       msg->sg.end - shift;
2728 out:
2729         sk_msg_reset_curr(msg);
2730         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2731         msg->data_end = msg->data + bytes;
2732         return 0;
2733 }
2734
2735 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2736         .func           = bpf_msg_pull_data,
2737         .gpl_only       = false,
2738         .ret_type       = RET_INTEGER,
2739         .arg1_type      = ARG_PTR_TO_CTX,
2740         .arg2_type      = ARG_ANYTHING,
2741         .arg3_type      = ARG_ANYTHING,
2742         .arg4_type      = ARG_ANYTHING,
2743 };
2744
2745 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2746            u32, len, u64, flags)
2747 {
2748         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2749         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2750         u8 *raw, *to, *from;
2751         struct page *page;
2752
2753         if (unlikely(flags))
2754                 return -EINVAL;
2755
2756         if (unlikely(len == 0))
2757                 return 0;
2758
2759         /* First find the starting scatterlist element */
2760         i = msg->sg.start;
2761         do {
2762                 offset += l;
2763                 l = sk_msg_elem(msg, i)->length;
2764
2765                 if (start < offset + l)
2766                         break;
2767                 sk_msg_iter_var_next(i);
2768         } while (i != msg->sg.end);
2769
2770         if (start >= offset + l)
2771                 return -EINVAL;
2772
2773         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2774
2775         /* If no space available will fallback to copy, we need at
2776          * least one scatterlist elem available to push data into
2777          * when start aligns to the beginning of an element or two
2778          * when it falls inside an element. We handle the start equals
2779          * offset case because its the common case for inserting a
2780          * header.
2781          */
2782         if (!space || (space == 1 && start != offset))
2783                 copy = msg->sg.data[i].length;
2784
2785         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2786                            get_order(copy + len));
2787         if (unlikely(!page))
2788                 return -ENOMEM;
2789
2790         if (copy) {
2791                 int front, back;
2792
2793                 raw = page_address(page);
2794
2795                 psge = sk_msg_elem(msg, i);
2796                 front = start - offset;
2797                 back = psge->length - front;
2798                 from = sg_virt(psge);
2799
2800                 if (front)
2801                         memcpy(raw, from, front);
2802
2803                 if (back) {
2804                         from += front;
2805                         to = raw + front + len;
2806
2807                         memcpy(to, from, back);
2808                 }
2809
2810                 put_page(sg_page(psge));
2811         } else if (start - offset) {
2812                 psge = sk_msg_elem(msg, i);
2813                 rsge = sk_msg_elem_cpy(msg, i);
2814
2815                 psge->length = start - offset;
2816                 rsge.length -= psge->length;
2817                 rsge.offset += start;
2818
2819                 sk_msg_iter_var_next(i);
2820                 sg_unmark_end(psge);
2821                 sg_unmark_end(&rsge);
2822                 sk_msg_iter_next(msg, end);
2823         }
2824
2825         /* Slot(s) to place newly allocated data */
2826         new = i;
2827
2828         /* Shift one or two slots as needed */
2829         if (!copy) {
2830                 sge = sk_msg_elem_cpy(msg, i);
2831
2832                 sk_msg_iter_var_next(i);
2833                 sg_unmark_end(&sge);
2834                 sk_msg_iter_next(msg, end);
2835
2836                 nsge = sk_msg_elem_cpy(msg, i);
2837                 if (rsge.length) {
2838                         sk_msg_iter_var_next(i);
2839                         nnsge = sk_msg_elem_cpy(msg, i);
2840                 }
2841
2842                 while (i != msg->sg.end) {
2843                         msg->sg.data[i] = sge;
2844                         sge = nsge;
2845                         sk_msg_iter_var_next(i);
2846                         if (rsge.length) {
2847                                 nsge = nnsge;
2848                                 nnsge = sk_msg_elem_cpy(msg, i);
2849                         } else {
2850                                 nsge = sk_msg_elem_cpy(msg, i);
2851                         }
2852                 }
2853         }
2854
2855         /* Place newly allocated data buffer */
2856         sk_mem_charge(msg->sk, len);
2857         msg->sg.size += len;
2858         __clear_bit(new, msg->sg.copy);
2859         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2860         if (rsge.length) {
2861                 get_page(sg_page(&rsge));
2862                 sk_msg_iter_var_next(new);
2863                 msg->sg.data[new] = rsge;
2864         }
2865
2866         sk_msg_reset_curr(msg);
2867         sk_msg_compute_data_pointers(msg);
2868         return 0;
2869 }
2870
2871 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2872         .func           = bpf_msg_push_data,
2873         .gpl_only       = false,
2874         .ret_type       = RET_INTEGER,
2875         .arg1_type      = ARG_PTR_TO_CTX,
2876         .arg2_type      = ARG_ANYTHING,
2877         .arg3_type      = ARG_ANYTHING,
2878         .arg4_type      = ARG_ANYTHING,
2879 };
2880
2881 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2882 {
2883         int prev;
2884
2885         do {
2886                 prev = i;
2887                 sk_msg_iter_var_next(i);
2888                 msg->sg.data[prev] = msg->sg.data[i];
2889         } while (i != msg->sg.end);
2890
2891         sk_msg_iter_prev(msg, end);
2892 }
2893
2894 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2895 {
2896         struct scatterlist tmp, sge;
2897
2898         sk_msg_iter_next(msg, end);
2899         sge = sk_msg_elem_cpy(msg, i);
2900         sk_msg_iter_var_next(i);
2901         tmp = sk_msg_elem_cpy(msg, i);
2902
2903         while (i != msg->sg.end) {
2904                 msg->sg.data[i] = sge;
2905                 sk_msg_iter_var_next(i);
2906                 sge = tmp;
2907                 tmp = sk_msg_elem_cpy(msg, i);
2908         }
2909 }
2910
2911 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2912            u32, len, u64, flags)
2913 {
2914         u32 i = 0, l = 0, space, offset = 0;
2915         u64 last = start + len;
2916         int pop;
2917
2918         if (unlikely(flags))
2919                 return -EINVAL;
2920
2921         /* First find the starting scatterlist element */
2922         i = msg->sg.start;
2923         do {
2924                 offset += l;
2925                 l = sk_msg_elem(msg, i)->length;
2926
2927                 if (start < offset + l)
2928                         break;
2929                 sk_msg_iter_var_next(i);
2930         } while (i != msg->sg.end);
2931
2932         /* Bounds checks: start and pop must be inside message */
2933         if (start >= offset + l || last >= msg->sg.size)
2934                 return -EINVAL;
2935
2936         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2937
2938         pop = len;
2939         /* --------------| offset
2940          * -| start      |-------- len -------|
2941          *
2942          *  |----- a ----|-------- pop -------|----- b ----|
2943          *  |______________________________________________| length
2944          *
2945          *
2946          * a:   region at front of scatter element to save
2947          * b:   region at back of scatter element to save when length > A + pop
2948          * pop: region to pop from element, same as input 'pop' here will be
2949          *      decremented below per iteration.
2950          *
2951          * Two top-level cases to handle when start != offset, first B is non
2952          * zero and second B is zero corresponding to when a pop includes more
2953          * than one element.
2954          *
2955          * Then if B is non-zero AND there is no space allocate space and
2956          * compact A, B regions into page. If there is space shift ring to
2957          * the rigth free'ing the next element in ring to place B, leaving
2958          * A untouched except to reduce length.
2959          */
2960         if (start != offset) {
2961                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2962                 int a = start;
2963                 int b = sge->length - pop - a;
2964
2965                 sk_msg_iter_var_next(i);
2966
2967                 if (pop < sge->length - a) {
2968                         if (space) {
2969                                 sge->length = a;
2970                                 sk_msg_shift_right(msg, i);
2971                                 nsge = sk_msg_elem(msg, i);
2972                                 get_page(sg_page(sge));
2973                                 sg_set_page(nsge,
2974                                             sg_page(sge),
2975                                             b, sge->offset + pop + a);
2976                         } else {
2977                                 struct page *page, *orig;
2978                                 u8 *to, *from;
2979
2980                                 page = alloc_pages(__GFP_NOWARN |
2981                                                    __GFP_COMP   | GFP_ATOMIC,
2982                                                    get_order(a + b));
2983                                 if (unlikely(!page))
2984                                         return -ENOMEM;
2985
2986                                 sge->length = a;
2987                                 orig = sg_page(sge);
2988                                 from = sg_virt(sge);
2989                                 to = page_address(page);
2990                                 memcpy(to, from, a);
2991                                 memcpy(to + a, from + a + pop, b);
2992                                 sg_set_page(sge, page, a + b, 0);
2993                                 put_page(orig);
2994                         }
2995                         pop = 0;
2996                 } else if (pop >= sge->length - a) {
2997                         pop -= (sge->length - a);
2998                         sge->length = a;
2999                 }
3000         }
3001
3002         /* From above the current layout _must_ be as follows,
3003          *
3004          * -| offset
3005          * -| start
3006          *
3007          *  |---- pop ---|---------------- b ------------|
3008          *  |____________________________________________| length
3009          *
3010          * Offset and start of the current msg elem are equal because in the
3011          * previous case we handled offset != start and either consumed the
3012          * entire element and advanced to the next element OR pop == 0.
3013          *
3014          * Two cases to handle here are first pop is less than the length
3015          * leaving some remainder b above. Simply adjust the element's layout
3016          * in this case. Or pop >= length of the element so that b = 0. In this
3017          * case advance to next element decrementing pop.
3018          */
3019         while (pop) {
3020                 struct scatterlist *sge = sk_msg_elem(msg, i);
3021
3022                 if (pop < sge->length) {
3023                         sge->length -= pop;
3024                         sge->offset += pop;
3025                         pop = 0;
3026                 } else {
3027                         pop -= sge->length;
3028                         sk_msg_shift_left(msg, i);
3029                 }
3030                 sk_msg_iter_var_next(i);
3031         }
3032
3033         sk_mem_uncharge(msg->sk, len - pop);
3034         msg->sg.size -= (len - pop);
3035         sk_msg_reset_curr(msg);
3036         sk_msg_compute_data_pointers(msg);
3037         return 0;
3038 }
3039
3040 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3041         .func           = bpf_msg_pop_data,
3042         .gpl_only       = false,
3043         .ret_type       = RET_INTEGER,
3044         .arg1_type      = ARG_PTR_TO_CTX,
3045         .arg2_type      = ARG_ANYTHING,
3046         .arg3_type      = ARG_ANYTHING,
3047         .arg4_type      = ARG_ANYTHING,
3048 };
3049
3050 #ifdef CONFIG_CGROUP_NET_CLASSID
3051 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3052 {
3053         return __task_get_classid(current);
3054 }
3055
3056 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3057         .func           = bpf_get_cgroup_classid_curr,
3058         .gpl_only       = false,
3059         .ret_type       = RET_INTEGER,
3060 };
3061
3062 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3063 {
3064         struct sock *sk = skb_to_full_sk(skb);
3065
3066         if (!sk || !sk_fullsock(sk))
3067                 return 0;
3068
3069         return sock_cgroup_classid(&sk->sk_cgrp_data);
3070 }
3071
3072 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3073         .func           = bpf_skb_cgroup_classid,
3074         .gpl_only       = false,
3075         .ret_type       = RET_INTEGER,
3076         .arg1_type      = ARG_PTR_TO_CTX,
3077 };
3078 #endif
3079
3080 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3081 {
3082         return task_get_classid(skb);
3083 }
3084
3085 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3086         .func           = bpf_get_cgroup_classid,
3087         .gpl_only       = false,
3088         .ret_type       = RET_INTEGER,
3089         .arg1_type      = ARG_PTR_TO_CTX,
3090 };
3091
3092 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3093 {
3094         return dst_tclassid(skb);
3095 }
3096
3097 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3098         .func           = bpf_get_route_realm,
3099         .gpl_only       = false,
3100         .ret_type       = RET_INTEGER,
3101         .arg1_type      = ARG_PTR_TO_CTX,
3102 };
3103
3104 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3105 {
3106         /* If skb_clear_hash() was called due to mangling, we can
3107          * trigger SW recalculation here. Later access to hash
3108          * can then use the inline skb->hash via context directly
3109          * instead of calling this helper again.
3110          */
3111         return skb_get_hash(skb);
3112 }
3113
3114 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3115         .func           = bpf_get_hash_recalc,
3116         .gpl_only       = false,
3117         .ret_type       = RET_INTEGER,
3118         .arg1_type      = ARG_PTR_TO_CTX,
3119 };
3120
3121 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3122 {
3123         /* After all direct packet write, this can be used once for
3124          * triggering a lazy recalc on next skb_get_hash() invocation.
3125          */
3126         skb_clear_hash(skb);
3127         return 0;
3128 }
3129
3130 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3131         .func           = bpf_set_hash_invalid,
3132         .gpl_only       = false,
3133         .ret_type       = RET_INTEGER,
3134         .arg1_type      = ARG_PTR_TO_CTX,
3135 };
3136
3137 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3138 {
3139         /* Set user specified hash as L4(+), so that it gets returned
3140          * on skb_get_hash() call unless BPF prog later on triggers a
3141          * skb_clear_hash().
3142          */
3143         __skb_set_sw_hash(skb, hash, true);
3144         return 0;
3145 }
3146
3147 static const struct bpf_func_proto bpf_set_hash_proto = {
3148         .func           = bpf_set_hash,
3149         .gpl_only       = false,
3150         .ret_type       = RET_INTEGER,
3151         .arg1_type      = ARG_PTR_TO_CTX,
3152         .arg2_type      = ARG_ANYTHING,
3153 };
3154
3155 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3156            u16, vlan_tci)
3157 {
3158         int ret;
3159
3160         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3161                      vlan_proto != htons(ETH_P_8021AD)))
3162                 vlan_proto = htons(ETH_P_8021Q);
3163
3164         bpf_push_mac_rcsum(skb);
3165         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3166         bpf_pull_mac_rcsum(skb);
3167
3168         bpf_compute_data_pointers(skb);
3169         return ret;
3170 }
3171
3172 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3173         .func           = bpf_skb_vlan_push,
3174         .gpl_only       = false,
3175         .ret_type       = RET_INTEGER,
3176         .arg1_type      = ARG_PTR_TO_CTX,
3177         .arg2_type      = ARG_ANYTHING,
3178         .arg3_type      = ARG_ANYTHING,
3179 };
3180
3181 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3182 {
3183         int ret;
3184
3185         bpf_push_mac_rcsum(skb);
3186         ret = skb_vlan_pop(skb);
3187         bpf_pull_mac_rcsum(skb);
3188
3189         bpf_compute_data_pointers(skb);
3190         return ret;
3191 }
3192
3193 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3194         .func           = bpf_skb_vlan_pop,
3195         .gpl_only       = false,
3196         .ret_type       = RET_INTEGER,
3197         .arg1_type      = ARG_PTR_TO_CTX,
3198 };
3199
3200 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3201 {
3202         /* Caller already did skb_cow() with len as headroom,
3203          * so no need to do it here.
3204          */
3205         skb_push(skb, len);
3206         memmove(skb->data, skb->data + len, off);
3207         memset(skb->data + off, 0, len);
3208
3209         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3210          * needed here as it does not change the skb->csum
3211          * result for checksum complete when summing over
3212          * zeroed blocks.
3213          */
3214         return 0;
3215 }
3216
3217 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3218 {
3219         void *old_data;
3220
3221         /* skb_ensure_writable() is not needed here, as we're
3222          * already working on an uncloned skb.
3223          */
3224         if (unlikely(!pskb_may_pull(skb, off + len)))
3225                 return -ENOMEM;
3226
3227         old_data = skb->data;
3228         __skb_pull(skb, len);
3229         skb_postpull_rcsum(skb, old_data + off, len);
3230         memmove(skb->data, old_data, off);
3231
3232         return 0;
3233 }
3234
3235 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3236 {
3237         bool trans_same = skb->transport_header == skb->network_header;
3238         int ret;
3239
3240         /* There's no need for __skb_push()/__skb_pull() pair to
3241          * get to the start of the mac header as we're guaranteed
3242          * to always start from here under eBPF.
3243          */
3244         ret = bpf_skb_generic_push(skb, off, len);
3245         if (likely(!ret)) {
3246                 skb->mac_header -= len;
3247                 skb->network_header -= len;
3248                 if (trans_same)
3249                         skb->transport_header = skb->network_header;
3250         }
3251
3252         return ret;
3253 }
3254
3255 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3256 {
3257         bool trans_same = skb->transport_header == skb->network_header;
3258         int ret;
3259
3260         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3261         ret = bpf_skb_generic_pop(skb, off, len);
3262         if (likely(!ret)) {
3263                 skb->mac_header += len;
3264                 skb->network_header += len;
3265                 if (trans_same)
3266                         skb->transport_header = skb->network_header;
3267         }
3268
3269         return ret;
3270 }
3271
3272 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3273 {
3274         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3275         u32 off = skb_mac_header_len(skb);
3276         int ret;
3277
3278         ret = skb_cow(skb, len_diff);
3279         if (unlikely(ret < 0))
3280                 return ret;
3281
3282         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3283         if (unlikely(ret < 0))
3284                 return ret;
3285
3286         if (skb_is_gso(skb)) {
3287                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3288
3289                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3290                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3291                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3292                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3293                 }
3294         }
3295
3296         skb->protocol = htons(ETH_P_IPV6);
3297         skb_clear_hash(skb);
3298
3299         return 0;
3300 }
3301
3302 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3303 {
3304         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3305         u32 off = skb_mac_header_len(skb);
3306         int ret;
3307
3308         ret = skb_unclone(skb, GFP_ATOMIC);
3309         if (unlikely(ret < 0))
3310                 return ret;
3311
3312         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3313         if (unlikely(ret < 0))
3314                 return ret;
3315
3316         if (skb_is_gso(skb)) {
3317                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3318
3319                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3320                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3321                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3322                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3323                 }
3324         }
3325
3326         skb->protocol = htons(ETH_P_IP);
3327         skb_clear_hash(skb);
3328
3329         return 0;
3330 }
3331
3332 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3333 {
3334         __be16 from_proto = skb->protocol;
3335
3336         if (from_proto == htons(ETH_P_IP) &&
3337               to_proto == htons(ETH_P_IPV6))
3338                 return bpf_skb_proto_4_to_6(skb);
3339
3340         if (from_proto == htons(ETH_P_IPV6) &&
3341               to_proto == htons(ETH_P_IP))
3342                 return bpf_skb_proto_6_to_4(skb);
3343
3344         return -ENOTSUPP;
3345 }
3346
3347 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3348            u64, flags)
3349 {
3350         int ret;
3351
3352         if (unlikely(flags))
3353                 return -EINVAL;
3354
3355         /* General idea is that this helper does the basic groundwork
3356          * needed for changing the protocol, and eBPF program fills the
3357          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3358          * and other helpers, rather than passing a raw buffer here.
3359          *
3360          * The rationale is to keep this minimal and without a need to
3361          * deal with raw packet data. F.e. even if we would pass buffers
3362          * here, the program still needs to call the bpf_lX_csum_replace()
3363          * helpers anyway. Plus, this way we keep also separation of
3364          * concerns, since f.e. bpf_skb_store_bytes() should only take
3365          * care of stores.
3366          *
3367          * Currently, additional options and extension header space are
3368          * not supported, but flags register is reserved so we can adapt
3369          * that. For offloads, we mark packet as dodgy, so that headers
3370          * need to be verified first.
3371          */
3372         ret = bpf_skb_proto_xlat(skb, proto);
3373         bpf_compute_data_pointers(skb);
3374         return ret;
3375 }
3376
3377 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3378         .func           = bpf_skb_change_proto,
3379         .gpl_only       = false,
3380         .ret_type       = RET_INTEGER,
3381         .arg1_type      = ARG_PTR_TO_CTX,
3382         .arg2_type      = ARG_ANYTHING,
3383         .arg3_type      = ARG_ANYTHING,
3384 };
3385
3386 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3387 {
3388         /* We only allow a restricted subset to be changed for now. */
3389         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3390                      !skb_pkt_type_ok(pkt_type)))
3391                 return -EINVAL;
3392
3393         skb->pkt_type = pkt_type;
3394         return 0;
3395 }
3396
3397 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3398         .func           = bpf_skb_change_type,
3399         .gpl_only       = false,
3400         .ret_type       = RET_INTEGER,
3401         .arg1_type      = ARG_PTR_TO_CTX,
3402         .arg2_type      = ARG_ANYTHING,
3403 };
3404
3405 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3406 {
3407         switch (skb->protocol) {
3408         case htons(ETH_P_IP):
3409                 return sizeof(struct iphdr);
3410         case htons(ETH_P_IPV6):
3411                 return sizeof(struct ipv6hdr);
3412         default:
3413                 return ~0U;
3414         }
3415 }
3416
3417 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3418                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3419
3420 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK    (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3421                                          BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3422
3423 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3424                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3425                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3426                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3427                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3428                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3429                                           BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3430                                          BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3431
3432 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3433                             u64 flags)
3434 {
3435         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3436         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3437         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3438         unsigned int gso_type = SKB_GSO_DODGY;
3439         int ret;
3440
3441         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3442                 /* udp gso_size delineates datagrams, only allow if fixed */
3443                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3444                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3445                         return -ENOTSUPP;
3446         }
3447
3448         ret = skb_cow_head(skb, len_diff);
3449         if (unlikely(ret < 0))
3450                 return ret;
3451
3452         if (encap) {
3453                 if (skb->protocol != htons(ETH_P_IP) &&
3454                     skb->protocol != htons(ETH_P_IPV6))
3455                         return -ENOTSUPP;
3456
3457                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3458                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3459                         return -EINVAL;
3460
3461                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3462                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3463                         return -EINVAL;
3464
3465                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3466                     inner_mac_len < ETH_HLEN)
3467                         return -EINVAL;
3468
3469                 if (skb->encapsulation)
3470                         return -EALREADY;
3471
3472                 mac_len = skb->network_header - skb->mac_header;
3473                 inner_net = skb->network_header;
3474                 if (inner_mac_len > len_diff)
3475                         return -EINVAL;
3476                 inner_trans = skb->transport_header;
3477         }
3478
3479         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3480         if (unlikely(ret < 0))
3481                 return ret;
3482
3483         if (encap) {
3484                 skb->inner_mac_header = inner_net - inner_mac_len;
3485                 skb->inner_network_header = inner_net;
3486                 skb->inner_transport_header = inner_trans;
3487
3488                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3489                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3490                 else
3491                         skb_set_inner_protocol(skb, skb->protocol);
3492
3493                 skb->encapsulation = 1;
3494                 skb_set_network_header(skb, mac_len);
3495
3496                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3497                         gso_type |= SKB_GSO_UDP_TUNNEL;
3498                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3499                         gso_type |= SKB_GSO_GRE;
3500                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3501                         gso_type |= SKB_GSO_IPXIP6;
3502                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3503                         gso_type |= SKB_GSO_IPXIP4;
3504
3505                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3506                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3507                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3508                                         sizeof(struct ipv6hdr) :
3509                                         sizeof(struct iphdr);
3510
3511                         skb_set_transport_header(skb, mac_len + nh_len);
3512                 }
3513
3514                 /* Match skb->protocol to new outer l3 protocol */
3515                 if (skb->protocol == htons(ETH_P_IP) &&
3516                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3517                         skb->protocol = htons(ETH_P_IPV6);
3518                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3519                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3520                         skb->protocol = htons(ETH_P_IP);
3521         }
3522
3523         if (skb_is_gso(skb)) {
3524                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3525
3526                 /* Due to header grow, MSS needs to be downgraded. */
3527                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3528                         skb_decrease_gso_size(shinfo, len_diff);
3529
3530                 /* Header must be checked, and gso_segs recomputed. */
3531                 shinfo->gso_type |= gso_type;
3532                 shinfo->gso_segs = 0;
3533         }
3534
3535         return 0;
3536 }
3537
3538 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3539                               u64 flags)
3540 {
3541         int ret;
3542
3543         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3544                                BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3545                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3546                 return -EINVAL;
3547
3548         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3549                 /* udp gso_size delineates datagrams, only allow if fixed */
3550                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3551                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3552                         return -ENOTSUPP;
3553         }
3554
3555         ret = skb_unclone(skb, GFP_ATOMIC);
3556         if (unlikely(ret < 0))
3557                 return ret;
3558
3559         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3560         if (unlikely(ret < 0))
3561                 return ret;
3562
3563         /* Match skb->protocol to new outer l3 protocol */
3564         if (skb->protocol == htons(ETH_P_IP) &&
3565             flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3566                 skb->protocol = htons(ETH_P_IPV6);
3567         else if (skb->protocol == htons(ETH_P_IPV6) &&
3568                  flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3569                 skb->protocol = htons(ETH_P_IP);
3570
3571         if (skb_is_gso(skb)) {
3572                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3573
3574                 /* Due to header shrink, MSS can be upgraded. */
3575                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3576                         skb_increase_gso_size(shinfo, len_diff);
3577
3578                 /* Header must be checked, and gso_segs recomputed. */
3579                 shinfo->gso_type |= SKB_GSO_DODGY;
3580                 shinfo->gso_segs = 0;
3581         }
3582
3583         return 0;
3584 }
3585
3586 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3587
3588 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3589            u32, mode, u64, flags)
3590 {
3591         u32 len_diff_abs = abs(len_diff);
3592         bool shrink = len_diff < 0;
3593         int ret = 0;
3594
3595         if (unlikely(flags || mode))
3596                 return -EINVAL;
3597         if (unlikely(len_diff_abs > 0xfffU))
3598                 return -EFAULT;
3599
3600         if (!shrink) {
3601                 ret = skb_cow(skb, len_diff);
3602                 if (unlikely(ret < 0))
3603                         return ret;
3604                 __skb_push(skb, len_diff_abs);
3605                 memset(skb->data, 0, len_diff_abs);
3606         } else {
3607                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3608                         return -ENOMEM;
3609                 __skb_pull(skb, len_diff_abs);
3610         }
3611         if (tls_sw_has_ctx_rx(skb->sk)) {
3612                 struct strp_msg *rxm = strp_msg(skb);
3613
3614                 rxm->full_len += len_diff;
3615         }
3616         return ret;
3617 }
3618
3619 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3620         .func           = sk_skb_adjust_room,
3621         .gpl_only       = false,
3622         .ret_type       = RET_INTEGER,
3623         .arg1_type      = ARG_PTR_TO_CTX,
3624         .arg2_type      = ARG_ANYTHING,
3625         .arg3_type      = ARG_ANYTHING,
3626         .arg4_type      = ARG_ANYTHING,
3627 };
3628
3629 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3630            u32, mode, u64, flags)
3631 {
3632         u32 len_cur, len_diff_abs = abs(len_diff);
3633         u32 len_min = bpf_skb_net_base_len(skb);
3634         u32 len_max = BPF_SKB_MAX_LEN;
3635         __be16 proto = skb->protocol;
3636         bool shrink = len_diff < 0;
3637         u32 off;
3638         int ret;
3639
3640         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3641                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3642                 return -EINVAL;
3643         if (unlikely(len_diff_abs > 0xfffU))
3644                 return -EFAULT;
3645         if (unlikely(proto != htons(ETH_P_IP) &&
3646                      proto != htons(ETH_P_IPV6)))
3647                 return -ENOTSUPP;
3648
3649         off = skb_mac_header_len(skb);
3650         switch (mode) {
3651         case BPF_ADJ_ROOM_NET:
3652                 off += bpf_skb_net_base_len(skb);
3653                 break;
3654         case BPF_ADJ_ROOM_MAC:
3655                 break;
3656         default:
3657                 return -ENOTSUPP;
3658         }
3659
3660         if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3661                 if (!shrink)
3662                         return -EINVAL;
3663
3664                 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3665                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3666                         len_min = sizeof(struct iphdr);
3667                         break;
3668                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3669                         len_min = sizeof(struct ipv6hdr);
3670                         break;
3671                 default:
3672                         return -EINVAL;
3673                 }
3674         }
3675
3676         len_cur = skb->len - skb_network_offset(skb);
3677         if ((shrink && (len_diff_abs >= len_cur ||
3678                         len_cur - len_diff_abs < len_min)) ||
3679             (!shrink && (skb->len + len_diff_abs > len_max &&
3680                          !skb_is_gso(skb))))
3681                 return -ENOTSUPP;
3682
3683         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3684                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3685         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3686                 __skb_reset_checksum_unnecessary(skb);
3687
3688         bpf_compute_data_pointers(skb);
3689         return ret;
3690 }
3691
3692 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3693         .func           = bpf_skb_adjust_room,
3694         .gpl_only       = false,
3695         .ret_type       = RET_INTEGER,
3696         .arg1_type      = ARG_PTR_TO_CTX,
3697         .arg2_type      = ARG_ANYTHING,
3698         .arg3_type      = ARG_ANYTHING,
3699         .arg4_type      = ARG_ANYTHING,
3700 };
3701
3702 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3703 {
3704         u32 min_len = skb_network_offset(skb);
3705
3706         if (skb_transport_header_was_set(skb))
3707                 min_len = skb_transport_offset(skb);
3708         if (skb->ip_summed == CHECKSUM_PARTIAL)
3709                 min_len = skb_checksum_start_offset(skb) +
3710                           skb->csum_offset + sizeof(__sum16);
3711         return min_len;
3712 }
3713
3714 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3715 {
3716         unsigned int old_len = skb->len;
3717         int ret;
3718
3719         ret = __skb_grow_rcsum(skb, new_len);
3720         if (!ret)
3721                 memset(skb->data + old_len, 0, new_len - old_len);
3722         return ret;
3723 }
3724
3725 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3726 {
3727         return __skb_trim_rcsum(skb, new_len);
3728 }
3729
3730 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3731                                         u64 flags)
3732 {
3733         u32 max_len = BPF_SKB_MAX_LEN;
3734         u32 min_len = __bpf_skb_min_len(skb);
3735         int ret;
3736
3737         if (unlikely(flags || new_len > max_len || new_len < min_len))
3738                 return -EINVAL;
3739         if (skb->encapsulation)
3740                 return -ENOTSUPP;
3741
3742         /* The basic idea of this helper is that it's performing the
3743          * needed work to either grow or trim an skb, and eBPF program
3744          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3745          * bpf_lX_csum_replace() and others rather than passing a raw
3746          * buffer here. This one is a slow path helper and intended
3747          * for replies with control messages.
3748          *
3749          * Like in bpf_skb_change_proto(), we want to keep this rather
3750          * minimal and without protocol specifics so that we are able
3751          * to separate concerns as in bpf_skb_store_bytes() should only
3752          * be the one responsible for writing buffers.
3753          *
3754          * It's really expected to be a slow path operation here for
3755          * control message replies, so we're implicitly linearizing,
3756          * uncloning and drop offloads from the skb by this.
3757          */
3758         ret = __bpf_try_make_writable(skb, skb->len);
3759         if (!ret) {
3760                 if (new_len > skb->len)
3761                         ret = bpf_skb_grow_rcsum(skb, new_len);
3762                 else if (new_len < skb->len)
3763                         ret = bpf_skb_trim_rcsum(skb, new_len);
3764                 if (!ret && skb_is_gso(skb))
3765                         skb_gso_reset(skb);
3766         }
3767         return ret;
3768 }
3769
3770 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3771            u64, flags)
3772 {
3773         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3774
3775         bpf_compute_data_pointers(skb);
3776         return ret;
3777 }
3778
3779 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3780         .func           = bpf_skb_change_tail,
3781         .gpl_only       = false,
3782         .ret_type       = RET_INTEGER,
3783         .arg1_type      = ARG_PTR_TO_CTX,
3784         .arg2_type      = ARG_ANYTHING,
3785         .arg3_type      = ARG_ANYTHING,
3786 };
3787
3788 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3789            u64, flags)
3790 {
3791         return __bpf_skb_change_tail(skb, new_len, flags);
3792 }
3793
3794 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3795         .func           = sk_skb_change_tail,
3796         .gpl_only       = false,
3797         .ret_type       = RET_INTEGER,
3798         .arg1_type      = ARG_PTR_TO_CTX,
3799         .arg2_type      = ARG_ANYTHING,
3800         .arg3_type      = ARG_ANYTHING,
3801 };
3802
3803 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3804                                         u64 flags)
3805 {
3806         u32 max_len = BPF_SKB_MAX_LEN;
3807         u32 new_len = skb->len + head_room;
3808         int ret;
3809
3810         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3811                      new_len < skb->len))
3812                 return -EINVAL;
3813
3814         ret = skb_cow(skb, head_room);
3815         if (likely(!ret)) {
3816                 /* Idea for this helper is that we currently only
3817                  * allow to expand on mac header. This means that
3818                  * skb->protocol network header, etc, stay as is.
3819                  * Compared to bpf_skb_change_tail(), we're more
3820                  * flexible due to not needing to linearize or
3821                  * reset GSO. Intention for this helper is to be
3822                  * used by an L3 skb that needs to push mac header
3823                  * for redirection into L2 device.
3824                  */
3825                 __skb_push(skb, head_room);
3826                 memset(skb->data, 0, head_room);
3827                 skb_reset_mac_header(skb);
3828                 skb_reset_mac_len(skb);
3829         }
3830
3831         return ret;
3832 }
3833
3834 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3835            u64, flags)
3836 {
3837         int ret = __bpf_skb_change_head(skb, head_room, flags);
3838
3839         bpf_compute_data_pointers(skb);
3840         return ret;
3841 }
3842
3843 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3844         .func           = bpf_skb_change_head,
3845         .gpl_only       = false,
3846         .ret_type       = RET_INTEGER,
3847         .arg1_type      = ARG_PTR_TO_CTX,
3848         .arg2_type      = ARG_ANYTHING,
3849         .arg3_type      = ARG_ANYTHING,
3850 };
3851
3852 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3853            u64, flags)
3854 {
3855         return __bpf_skb_change_head(skb, head_room, flags);
3856 }
3857
3858 static const struct bpf_func_proto sk_skb_change_head_proto = {
3859         .func           = sk_skb_change_head,
3860         .gpl_only       = false,
3861         .ret_type       = RET_INTEGER,
3862         .arg1_type      = ARG_PTR_TO_CTX,
3863         .arg2_type      = ARG_ANYTHING,
3864         .arg3_type      = ARG_ANYTHING,
3865 };
3866
3867 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3868 {
3869         return xdp_get_buff_len(xdp);
3870 }
3871
3872 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3873         .func           = bpf_xdp_get_buff_len,
3874         .gpl_only       = false,
3875         .ret_type       = RET_INTEGER,
3876         .arg1_type      = ARG_PTR_TO_CTX,
3877 };
3878
3879 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3880
3881 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3882         .func           = bpf_xdp_get_buff_len,
3883         .gpl_only       = false,
3884         .arg1_type      = ARG_PTR_TO_BTF_ID,
3885         .arg1_btf_id    = &bpf_xdp_get_buff_len_bpf_ids[0],
3886 };
3887
3888 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3889 {
3890         return xdp_data_meta_unsupported(xdp) ? 0 :
3891                xdp->data - xdp->data_meta;
3892 }
3893
3894 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3895 {
3896         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3897         unsigned long metalen = xdp_get_metalen(xdp);
3898         void *data_start = xdp_frame_end + metalen;
3899         void *data = xdp->data + offset;
3900
3901         if (unlikely(data < data_start ||
3902                      data > xdp->data_end - ETH_HLEN))
3903                 return -EINVAL;
3904
3905         if (metalen)
3906                 memmove(xdp->data_meta + offset,
3907                         xdp->data_meta, metalen);
3908         xdp->data_meta += offset;
3909         xdp->data = data;
3910
3911         return 0;
3912 }
3913
3914 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3915         .func           = bpf_xdp_adjust_head,
3916         .gpl_only       = false,
3917         .ret_type       = RET_INTEGER,
3918         .arg1_type      = ARG_PTR_TO_CTX,
3919         .arg2_type      = ARG_ANYTHING,
3920 };
3921
3922 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3923                       void *buf, unsigned long len, bool flush)
3924 {
3925         unsigned long ptr_len, ptr_off = 0;
3926         skb_frag_t *next_frag, *end_frag;
3927         struct skb_shared_info *sinfo;
3928         void *src, *dst;
3929         u8 *ptr_buf;
3930
3931         if (likely(xdp->data_end - xdp->data >= off + len)) {
3932                 src = flush ? buf : xdp->data + off;
3933                 dst = flush ? xdp->data + off : buf;
3934                 memcpy(dst, src, len);
3935                 return;
3936         }
3937
3938         sinfo = xdp_get_shared_info_from_buff(xdp);
3939         end_frag = &sinfo->frags[sinfo->nr_frags];
3940         next_frag = &sinfo->frags[0];
3941
3942         ptr_len = xdp->data_end - xdp->data;
3943         ptr_buf = xdp->data;
3944
3945         while (true) {
3946                 if (off < ptr_off + ptr_len) {
3947                         unsigned long copy_off = off - ptr_off;
3948                         unsigned long copy_len = min(len, ptr_len - copy_off);
3949
3950                         src = flush ? buf : ptr_buf + copy_off;
3951                         dst = flush ? ptr_buf + copy_off : buf;
3952                         memcpy(dst, src, copy_len);
3953
3954                         off += copy_len;
3955                         len -= copy_len;
3956                         buf += copy_len;
3957                 }
3958
3959                 if (!len || next_frag == end_frag)
3960                         break;
3961
3962                 ptr_off += ptr_len;
3963                 ptr_buf = skb_frag_address(next_frag);
3964                 ptr_len = skb_frag_size(next_frag);
3965                 next_frag++;
3966         }
3967 }
3968
3969 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3970 {
3971         u32 size = xdp->data_end - xdp->data;
3972         struct skb_shared_info *sinfo;
3973         void *addr = xdp->data;
3974         int i;
3975
3976         if (unlikely(offset > 0xffff || len > 0xffff))
3977                 return ERR_PTR(-EFAULT);
3978
3979         if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3980                 return ERR_PTR(-EINVAL);
3981
3982         if (likely(offset < size)) /* linear area */
3983                 goto out;
3984
3985         sinfo = xdp_get_shared_info_from_buff(xdp);
3986         offset -= size;
3987         for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3988                 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3989
3990                 if  (offset < frag_size) {
3991                         addr = skb_frag_address(&sinfo->frags[i]);
3992                         size = frag_size;
3993                         break;
3994                 }
3995                 offset -= frag_size;
3996         }
3997 out:
3998         return offset + len <= size ? addr + offset : NULL;
3999 }
4000
4001 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
4002            void *, buf, u32, len)
4003 {
4004         void *ptr;
4005
4006         ptr = bpf_xdp_pointer(xdp, offset, len);
4007         if (IS_ERR(ptr))
4008                 return PTR_ERR(ptr);
4009
4010         if (!ptr)
4011                 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
4012         else
4013                 memcpy(buf, ptr, len);
4014
4015         return 0;
4016 }
4017
4018 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
4019         .func           = bpf_xdp_load_bytes,
4020         .gpl_only       = false,
4021         .ret_type       = RET_INTEGER,
4022         .arg1_type      = ARG_PTR_TO_CTX,
4023         .arg2_type      = ARG_ANYTHING,
4024         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4025         .arg4_type      = ARG_CONST_SIZE,
4026 };
4027
4028 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4029 {
4030         return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4031 }
4032
4033 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4034            void *, buf, u32, len)
4035 {
4036         void *ptr;
4037
4038         ptr = bpf_xdp_pointer(xdp, offset, len);
4039         if (IS_ERR(ptr))
4040                 return PTR_ERR(ptr);
4041
4042         if (!ptr)
4043                 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4044         else
4045                 memcpy(ptr, buf, len);
4046
4047         return 0;
4048 }
4049
4050 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4051         .func           = bpf_xdp_store_bytes,
4052         .gpl_only       = false,
4053         .ret_type       = RET_INTEGER,
4054         .arg1_type      = ARG_PTR_TO_CTX,
4055         .arg2_type      = ARG_ANYTHING,
4056         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4057         .arg4_type      = ARG_CONST_SIZE,
4058 };
4059
4060 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4061 {
4062         return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4063 }
4064
4065 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4066 {
4067         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4068         skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4069         struct xdp_rxq_info *rxq = xdp->rxq;
4070         unsigned int tailroom;
4071
4072         if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4073                 return -EOPNOTSUPP;
4074
4075         tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4076         if (unlikely(offset > tailroom))
4077                 return -EINVAL;
4078
4079         memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4080         skb_frag_size_add(frag, offset);
4081         sinfo->xdp_frags_size += offset;
4082
4083         return 0;
4084 }
4085
4086 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4087 {
4088         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4089         int i, n_frags_free = 0, len_free = 0;
4090
4091         if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4092                 return -EINVAL;
4093
4094         for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4095                 skb_frag_t *frag = &sinfo->frags[i];
4096                 int shrink = min_t(int, offset, skb_frag_size(frag));
4097
4098                 len_free += shrink;
4099                 offset -= shrink;
4100
4101                 if (skb_frag_size(frag) == shrink) {
4102                         struct page *page = skb_frag_page(frag);
4103
4104                         __xdp_return(page_address(page), &xdp->rxq->mem,
4105                                      false, NULL);
4106                         n_frags_free++;
4107                 } else {
4108                         skb_frag_size_sub(frag, shrink);
4109                         break;
4110                 }
4111         }
4112         sinfo->nr_frags -= n_frags_free;
4113         sinfo->xdp_frags_size -= len_free;
4114
4115         if (unlikely(!sinfo->nr_frags)) {
4116                 xdp_buff_clear_frags_flag(xdp);
4117                 xdp->data_end -= offset;
4118         }
4119
4120         return 0;
4121 }
4122
4123 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4124 {
4125         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4126         void *data_end = xdp->data_end + offset;
4127
4128         if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4129                 if (offset < 0)
4130                         return bpf_xdp_frags_shrink_tail(xdp, -offset);
4131
4132                 return bpf_xdp_frags_increase_tail(xdp, offset);
4133         }
4134
4135         /* Notice that xdp_data_hard_end have reserved some tailroom */
4136         if (unlikely(data_end > data_hard_end))
4137                 return -EINVAL;
4138
4139         if (unlikely(data_end < xdp->data + ETH_HLEN))
4140                 return -EINVAL;
4141
4142         /* Clear memory area on grow, can contain uninit kernel memory */
4143         if (offset > 0)
4144                 memset(xdp->data_end, 0, offset);
4145
4146         xdp->data_end = data_end;
4147
4148         return 0;
4149 }
4150
4151 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4152         .func           = bpf_xdp_adjust_tail,
4153         .gpl_only       = false,
4154         .ret_type       = RET_INTEGER,
4155         .arg1_type      = ARG_PTR_TO_CTX,
4156         .arg2_type      = ARG_ANYTHING,
4157 };
4158
4159 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4160 {
4161         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4162         void *meta = xdp->data_meta + offset;
4163         unsigned long metalen = xdp->data - meta;
4164
4165         if (xdp_data_meta_unsupported(xdp))
4166                 return -ENOTSUPP;
4167         if (unlikely(meta < xdp_frame_end ||
4168                      meta > xdp->data))
4169                 return -EINVAL;
4170         if (unlikely(xdp_metalen_invalid(metalen)))
4171                 return -EACCES;
4172
4173         xdp->data_meta = meta;
4174
4175         return 0;
4176 }
4177
4178 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4179         .func           = bpf_xdp_adjust_meta,
4180         .gpl_only       = false,
4181         .ret_type       = RET_INTEGER,
4182         .arg1_type      = ARG_PTR_TO_CTX,
4183         .arg2_type      = ARG_ANYTHING,
4184 };
4185
4186 /**
4187  * DOC: xdp redirect
4188  *
4189  * XDP_REDIRECT works by a three-step process, implemented in the functions
4190  * below:
4191  *
4192  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4193  *    of the redirect and store it (along with some other metadata) in a per-CPU
4194  *    struct bpf_redirect_info.
4195  *
4196  * 2. When the program returns the XDP_REDIRECT return code, the driver will
4197  *    call xdp_do_redirect() which will use the information in struct
4198  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
4199  *    bulk queue structure.
4200  *
4201  * 3. Before exiting its NAPI poll loop, the driver will call
4202  *    xdp_do_flush(), which will flush all the different bulk queues,
4203  *    thus completing the redirect. Note that xdp_do_flush() must be
4204  *    called before napi_complete_done() in the driver, as the
4205  *    XDP_REDIRECT logic relies on being inside a single NAPI instance
4206  *    through to the xdp_do_flush() call for RCU protection of all
4207  *    in-kernel data structures.
4208  */
4209 /*
4210  * Pointers to the map entries will be kept around for this whole sequence of
4211  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4212  * the core code; instead, the RCU protection relies on everything happening
4213  * inside a single NAPI poll sequence, which means it's between a pair of calls
4214  * to local_bh_disable()/local_bh_enable().
4215  *
4216  * The map entries are marked as __rcu and the map code makes sure to
4217  * dereference those pointers with rcu_dereference_check() in a way that works
4218  * for both sections that to hold an rcu_read_lock() and sections that are
4219  * called from NAPI without a separate rcu_read_lock(). The code below does not
4220  * use RCU annotations, but relies on those in the map code.
4221  */
4222 void xdp_do_flush(void)
4223 {
4224         __dev_flush();
4225         __cpu_map_flush();
4226         __xsk_map_flush();
4227 }
4228 EXPORT_SYMBOL_GPL(xdp_do_flush);
4229
4230 void bpf_clear_redirect_map(struct bpf_map *map)
4231 {
4232         struct bpf_redirect_info *ri;
4233         int cpu;
4234
4235         for_each_possible_cpu(cpu) {
4236                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4237                 /* Avoid polluting remote cacheline due to writes if
4238                  * not needed. Once we pass this test, we need the
4239                  * cmpxchg() to make sure it hasn't been changed in
4240                  * the meantime by remote CPU.
4241                  */
4242                 if (unlikely(READ_ONCE(ri->map) == map))
4243                         cmpxchg(&ri->map, map, NULL);
4244         }
4245 }
4246
4247 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4248 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4249
4250 u32 xdp_master_redirect(struct xdp_buff *xdp)
4251 {
4252         struct net_device *master, *slave;
4253         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4254
4255         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4256         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4257         if (slave && slave != xdp->rxq->dev) {
4258                 /* The target device is different from the receiving device, so
4259                  * redirect it to the new device.
4260                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4261                  * drivers to unmap the packet from their rx ring.
4262                  */
4263                 ri->tgt_index = slave->ifindex;
4264                 ri->map_id = INT_MAX;
4265                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4266                 return XDP_REDIRECT;
4267         }
4268         return XDP_TX;
4269 }
4270 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4271
4272 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4273                                         struct net_device *dev,
4274                                         struct xdp_buff *xdp,
4275                                         struct bpf_prog *xdp_prog)
4276 {
4277         enum bpf_map_type map_type = ri->map_type;
4278         void *fwd = ri->tgt_value;
4279         u32 map_id = ri->map_id;
4280         int err;
4281
4282         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4283         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4284
4285         err = __xsk_map_redirect(fwd, xdp);
4286         if (unlikely(err))
4287                 goto err;
4288
4289         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4290         return 0;
4291 err:
4292         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4293         return err;
4294 }
4295
4296 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4297                                                    struct net_device *dev,
4298                                                    struct xdp_frame *xdpf,
4299                                                    struct bpf_prog *xdp_prog)
4300 {
4301         enum bpf_map_type map_type = ri->map_type;
4302         void *fwd = ri->tgt_value;
4303         u32 map_id = ri->map_id;
4304         struct bpf_map *map;
4305         int err;
4306
4307         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4308         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4309
4310         if (unlikely(!xdpf)) {
4311                 err = -EOVERFLOW;
4312                 goto err;
4313         }
4314
4315         switch (map_type) {
4316         case BPF_MAP_TYPE_DEVMAP:
4317                 fallthrough;
4318         case BPF_MAP_TYPE_DEVMAP_HASH:
4319                 map = READ_ONCE(ri->map);
4320                 if (unlikely(map)) {
4321                         WRITE_ONCE(ri->map, NULL);
4322                         err = dev_map_enqueue_multi(xdpf, dev, map,
4323                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
4324                 } else {
4325                         err = dev_map_enqueue(fwd, xdpf, dev);
4326                 }
4327                 break;
4328         case BPF_MAP_TYPE_CPUMAP:
4329                 err = cpu_map_enqueue(fwd, xdpf, dev);
4330                 break;
4331         case BPF_MAP_TYPE_UNSPEC:
4332                 if (map_id == INT_MAX) {
4333                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4334                         if (unlikely(!fwd)) {
4335                                 err = -EINVAL;
4336                                 break;
4337                         }
4338                         err = dev_xdp_enqueue(fwd, xdpf, dev);
4339                         break;
4340                 }
4341                 fallthrough;
4342         default:
4343                 err = -EBADRQC;
4344         }
4345
4346         if (unlikely(err))
4347                 goto err;
4348
4349         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4350         return 0;
4351 err:
4352         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4353         return err;
4354 }
4355
4356 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4357                     struct bpf_prog *xdp_prog)
4358 {
4359         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4360         enum bpf_map_type map_type = ri->map_type;
4361
4362         if (map_type == BPF_MAP_TYPE_XSKMAP)
4363                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4364
4365         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4366                                        xdp_prog);
4367 }
4368 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4369
4370 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4371                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4372 {
4373         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4374         enum bpf_map_type map_type = ri->map_type;
4375
4376         if (map_type == BPF_MAP_TYPE_XSKMAP)
4377                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4378
4379         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4380 }
4381 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4382
4383 static int xdp_do_generic_redirect_map(struct net_device *dev,
4384                                        struct sk_buff *skb,
4385                                        struct xdp_buff *xdp,
4386                                        struct bpf_prog *xdp_prog,
4387                                        void *fwd,
4388                                        enum bpf_map_type map_type, u32 map_id)
4389 {
4390         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4391         struct bpf_map *map;
4392         int err;
4393
4394         switch (map_type) {
4395         case BPF_MAP_TYPE_DEVMAP:
4396                 fallthrough;
4397         case BPF_MAP_TYPE_DEVMAP_HASH:
4398                 map = READ_ONCE(ri->map);
4399                 if (unlikely(map)) {
4400                         WRITE_ONCE(ri->map, NULL);
4401                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4402                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4403                 } else {
4404                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4405                 }
4406                 if (unlikely(err))
4407                         goto err;
4408                 break;
4409         case BPF_MAP_TYPE_XSKMAP:
4410                 err = xsk_generic_rcv(fwd, xdp);
4411                 if (err)
4412                         goto err;
4413                 consume_skb(skb);
4414                 break;
4415         case BPF_MAP_TYPE_CPUMAP:
4416                 err = cpu_map_generic_redirect(fwd, skb);
4417                 if (unlikely(err))
4418                         goto err;
4419                 break;
4420         default:
4421                 err = -EBADRQC;
4422                 goto err;
4423         }
4424
4425         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4426         return 0;
4427 err:
4428         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4429         return err;
4430 }
4431
4432 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4433                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4434 {
4435         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4436         enum bpf_map_type map_type = ri->map_type;
4437         void *fwd = ri->tgt_value;
4438         u32 map_id = ri->map_id;
4439         int err;
4440
4441         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4442         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4443
4444         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4445                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4446                 if (unlikely(!fwd)) {
4447                         err = -EINVAL;
4448                         goto err;
4449                 }
4450
4451                 err = xdp_ok_fwd_dev(fwd, skb->len);
4452                 if (unlikely(err))
4453                         goto err;
4454
4455                 skb->dev = fwd;
4456                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4457                 generic_xdp_tx(skb, xdp_prog);
4458                 return 0;
4459         }
4460
4461         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4462 err:
4463         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4464         return err;
4465 }
4466
4467 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4468 {
4469         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4470
4471         if (unlikely(flags))
4472                 return XDP_ABORTED;
4473
4474         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4475          * by map_idr) is used for ifindex based XDP redirect.
4476          */
4477         ri->tgt_index = ifindex;
4478         ri->map_id = INT_MAX;
4479         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4480
4481         return XDP_REDIRECT;
4482 }
4483
4484 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4485         .func           = bpf_xdp_redirect,
4486         .gpl_only       = false,
4487         .ret_type       = RET_INTEGER,
4488         .arg1_type      = ARG_ANYTHING,
4489         .arg2_type      = ARG_ANYTHING,
4490 };
4491
4492 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4493            u64, flags)
4494 {
4495         return map->ops->map_redirect(map, key, flags);
4496 }
4497
4498 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4499         .func           = bpf_xdp_redirect_map,
4500         .gpl_only       = false,
4501         .ret_type       = RET_INTEGER,
4502         .arg1_type      = ARG_CONST_MAP_PTR,
4503         .arg2_type      = ARG_ANYTHING,
4504         .arg3_type      = ARG_ANYTHING,
4505 };
4506
4507 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4508                                   unsigned long off, unsigned long len)
4509 {
4510         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4511
4512         if (unlikely(!ptr))
4513                 return len;
4514         if (ptr != dst_buff)
4515                 memcpy(dst_buff, ptr, len);
4516
4517         return 0;
4518 }
4519
4520 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4521            u64, flags, void *, meta, u64, meta_size)
4522 {
4523         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4524
4525         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4526                 return -EINVAL;
4527         if (unlikely(!skb || skb_size > skb->len))
4528                 return -EFAULT;
4529
4530         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4531                                 bpf_skb_copy);
4532 }
4533
4534 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4535         .func           = bpf_skb_event_output,
4536         .gpl_only       = true,
4537         .ret_type       = RET_INTEGER,
4538         .arg1_type      = ARG_PTR_TO_CTX,
4539         .arg2_type      = ARG_CONST_MAP_PTR,
4540         .arg3_type      = ARG_ANYTHING,
4541         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4542         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4543 };
4544
4545 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4546
4547 const struct bpf_func_proto bpf_skb_output_proto = {
4548         .func           = bpf_skb_event_output,
4549         .gpl_only       = true,
4550         .ret_type       = RET_INTEGER,
4551         .arg1_type      = ARG_PTR_TO_BTF_ID,
4552         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4553         .arg2_type      = ARG_CONST_MAP_PTR,
4554         .arg3_type      = ARG_ANYTHING,
4555         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4556         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4557 };
4558
4559 static unsigned short bpf_tunnel_key_af(u64 flags)
4560 {
4561         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4562 }
4563
4564 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4565            u32, size, u64, flags)
4566 {
4567         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4568         u8 compat[sizeof(struct bpf_tunnel_key)];
4569         void *to_orig = to;
4570         int err;
4571
4572         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4573                                          BPF_F_TUNINFO_FLAGS)))) {
4574                 err = -EINVAL;
4575                 goto err_clear;
4576         }
4577         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4578                 err = -EPROTO;
4579                 goto err_clear;
4580         }
4581         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4582                 err = -EINVAL;
4583                 switch (size) {
4584                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4585                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4586                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4587                         goto set_compat;
4588                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4589                         /* Fixup deprecated structure layouts here, so we have
4590                          * a common path later on.
4591                          */
4592                         if (ip_tunnel_info_af(info) != AF_INET)
4593                                 goto err_clear;
4594 set_compat:
4595                         to = (struct bpf_tunnel_key *)compat;
4596                         break;
4597                 default:
4598                         goto err_clear;
4599                 }
4600         }
4601
4602         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4603         to->tunnel_tos = info->key.tos;
4604         to->tunnel_ttl = info->key.ttl;
4605         if (flags & BPF_F_TUNINFO_FLAGS)
4606                 to->tunnel_flags = info->key.tun_flags;
4607         else
4608                 to->tunnel_ext = 0;
4609
4610         if (flags & BPF_F_TUNINFO_IPV6) {
4611                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4612                        sizeof(to->remote_ipv6));
4613                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4614                        sizeof(to->local_ipv6));
4615                 to->tunnel_label = be32_to_cpu(info->key.label);
4616         } else {
4617                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4618                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4619                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4620                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4621                 to->tunnel_label = 0;
4622         }
4623
4624         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4625                 memcpy(to_orig, to, size);
4626
4627         return 0;
4628 err_clear:
4629         memset(to_orig, 0, size);
4630         return err;
4631 }
4632
4633 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4634         .func           = bpf_skb_get_tunnel_key,
4635         .gpl_only       = false,
4636         .ret_type       = RET_INTEGER,
4637         .arg1_type      = ARG_PTR_TO_CTX,
4638         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4639         .arg3_type      = ARG_CONST_SIZE,
4640         .arg4_type      = ARG_ANYTHING,
4641 };
4642
4643 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4644 {
4645         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4646         int err;
4647
4648         if (unlikely(!info ||
4649                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4650                 err = -ENOENT;
4651                 goto err_clear;
4652         }
4653         if (unlikely(size < info->options_len)) {
4654                 err = -ENOMEM;
4655                 goto err_clear;
4656         }
4657
4658         ip_tunnel_info_opts_get(to, info);
4659         if (size > info->options_len)
4660                 memset(to + info->options_len, 0, size - info->options_len);
4661
4662         return info->options_len;
4663 err_clear:
4664         memset(to, 0, size);
4665         return err;
4666 }
4667
4668 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4669         .func           = bpf_skb_get_tunnel_opt,
4670         .gpl_only       = false,
4671         .ret_type       = RET_INTEGER,
4672         .arg1_type      = ARG_PTR_TO_CTX,
4673         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4674         .arg3_type      = ARG_CONST_SIZE,
4675 };
4676
4677 static struct metadata_dst __percpu *md_dst;
4678
4679 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4680            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4681 {
4682         struct metadata_dst *md = this_cpu_ptr(md_dst);
4683         u8 compat[sizeof(struct bpf_tunnel_key)];
4684         struct ip_tunnel_info *info;
4685
4686         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4687                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4688                                BPF_F_NO_TUNNEL_KEY)))
4689                 return -EINVAL;
4690         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4691                 switch (size) {
4692                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4693                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4694                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4695                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4696                         /* Fixup deprecated structure layouts here, so we have
4697                          * a common path later on.
4698                          */
4699                         memcpy(compat, from, size);
4700                         memset(compat + size, 0, sizeof(compat) - size);
4701                         from = (const struct bpf_tunnel_key *) compat;
4702                         break;
4703                 default:
4704                         return -EINVAL;
4705                 }
4706         }
4707         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4708                      from->tunnel_ext))
4709                 return -EINVAL;
4710
4711         skb_dst_drop(skb);
4712         dst_hold((struct dst_entry *) md);
4713         skb_dst_set(skb, (struct dst_entry *) md);
4714
4715         info = &md->u.tun_info;
4716         memset(info, 0, sizeof(*info));
4717         info->mode = IP_TUNNEL_INFO_TX;
4718
4719         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4720         if (flags & BPF_F_DONT_FRAGMENT)
4721                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4722         if (flags & BPF_F_ZERO_CSUM_TX)
4723                 info->key.tun_flags &= ~TUNNEL_CSUM;
4724         if (flags & BPF_F_SEQ_NUMBER)
4725                 info->key.tun_flags |= TUNNEL_SEQ;
4726         if (flags & BPF_F_NO_TUNNEL_KEY)
4727                 info->key.tun_flags &= ~TUNNEL_KEY;
4728
4729         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4730         info->key.tos = from->tunnel_tos;
4731         info->key.ttl = from->tunnel_ttl;
4732
4733         if (flags & BPF_F_TUNINFO_IPV6) {
4734                 info->mode |= IP_TUNNEL_INFO_IPV6;
4735                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4736                        sizeof(from->remote_ipv6));
4737                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4738                        sizeof(from->local_ipv6));
4739                 info->key.label = cpu_to_be32(from->tunnel_label) &
4740                                   IPV6_FLOWLABEL_MASK;
4741         } else {
4742                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4743                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4744                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4745         }
4746
4747         return 0;
4748 }
4749
4750 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4751         .func           = bpf_skb_set_tunnel_key,
4752         .gpl_only       = false,
4753         .ret_type       = RET_INTEGER,
4754         .arg1_type      = ARG_PTR_TO_CTX,
4755         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4756         .arg3_type      = ARG_CONST_SIZE,
4757         .arg4_type      = ARG_ANYTHING,
4758 };
4759
4760 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4761            const u8 *, from, u32, size)
4762 {
4763         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4764         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4765
4766         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4767                 return -EINVAL;
4768         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4769                 return -ENOMEM;
4770
4771         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4772
4773         return 0;
4774 }
4775
4776 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4777         .func           = bpf_skb_set_tunnel_opt,
4778         .gpl_only       = false,
4779         .ret_type       = RET_INTEGER,
4780         .arg1_type      = ARG_PTR_TO_CTX,
4781         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4782         .arg3_type      = ARG_CONST_SIZE,
4783 };
4784
4785 static const struct bpf_func_proto *
4786 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4787 {
4788         if (!md_dst) {
4789                 struct metadata_dst __percpu *tmp;
4790
4791                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4792                                                 METADATA_IP_TUNNEL,
4793                                                 GFP_KERNEL);
4794                 if (!tmp)
4795                         return NULL;
4796                 if (cmpxchg(&md_dst, NULL, tmp))
4797                         metadata_dst_free_percpu(tmp);
4798         }
4799
4800         switch (which) {
4801         case BPF_FUNC_skb_set_tunnel_key:
4802                 return &bpf_skb_set_tunnel_key_proto;
4803         case BPF_FUNC_skb_set_tunnel_opt:
4804                 return &bpf_skb_set_tunnel_opt_proto;
4805         default:
4806                 return NULL;
4807         }
4808 }
4809
4810 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4811            u32, idx)
4812 {
4813         struct bpf_array *array = container_of(map, struct bpf_array, map);
4814         struct cgroup *cgrp;
4815         struct sock *sk;
4816
4817         sk = skb_to_full_sk(skb);
4818         if (!sk || !sk_fullsock(sk))
4819                 return -ENOENT;
4820         if (unlikely(idx >= array->map.max_entries))
4821                 return -E2BIG;
4822
4823         cgrp = READ_ONCE(array->ptrs[idx]);
4824         if (unlikely(!cgrp))
4825                 return -EAGAIN;
4826
4827         return sk_under_cgroup_hierarchy(sk, cgrp);
4828 }
4829
4830 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4831         .func           = bpf_skb_under_cgroup,
4832         .gpl_only       = false,
4833         .ret_type       = RET_INTEGER,
4834         .arg1_type      = ARG_PTR_TO_CTX,
4835         .arg2_type      = ARG_CONST_MAP_PTR,
4836         .arg3_type      = ARG_ANYTHING,
4837 };
4838
4839 #ifdef CONFIG_SOCK_CGROUP_DATA
4840 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4841 {
4842         struct cgroup *cgrp;
4843
4844         sk = sk_to_full_sk(sk);
4845         if (!sk || !sk_fullsock(sk))
4846                 return 0;
4847
4848         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4849         return cgroup_id(cgrp);
4850 }
4851
4852 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4853 {
4854         return __bpf_sk_cgroup_id(skb->sk);
4855 }
4856
4857 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4858         .func           = bpf_skb_cgroup_id,
4859         .gpl_only       = false,
4860         .ret_type       = RET_INTEGER,
4861         .arg1_type      = ARG_PTR_TO_CTX,
4862 };
4863
4864 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4865                                               int ancestor_level)
4866 {
4867         struct cgroup *ancestor;
4868         struct cgroup *cgrp;
4869
4870         sk = sk_to_full_sk(sk);
4871         if (!sk || !sk_fullsock(sk))
4872                 return 0;
4873
4874         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4875         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4876         if (!ancestor)
4877                 return 0;
4878
4879         return cgroup_id(ancestor);
4880 }
4881
4882 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4883            ancestor_level)
4884 {
4885         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4886 }
4887
4888 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4889         .func           = bpf_skb_ancestor_cgroup_id,
4890         .gpl_only       = false,
4891         .ret_type       = RET_INTEGER,
4892         .arg1_type      = ARG_PTR_TO_CTX,
4893         .arg2_type      = ARG_ANYTHING,
4894 };
4895
4896 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4897 {
4898         return __bpf_sk_cgroup_id(sk);
4899 }
4900
4901 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4902         .func           = bpf_sk_cgroup_id,
4903         .gpl_only       = false,
4904         .ret_type       = RET_INTEGER,
4905         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4906 };
4907
4908 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4909 {
4910         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4911 }
4912
4913 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4914         .func           = bpf_sk_ancestor_cgroup_id,
4915         .gpl_only       = false,
4916         .ret_type       = RET_INTEGER,
4917         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4918         .arg2_type      = ARG_ANYTHING,
4919 };
4920 #endif
4921
4922 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4923                                   unsigned long off, unsigned long len)
4924 {
4925         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4926
4927         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4928         return 0;
4929 }
4930
4931 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4932            u64, flags, void *, meta, u64, meta_size)
4933 {
4934         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4935
4936         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4937                 return -EINVAL;
4938
4939         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4940                 return -EFAULT;
4941
4942         return bpf_event_output(map, flags, meta, meta_size, xdp,
4943                                 xdp_size, bpf_xdp_copy);
4944 }
4945
4946 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4947         .func           = bpf_xdp_event_output,
4948         .gpl_only       = true,
4949         .ret_type       = RET_INTEGER,
4950         .arg1_type      = ARG_PTR_TO_CTX,
4951         .arg2_type      = ARG_CONST_MAP_PTR,
4952         .arg3_type      = ARG_ANYTHING,
4953         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4954         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4955 };
4956
4957 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4958
4959 const struct bpf_func_proto bpf_xdp_output_proto = {
4960         .func           = bpf_xdp_event_output,
4961         .gpl_only       = true,
4962         .ret_type       = RET_INTEGER,
4963         .arg1_type      = ARG_PTR_TO_BTF_ID,
4964         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4965         .arg2_type      = ARG_CONST_MAP_PTR,
4966         .arg3_type      = ARG_ANYTHING,
4967         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4968         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4969 };
4970
4971 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4972 {
4973         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4974 }
4975
4976 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4977         .func           = bpf_get_socket_cookie,
4978         .gpl_only       = false,
4979         .ret_type       = RET_INTEGER,
4980         .arg1_type      = ARG_PTR_TO_CTX,
4981 };
4982
4983 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4984 {
4985         return __sock_gen_cookie(ctx->sk);
4986 }
4987
4988 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4989         .func           = bpf_get_socket_cookie_sock_addr,
4990         .gpl_only       = false,
4991         .ret_type       = RET_INTEGER,
4992         .arg1_type      = ARG_PTR_TO_CTX,
4993 };
4994
4995 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4996 {
4997         return __sock_gen_cookie(ctx);
4998 }
4999
5000 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
5001         .func           = bpf_get_socket_cookie_sock,
5002         .gpl_only       = false,
5003         .ret_type       = RET_INTEGER,
5004         .arg1_type      = ARG_PTR_TO_CTX,
5005 };
5006
5007 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
5008 {
5009         return sk ? sock_gen_cookie(sk) : 0;
5010 }
5011
5012 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5013         .func           = bpf_get_socket_ptr_cookie,
5014         .gpl_only       = false,
5015         .ret_type       = RET_INTEGER,
5016         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5017 };
5018
5019 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5020 {
5021         return __sock_gen_cookie(ctx->sk);
5022 }
5023
5024 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5025         .func           = bpf_get_socket_cookie_sock_ops,
5026         .gpl_only       = false,
5027         .ret_type       = RET_INTEGER,
5028         .arg1_type      = ARG_PTR_TO_CTX,
5029 };
5030
5031 static u64 __bpf_get_netns_cookie(struct sock *sk)
5032 {
5033         const struct net *net = sk ? sock_net(sk) : &init_net;
5034
5035         return net->net_cookie;
5036 }
5037
5038 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5039 {
5040         return __bpf_get_netns_cookie(ctx);
5041 }
5042
5043 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5044         .func           = bpf_get_netns_cookie_sock,
5045         .gpl_only       = false,
5046         .ret_type       = RET_INTEGER,
5047         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5048 };
5049
5050 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5051 {
5052         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5053 }
5054
5055 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5056         .func           = bpf_get_netns_cookie_sock_addr,
5057         .gpl_only       = false,
5058         .ret_type       = RET_INTEGER,
5059         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5060 };
5061
5062 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5063 {
5064         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5065 }
5066
5067 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5068         .func           = bpf_get_netns_cookie_sock_ops,
5069         .gpl_only       = false,
5070         .ret_type       = RET_INTEGER,
5071         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5072 };
5073
5074 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5075 {
5076         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5077 }
5078
5079 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5080         .func           = bpf_get_netns_cookie_sk_msg,
5081         .gpl_only       = false,
5082         .ret_type       = RET_INTEGER,
5083         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5084 };
5085
5086 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5087 {
5088         struct sock *sk = sk_to_full_sk(skb->sk);
5089         kuid_t kuid;
5090
5091         if (!sk || !sk_fullsock(sk))
5092                 return overflowuid;
5093         kuid = sock_net_uid(sock_net(sk), sk);
5094         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5095 }
5096
5097 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5098         .func           = bpf_get_socket_uid,
5099         .gpl_only       = false,
5100         .ret_type       = RET_INTEGER,
5101         .arg1_type      = ARG_PTR_TO_CTX,
5102 };
5103
5104 static int sol_socket_sockopt(struct sock *sk, int optname,
5105                               char *optval, int *optlen,
5106                               bool getopt)
5107 {
5108         switch (optname) {
5109         case SO_REUSEADDR:
5110         case SO_SNDBUF:
5111         case SO_RCVBUF:
5112         case SO_KEEPALIVE:
5113         case SO_PRIORITY:
5114         case SO_REUSEPORT:
5115         case SO_RCVLOWAT:
5116         case SO_MARK:
5117         case SO_MAX_PACING_RATE:
5118         case SO_BINDTOIFINDEX:
5119         case SO_TXREHASH:
5120                 if (*optlen != sizeof(int))
5121                         return -EINVAL;
5122                 break;
5123         case SO_BINDTODEVICE:
5124                 break;
5125         default:
5126                 return -EINVAL;
5127         }
5128
5129         if (getopt) {
5130                 if (optname == SO_BINDTODEVICE)
5131                         return -EINVAL;
5132                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5133                                      KERNEL_SOCKPTR(optval),
5134                                      KERNEL_SOCKPTR(optlen));
5135         }
5136
5137         return sk_setsockopt(sk, SOL_SOCKET, optname,
5138                              KERNEL_SOCKPTR(optval), *optlen);
5139 }
5140
5141 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5142                                   char *optval, int optlen)
5143 {
5144         struct tcp_sock *tp = tcp_sk(sk);
5145         unsigned long timeout;
5146         int val;
5147
5148         if (optlen != sizeof(int))
5149                 return -EINVAL;
5150
5151         val = *(int *)optval;
5152
5153         /* Only some options are supported */
5154         switch (optname) {
5155         case TCP_BPF_IW:
5156                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5157                         return -EINVAL;
5158                 tcp_snd_cwnd_set(tp, val);
5159                 break;
5160         case TCP_BPF_SNDCWND_CLAMP:
5161                 if (val <= 0)
5162                         return -EINVAL;
5163                 tp->snd_cwnd_clamp = val;
5164                 tp->snd_ssthresh = val;
5165                 break;
5166         case TCP_BPF_DELACK_MAX:
5167                 timeout = usecs_to_jiffies(val);
5168                 if (timeout > TCP_DELACK_MAX ||
5169                     timeout < TCP_TIMEOUT_MIN)
5170                         return -EINVAL;
5171                 inet_csk(sk)->icsk_delack_max = timeout;
5172                 break;
5173         case TCP_BPF_RTO_MIN:
5174                 timeout = usecs_to_jiffies(val);
5175                 if (timeout > TCP_RTO_MIN ||
5176                     timeout < TCP_TIMEOUT_MIN)
5177                         return -EINVAL;
5178                 inet_csk(sk)->icsk_rto_min = timeout;
5179                 break;
5180         default:
5181                 return -EINVAL;
5182         }
5183
5184         return 0;
5185 }
5186
5187 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5188                                       int *optlen, bool getopt)
5189 {
5190         struct tcp_sock *tp;
5191         int ret;
5192
5193         if (*optlen < 2)
5194                 return -EINVAL;
5195
5196         if (getopt) {
5197                 if (!inet_csk(sk)->icsk_ca_ops)
5198                         return -EINVAL;
5199                 /* BPF expects NULL-terminated tcp-cc string */
5200                 optval[--(*optlen)] = '\0';
5201                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5202                                          KERNEL_SOCKPTR(optval),
5203                                          KERNEL_SOCKPTR(optlen));
5204         }
5205
5206         /* "cdg" is the only cc that alloc a ptr
5207          * in inet_csk_ca area.  The bpf-tcp-cc may
5208          * overwrite this ptr after switching to cdg.
5209          */
5210         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5211                 return -ENOTSUPP;
5212
5213         /* It stops this looping
5214          *
5215          * .init => bpf_setsockopt(tcp_cc) => .init =>
5216          * bpf_setsockopt(tcp_cc)" => .init => ....
5217          *
5218          * The second bpf_setsockopt(tcp_cc) is not allowed
5219          * in order to break the loop when both .init
5220          * are the same bpf prog.
5221          *
5222          * This applies even the second bpf_setsockopt(tcp_cc)
5223          * does not cause a loop.  This limits only the first
5224          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5225          * pick a fallback cc (eg. peer does not support ECN)
5226          * and the second '.init' cannot fallback to
5227          * another.
5228          */
5229         tp = tcp_sk(sk);
5230         if (tp->bpf_chg_cc_inprogress)
5231                 return -EBUSY;
5232
5233         tp->bpf_chg_cc_inprogress = 1;
5234         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5235                                 KERNEL_SOCKPTR(optval), *optlen);
5236         tp->bpf_chg_cc_inprogress = 0;
5237         return ret;
5238 }
5239
5240 static int sol_tcp_sockopt(struct sock *sk, int optname,
5241                            char *optval, int *optlen,
5242                            bool getopt)
5243 {
5244         if (sk->sk_protocol != IPPROTO_TCP)
5245                 return -EINVAL;
5246
5247         switch (optname) {
5248         case TCP_NODELAY:
5249         case TCP_MAXSEG:
5250         case TCP_KEEPIDLE:
5251         case TCP_KEEPINTVL:
5252         case TCP_KEEPCNT:
5253         case TCP_SYNCNT:
5254         case TCP_WINDOW_CLAMP:
5255         case TCP_THIN_LINEAR_TIMEOUTS:
5256         case TCP_USER_TIMEOUT:
5257         case TCP_NOTSENT_LOWAT:
5258         case TCP_SAVE_SYN:
5259                 if (*optlen != sizeof(int))
5260                         return -EINVAL;
5261                 break;
5262         case TCP_CONGESTION:
5263                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5264         case TCP_SAVED_SYN:
5265                 if (*optlen < 1)
5266                         return -EINVAL;
5267                 break;
5268         default:
5269                 if (getopt)
5270                         return -EINVAL;
5271                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5272         }
5273
5274         if (getopt) {
5275                 if (optname == TCP_SAVED_SYN) {
5276                         struct tcp_sock *tp = tcp_sk(sk);
5277
5278                         if (!tp->saved_syn ||
5279                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5280                                 return -EINVAL;
5281                         memcpy(optval, tp->saved_syn->data, *optlen);
5282                         /* It cannot free tp->saved_syn here because it
5283                          * does not know if the user space still needs it.
5284                          */
5285                         return 0;
5286                 }
5287
5288                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5289                                          KERNEL_SOCKPTR(optval),
5290                                          KERNEL_SOCKPTR(optlen));
5291         }
5292
5293         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5294                                  KERNEL_SOCKPTR(optval), *optlen);
5295 }
5296
5297 static int sol_ip_sockopt(struct sock *sk, int optname,
5298                           char *optval, int *optlen,
5299                           bool getopt)
5300 {
5301         if (sk->sk_family != AF_INET)
5302                 return -EINVAL;
5303
5304         switch (optname) {
5305         case IP_TOS:
5306                 if (*optlen != sizeof(int))
5307                         return -EINVAL;
5308                 break;
5309         default:
5310                 return -EINVAL;
5311         }
5312
5313         if (getopt)
5314                 return do_ip_getsockopt(sk, SOL_IP, optname,
5315                                         KERNEL_SOCKPTR(optval),
5316                                         KERNEL_SOCKPTR(optlen));
5317
5318         return do_ip_setsockopt(sk, SOL_IP, optname,
5319                                 KERNEL_SOCKPTR(optval), *optlen);
5320 }
5321
5322 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5323                             char *optval, int *optlen,
5324                             bool getopt)
5325 {
5326         if (sk->sk_family != AF_INET6)
5327                 return -EINVAL;
5328
5329         switch (optname) {
5330         case IPV6_TCLASS:
5331         case IPV6_AUTOFLOWLABEL:
5332                 if (*optlen != sizeof(int))
5333                         return -EINVAL;
5334                 break;
5335         default:
5336                 return -EINVAL;
5337         }
5338
5339         if (getopt)
5340                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5341                                                       KERNEL_SOCKPTR(optval),
5342                                                       KERNEL_SOCKPTR(optlen));
5343
5344         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5345                                               KERNEL_SOCKPTR(optval), *optlen);
5346 }
5347
5348 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5349                             char *optval, int optlen)
5350 {
5351         if (!sk_fullsock(sk))
5352                 return -EINVAL;
5353
5354         if (level == SOL_SOCKET)
5355                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5356         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5357                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5358         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5359                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5360         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5361                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5362
5363         return -EINVAL;
5364 }
5365
5366 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5367                            char *optval, int optlen)
5368 {
5369         if (sk_fullsock(sk))
5370                 sock_owned_by_me(sk);
5371         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5372 }
5373
5374 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5375                             char *optval, int optlen)
5376 {
5377         int err, saved_optlen = optlen;
5378
5379         if (!sk_fullsock(sk)) {
5380                 err = -EINVAL;
5381                 goto done;
5382         }
5383
5384         if (level == SOL_SOCKET)
5385                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5386         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5387                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5388         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5389                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5390         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5391                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5392         else
5393                 err = -EINVAL;
5394
5395 done:
5396         if (err)
5397                 optlen = 0;
5398         if (optlen < saved_optlen)
5399                 memset(optval + optlen, 0, saved_optlen - optlen);
5400         return err;
5401 }
5402
5403 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5404                            char *optval, int optlen)
5405 {
5406         if (sk_fullsock(sk))
5407                 sock_owned_by_me(sk);
5408         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5409 }
5410
5411 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5412            int, optname, char *, optval, int, optlen)
5413 {
5414         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5415 }
5416
5417 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5418         .func           = bpf_sk_setsockopt,
5419         .gpl_only       = false,
5420         .ret_type       = RET_INTEGER,
5421         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5422         .arg2_type      = ARG_ANYTHING,
5423         .arg3_type      = ARG_ANYTHING,
5424         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5425         .arg5_type      = ARG_CONST_SIZE,
5426 };
5427
5428 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5429            int, optname, char *, optval, int, optlen)
5430 {
5431         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5432 }
5433
5434 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5435         .func           = bpf_sk_getsockopt,
5436         .gpl_only       = false,
5437         .ret_type       = RET_INTEGER,
5438         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5439         .arg2_type      = ARG_ANYTHING,
5440         .arg3_type      = ARG_ANYTHING,
5441         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5442         .arg5_type      = ARG_CONST_SIZE,
5443 };
5444
5445 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5446            int, optname, char *, optval, int, optlen)
5447 {
5448         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5449 }
5450
5451 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5452         .func           = bpf_unlocked_sk_setsockopt,
5453         .gpl_only       = false,
5454         .ret_type       = RET_INTEGER,
5455         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5456         .arg2_type      = ARG_ANYTHING,
5457         .arg3_type      = ARG_ANYTHING,
5458         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5459         .arg5_type      = ARG_CONST_SIZE,
5460 };
5461
5462 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5463            int, optname, char *, optval, int, optlen)
5464 {
5465         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5466 }
5467
5468 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5469         .func           = bpf_unlocked_sk_getsockopt,
5470         .gpl_only       = false,
5471         .ret_type       = RET_INTEGER,
5472         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5473         .arg2_type      = ARG_ANYTHING,
5474         .arg3_type      = ARG_ANYTHING,
5475         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5476         .arg5_type      = ARG_CONST_SIZE,
5477 };
5478
5479 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5480            int, level, int, optname, char *, optval, int, optlen)
5481 {
5482         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5483 }
5484
5485 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5486         .func           = bpf_sock_addr_setsockopt,
5487         .gpl_only       = false,
5488         .ret_type       = RET_INTEGER,
5489         .arg1_type      = ARG_PTR_TO_CTX,
5490         .arg2_type      = ARG_ANYTHING,
5491         .arg3_type      = ARG_ANYTHING,
5492         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5493         .arg5_type      = ARG_CONST_SIZE,
5494 };
5495
5496 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5497            int, level, int, optname, char *, optval, int, optlen)
5498 {
5499         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5500 }
5501
5502 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5503         .func           = bpf_sock_addr_getsockopt,
5504         .gpl_only       = false,
5505         .ret_type       = RET_INTEGER,
5506         .arg1_type      = ARG_PTR_TO_CTX,
5507         .arg2_type      = ARG_ANYTHING,
5508         .arg3_type      = ARG_ANYTHING,
5509         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5510         .arg5_type      = ARG_CONST_SIZE,
5511 };
5512
5513 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5514            int, level, int, optname, char *, optval, int, optlen)
5515 {
5516         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5517 }
5518
5519 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5520         .func           = bpf_sock_ops_setsockopt,
5521         .gpl_only       = false,
5522         .ret_type       = RET_INTEGER,
5523         .arg1_type      = ARG_PTR_TO_CTX,
5524         .arg2_type      = ARG_ANYTHING,
5525         .arg3_type      = ARG_ANYTHING,
5526         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5527         .arg5_type      = ARG_CONST_SIZE,
5528 };
5529
5530 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5531                                 int optname, const u8 **start)
5532 {
5533         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5534         const u8 *hdr_start;
5535         int ret;
5536
5537         if (syn_skb) {
5538                 /* sk is a request_sock here */
5539
5540                 if (optname == TCP_BPF_SYN) {
5541                         hdr_start = syn_skb->data;
5542                         ret = tcp_hdrlen(syn_skb);
5543                 } else if (optname == TCP_BPF_SYN_IP) {
5544                         hdr_start = skb_network_header(syn_skb);
5545                         ret = skb_network_header_len(syn_skb) +
5546                                 tcp_hdrlen(syn_skb);
5547                 } else {
5548                         /* optname == TCP_BPF_SYN_MAC */
5549                         hdr_start = skb_mac_header(syn_skb);
5550                         ret = skb_mac_header_len(syn_skb) +
5551                                 skb_network_header_len(syn_skb) +
5552                                 tcp_hdrlen(syn_skb);
5553                 }
5554         } else {
5555                 struct sock *sk = bpf_sock->sk;
5556                 struct saved_syn *saved_syn;
5557
5558                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5559                         /* synack retransmit. bpf_sock->syn_skb will
5560                          * not be available.  It has to resort to
5561                          * saved_syn (if it is saved).
5562                          */
5563                         saved_syn = inet_reqsk(sk)->saved_syn;
5564                 else
5565                         saved_syn = tcp_sk(sk)->saved_syn;
5566
5567                 if (!saved_syn)
5568                         return -ENOENT;
5569
5570                 if (optname == TCP_BPF_SYN) {
5571                         hdr_start = saved_syn->data +
5572                                 saved_syn->mac_hdrlen +
5573                                 saved_syn->network_hdrlen;
5574                         ret = saved_syn->tcp_hdrlen;
5575                 } else if (optname == TCP_BPF_SYN_IP) {
5576                         hdr_start = saved_syn->data +
5577                                 saved_syn->mac_hdrlen;
5578                         ret = saved_syn->network_hdrlen +
5579                                 saved_syn->tcp_hdrlen;
5580                 } else {
5581                         /* optname == TCP_BPF_SYN_MAC */
5582
5583                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5584                         if (!saved_syn->mac_hdrlen)
5585                                 return -ENOENT;
5586
5587                         hdr_start = saved_syn->data;
5588                         ret = saved_syn->mac_hdrlen +
5589                                 saved_syn->network_hdrlen +
5590                                 saved_syn->tcp_hdrlen;
5591                 }
5592         }
5593
5594         *start = hdr_start;
5595         return ret;
5596 }
5597
5598 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5599            int, level, int, optname, char *, optval, int, optlen)
5600 {
5601         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5602             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5603                 int ret, copy_len = 0;
5604                 const u8 *start;
5605
5606                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5607                 if (ret > 0) {
5608                         copy_len = ret;
5609                         if (optlen < copy_len) {
5610                                 copy_len = optlen;
5611                                 ret = -ENOSPC;
5612                         }
5613
5614                         memcpy(optval, start, copy_len);
5615                 }
5616
5617                 /* Zero out unused buffer at the end */
5618                 memset(optval + copy_len, 0, optlen - copy_len);
5619
5620                 return ret;
5621         }
5622
5623         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5624 }
5625
5626 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5627         .func           = bpf_sock_ops_getsockopt,
5628         .gpl_only       = false,
5629         .ret_type       = RET_INTEGER,
5630         .arg1_type      = ARG_PTR_TO_CTX,
5631         .arg2_type      = ARG_ANYTHING,
5632         .arg3_type      = ARG_ANYTHING,
5633         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5634         .arg5_type      = ARG_CONST_SIZE,
5635 };
5636
5637 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5638            int, argval)
5639 {
5640         struct sock *sk = bpf_sock->sk;
5641         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5642
5643         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5644                 return -EINVAL;
5645
5646         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5647
5648         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5649 }
5650
5651 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5652         .func           = bpf_sock_ops_cb_flags_set,
5653         .gpl_only       = false,
5654         .ret_type       = RET_INTEGER,
5655         .arg1_type      = ARG_PTR_TO_CTX,
5656         .arg2_type      = ARG_ANYTHING,
5657 };
5658
5659 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5660 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5661
5662 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5663            int, addr_len)
5664 {
5665 #ifdef CONFIG_INET
5666         struct sock *sk = ctx->sk;
5667         u32 flags = BIND_FROM_BPF;
5668         int err;
5669
5670         err = -EINVAL;
5671         if (addr_len < offsetofend(struct sockaddr, sa_family))
5672                 return err;
5673         if (addr->sa_family == AF_INET) {
5674                 if (addr_len < sizeof(struct sockaddr_in))
5675                         return err;
5676                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5677                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5678                 return __inet_bind(sk, addr, addr_len, flags);
5679 #if IS_ENABLED(CONFIG_IPV6)
5680         } else if (addr->sa_family == AF_INET6) {
5681                 if (addr_len < SIN6_LEN_RFC2133)
5682                         return err;
5683                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5684                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5685                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5686                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5687                  */
5688                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5689 #endif /* CONFIG_IPV6 */
5690         }
5691 #endif /* CONFIG_INET */
5692
5693         return -EAFNOSUPPORT;
5694 }
5695
5696 static const struct bpf_func_proto bpf_bind_proto = {
5697         .func           = bpf_bind,
5698         .gpl_only       = false,
5699         .ret_type       = RET_INTEGER,
5700         .arg1_type      = ARG_PTR_TO_CTX,
5701         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5702         .arg3_type      = ARG_CONST_SIZE,
5703 };
5704
5705 #ifdef CONFIG_XFRM
5706
5707 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5708     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5709
5710 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5711 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5712
5713 #endif
5714
5715 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5716            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5717 {
5718         const struct sec_path *sp = skb_sec_path(skb);
5719         const struct xfrm_state *x;
5720
5721         if (!sp || unlikely(index >= sp->len || flags))
5722                 goto err_clear;
5723
5724         x = sp->xvec[index];
5725
5726         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5727                 goto err_clear;
5728
5729         to->reqid = x->props.reqid;
5730         to->spi = x->id.spi;
5731         to->family = x->props.family;
5732         to->ext = 0;
5733
5734         if (to->family == AF_INET6) {
5735                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5736                        sizeof(to->remote_ipv6));
5737         } else {
5738                 to->remote_ipv4 = x->props.saddr.a4;
5739                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5740         }
5741
5742         return 0;
5743 err_clear:
5744         memset(to, 0, size);
5745         return -EINVAL;
5746 }
5747
5748 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5749         .func           = bpf_skb_get_xfrm_state,
5750         .gpl_only       = false,
5751         .ret_type       = RET_INTEGER,
5752         .arg1_type      = ARG_PTR_TO_CTX,
5753         .arg2_type      = ARG_ANYTHING,
5754         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5755         .arg4_type      = ARG_CONST_SIZE,
5756         .arg5_type      = ARG_ANYTHING,
5757 };
5758 #endif
5759
5760 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5761 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5762 {
5763         params->h_vlan_TCI = 0;
5764         params->h_vlan_proto = 0;
5765         if (mtu)
5766                 params->mtu_result = mtu; /* union with tot_len */
5767
5768         return 0;
5769 }
5770 #endif
5771
5772 #if IS_ENABLED(CONFIG_INET)
5773 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5774                                u32 flags, bool check_mtu)
5775 {
5776         struct fib_nh_common *nhc;
5777         struct in_device *in_dev;
5778         struct neighbour *neigh;
5779         struct net_device *dev;
5780         struct fib_result res;
5781         struct flowi4 fl4;
5782         u32 mtu = 0;
5783         int err;
5784
5785         dev = dev_get_by_index_rcu(net, params->ifindex);
5786         if (unlikely(!dev))
5787                 return -ENODEV;
5788
5789         /* verify forwarding is enabled on this interface */
5790         in_dev = __in_dev_get_rcu(dev);
5791         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5792                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5793
5794         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5795                 fl4.flowi4_iif = 1;
5796                 fl4.flowi4_oif = params->ifindex;
5797         } else {
5798                 fl4.flowi4_iif = params->ifindex;
5799                 fl4.flowi4_oif = 0;
5800         }
5801         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5802         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5803         fl4.flowi4_flags = 0;
5804
5805         fl4.flowi4_proto = params->l4_protocol;
5806         fl4.daddr = params->ipv4_dst;
5807         fl4.saddr = params->ipv4_src;
5808         fl4.fl4_sport = params->sport;
5809         fl4.fl4_dport = params->dport;
5810         fl4.flowi4_multipath_hash = 0;
5811
5812         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5813                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5814                 struct fib_table *tb;
5815
5816                 if (flags & BPF_FIB_LOOKUP_TBID) {
5817                         tbid = params->tbid;
5818                         /* zero out for vlan output */
5819                         params->tbid = 0;
5820                 }
5821
5822                 tb = fib_get_table(net, tbid);
5823                 if (unlikely(!tb))
5824                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5825
5826                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5827         } else {
5828                 fl4.flowi4_mark = 0;
5829                 fl4.flowi4_secid = 0;
5830                 fl4.flowi4_tun_key.tun_id = 0;
5831                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5832
5833                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5834         }
5835
5836         if (err) {
5837                 /* map fib lookup errors to RTN_ type */
5838                 if (err == -EINVAL)
5839                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5840                 if (err == -EHOSTUNREACH)
5841                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5842                 if (err == -EACCES)
5843                         return BPF_FIB_LKUP_RET_PROHIBIT;
5844
5845                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5846         }
5847
5848         if (res.type != RTN_UNICAST)
5849                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5850
5851         if (fib_info_num_path(res.fi) > 1)
5852                 fib_select_path(net, &res, &fl4, NULL);
5853
5854         if (check_mtu) {
5855                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5856                 if (params->tot_len > mtu) {
5857                         params->mtu_result = mtu; /* union with tot_len */
5858                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5859                 }
5860         }
5861
5862         nhc = res.nhc;
5863
5864         /* do not handle lwt encaps right now */
5865         if (nhc->nhc_lwtstate)
5866                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5867
5868         dev = nhc->nhc_dev;
5869
5870         params->rt_metric = res.fi->fib_priority;
5871         params->ifindex = dev->ifindex;
5872
5873         /* xdp and cls_bpf programs are run in RCU-bh so
5874          * rcu_read_lock_bh is not needed here
5875          */
5876         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5877                 if (nhc->nhc_gw_family)
5878                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5879         } else {
5880                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5881
5882                 params->family = AF_INET6;
5883                 *dst = nhc->nhc_gw.ipv6;
5884         }
5885
5886         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5887                 goto set_fwd_params;
5888
5889         if (likely(nhc->nhc_gw_family != AF_INET6))
5890                 neigh = __ipv4_neigh_lookup_noref(dev,
5891                                                   (__force u32)params->ipv4_dst);
5892         else
5893                 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5894
5895         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5896                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5897         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5898         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5899
5900 set_fwd_params:
5901         return bpf_fib_set_fwd_params(params, mtu);
5902 }
5903 #endif
5904
5905 #if IS_ENABLED(CONFIG_IPV6)
5906 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5907                                u32 flags, bool check_mtu)
5908 {
5909         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5910         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5911         struct fib6_result res = {};
5912         struct neighbour *neigh;
5913         struct net_device *dev;
5914         struct inet6_dev *idev;
5915         struct flowi6 fl6;
5916         int strict = 0;
5917         int oif, err;
5918         u32 mtu = 0;
5919
5920         /* link local addresses are never forwarded */
5921         if (rt6_need_strict(dst) || rt6_need_strict(src))
5922                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5923
5924         dev = dev_get_by_index_rcu(net, params->ifindex);
5925         if (unlikely(!dev))
5926                 return -ENODEV;
5927
5928         idev = __in6_dev_get_safely(dev);
5929         if (unlikely(!idev || !idev->cnf.forwarding))
5930                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5931
5932         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5933                 fl6.flowi6_iif = 1;
5934                 oif = fl6.flowi6_oif = params->ifindex;
5935         } else {
5936                 oif = fl6.flowi6_iif = params->ifindex;
5937                 fl6.flowi6_oif = 0;
5938                 strict = RT6_LOOKUP_F_HAS_SADDR;
5939         }
5940         fl6.flowlabel = params->flowinfo;
5941         fl6.flowi6_scope = 0;
5942         fl6.flowi6_flags = 0;
5943         fl6.mp_hash = 0;
5944
5945         fl6.flowi6_proto = params->l4_protocol;
5946         fl6.daddr = *dst;
5947         fl6.saddr = *src;
5948         fl6.fl6_sport = params->sport;
5949         fl6.fl6_dport = params->dport;
5950
5951         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5952                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5953                 struct fib6_table *tb;
5954
5955                 if (flags & BPF_FIB_LOOKUP_TBID) {
5956                         tbid = params->tbid;
5957                         /* zero out for vlan output */
5958                         params->tbid = 0;
5959                 }
5960
5961                 tb = ipv6_stub->fib6_get_table(net, tbid);
5962                 if (unlikely(!tb))
5963                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5964
5965                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5966                                                    strict);
5967         } else {
5968                 fl6.flowi6_mark = 0;
5969                 fl6.flowi6_secid = 0;
5970                 fl6.flowi6_tun_key.tun_id = 0;
5971                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5972
5973                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5974         }
5975
5976         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5977                      res.f6i == net->ipv6.fib6_null_entry))
5978                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5979
5980         switch (res.fib6_type) {
5981         /* only unicast is forwarded */
5982         case RTN_UNICAST:
5983                 break;
5984         case RTN_BLACKHOLE:
5985                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5986         case RTN_UNREACHABLE:
5987                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5988         case RTN_PROHIBIT:
5989                 return BPF_FIB_LKUP_RET_PROHIBIT;
5990         default:
5991                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5992         }
5993
5994         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5995                                     fl6.flowi6_oif != 0, NULL, strict);
5996
5997         if (check_mtu) {
5998                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5999                 if (params->tot_len > mtu) {
6000                         params->mtu_result = mtu; /* union with tot_len */
6001                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
6002                 }
6003         }
6004
6005         if (res.nh->fib_nh_lws)
6006                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
6007
6008         if (res.nh->fib_nh_gw_family)
6009                 *dst = res.nh->fib_nh_gw6;
6010
6011         dev = res.nh->fib_nh_dev;
6012         params->rt_metric = res.f6i->fib6_metric;
6013         params->ifindex = dev->ifindex;
6014
6015         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6016                 goto set_fwd_params;
6017
6018         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6019          * not needed here.
6020          */
6021         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6022         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6023                 return BPF_FIB_LKUP_RET_NO_NEIGH;
6024         memcpy(params->dmac, neigh->ha, ETH_ALEN);
6025         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6026
6027 set_fwd_params:
6028         return bpf_fib_set_fwd_params(params, mtu);
6029 }
6030 #endif
6031
6032 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6033                              BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6034
6035 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6036            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6037 {
6038         if (plen < sizeof(*params))
6039                 return -EINVAL;
6040
6041         if (flags & ~BPF_FIB_LOOKUP_MASK)
6042                 return -EINVAL;
6043
6044         switch (params->family) {
6045 #if IS_ENABLED(CONFIG_INET)
6046         case AF_INET:
6047                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6048                                            flags, true);
6049 #endif
6050 #if IS_ENABLED(CONFIG_IPV6)
6051         case AF_INET6:
6052                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6053                                            flags, true);
6054 #endif
6055         }
6056         return -EAFNOSUPPORT;
6057 }
6058
6059 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6060         .func           = bpf_xdp_fib_lookup,
6061         .gpl_only       = true,
6062         .ret_type       = RET_INTEGER,
6063         .arg1_type      = ARG_PTR_TO_CTX,
6064         .arg2_type      = ARG_PTR_TO_MEM,
6065         .arg3_type      = ARG_CONST_SIZE,
6066         .arg4_type      = ARG_ANYTHING,
6067 };
6068
6069 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6070            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6071 {
6072         struct net *net = dev_net(skb->dev);
6073         int rc = -EAFNOSUPPORT;
6074         bool check_mtu = false;
6075
6076         if (plen < sizeof(*params))
6077                 return -EINVAL;
6078
6079         if (flags & ~BPF_FIB_LOOKUP_MASK)
6080                 return -EINVAL;
6081
6082         if (params->tot_len)
6083                 check_mtu = true;
6084
6085         switch (params->family) {
6086 #if IS_ENABLED(CONFIG_INET)
6087         case AF_INET:
6088                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6089                 break;
6090 #endif
6091 #if IS_ENABLED(CONFIG_IPV6)
6092         case AF_INET6:
6093                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6094                 break;
6095 #endif
6096         }
6097
6098         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6099                 struct net_device *dev;
6100
6101                 /* When tot_len isn't provided by user, check skb
6102                  * against MTU of FIB lookup resulting net_device
6103                  */
6104                 dev = dev_get_by_index_rcu(net, params->ifindex);
6105                 if (!is_skb_forwardable(dev, skb))
6106                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6107
6108                 params->mtu_result = dev->mtu; /* union with tot_len */
6109         }
6110
6111         return rc;
6112 }
6113
6114 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6115         .func           = bpf_skb_fib_lookup,
6116         .gpl_only       = true,
6117         .ret_type       = RET_INTEGER,
6118         .arg1_type      = ARG_PTR_TO_CTX,
6119         .arg2_type      = ARG_PTR_TO_MEM,
6120         .arg3_type      = ARG_CONST_SIZE,
6121         .arg4_type      = ARG_ANYTHING,
6122 };
6123
6124 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6125                                             u32 ifindex)
6126 {
6127         struct net *netns = dev_net(dev_curr);
6128
6129         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6130         if (ifindex == 0)
6131                 return dev_curr;
6132
6133         return dev_get_by_index_rcu(netns, ifindex);
6134 }
6135
6136 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6137            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6138 {
6139         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6140         struct net_device *dev = skb->dev;
6141         int skb_len, dev_len;
6142         int mtu;
6143
6144         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6145                 return -EINVAL;
6146
6147         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6148                 return -EINVAL;
6149
6150         dev = __dev_via_ifindex(dev, ifindex);
6151         if (unlikely(!dev))
6152                 return -ENODEV;
6153
6154         mtu = READ_ONCE(dev->mtu);
6155
6156         dev_len = mtu + dev->hard_header_len;
6157
6158         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6159         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6160
6161         skb_len += len_diff; /* minus result pass check */
6162         if (skb_len <= dev_len) {
6163                 ret = BPF_MTU_CHK_RET_SUCCESS;
6164                 goto out;
6165         }
6166         /* At this point, skb->len exceed MTU, but as it include length of all
6167          * segments, it can still be below MTU.  The SKB can possibly get
6168          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6169          * must choose if segs are to be MTU checked.
6170          */
6171         if (skb_is_gso(skb)) {
6172                 ret = BPF_MTU_CHK_RET_SUCCESS;
6173
6174                 if (flags & BPF_MTU_CHK_SEGS &&
6175                     !skb_gso_validate_network_len(skb, mtu))
6176                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6177         }
6178 out:
6179         /* BPF verifier guarantees valid pointer */
6180         *mtu_len = mtu;
6181
6182         return ret;
6183 }
6184
6185 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6186            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6187 {
6188         struct net_device *dev = xdp->rxq->dev;
6189         int xdp_len = xdp->data_end - xdp->data;
6190         int ret = BPF_MTU_CHK_RET_SUCCESS;
6191         int mtu, dev_len;
6192
6193         /* XDP variant doesn't support multi-buffer segment check (yet) */
6194         if (unlikely(flags))
6195                 return -EINVAL;
6196
6197         dev = __dev_via_ifindex(dev, ifindex);
6198         if (unlikely(!dev))
6199                 return -ENODEV;
6200
6201         mtu = READ_ONCE(dev->mtu);
6202
6203         /* Add L2-header as dev MTU is L3 size */
6204         dev_len = mtu + dev->hard_header_len;
6205
6206         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6207         if (*mtu_len)
6208                 xdp_len = *mtu_len + dev->hard_header_len;
6209
6210         xdp_len += len_diff; /* minus result pass check */
6211         if (xdp_len > dev_len)
6212                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6213
6214         /* BPF verifier guarantees valid pointer */
6215         *mtu_len = mtu;
6216
6217         return ret;
6218 }
6219
6220 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6221         .func           = bpf_skb_check_mtu,
6222         .gpl_only       = true,
6223         .ret_type       = RET_INTEGER,
6224         .arg1_type      = ARG_PTR_TO_CTX,
6225         .arg2_type      = ARG_ANYTHING,
6226         .arg3_type      = ARG_PTR_TO_INT,
6227         .arg4_type      = ARG_ANYTHING,
6228         .arg5_type      = ARG_ANYTHING,
6229 };
6230
6231 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6232         .func           = bpf_xdp_check_mtu,
6233         .gpl_only       = true,
6234         .ret_type       = RET_INTEGER,
6235         .arg1_type      = ARG_PTR_TO_CTX,
6236         .arg2_type      = ARG_ANYTHING,
6237         .arg3_type      = ARG_PTR_TO_INT,
6238         .arg4_type      = ARG_ANYTHING,
6239         .arg5_type      = ARG_ANYTHING,
6240 };
6241
6242 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6243 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6244 {
6245         int err;
6246         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6247
6248         if (!seg6_validate_srh(srh, len, false))
6249                 return -EINVAL;
6250
6251         switch (type) {
6252         case BPF_LWT_ENCAP_SEG6_INLINE:
6253                 if (skb->protocol != htons(ETH_P_IPV6))
6254                         return -EBADMSG;
6255
6256                 err = seg6_do_srh_inline(skb, srh);
6257                 break;
6258         case BPF_LWT_ENCAP_SEG6:
6259                 skb_reset_inner_headers(skb);
6260                 skb->encapsulation = 1;
6261                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6262                 break;
6263         default:
6264                 return -EINVAL;
6265         }
6266
6267         bpf_compute_data_pointers(skb);
6268         if (err)
6269                 return err;
6270
6271         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6272
6273         return seg6_lookup_nexthop(skb, NULL, 0);
6274 }
6275 #endif /* CONFIG_IPV6_SEG6_BPF */
6276
6277 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6278 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6279                              bool ingress)
6280 {
6281         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6282 }
6283 #endif
6284
6285 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6286            u32, len)
6287 {
6288         switch (type) {
6289 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6290         case BPF_LWT_ENCAP_SEG6:
6291         case BPF_LWT_ENCAP_SEG6_INLINE:
6292                 return bpf_push_seg6_encap(skb, type, hdr, len);
6293 #endif
6294 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6295         case BPF_LWT_ENCAP_IP:
6296                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6297 #endif
6298         default:
6299                 return -EINVAL;
6300         }
6301 }
6302
6303 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6304            void *, hdr, u32, len)
6305 {
6306         switch (type) {
6307 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6308         case BPF_LWT_ENCAP_IP:
6309                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6310 #endif
6311         default:
6312                 return -EINVAL;
6313         }
6314 }
6315
6316 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6317         .func           = bpf_lwt_in_push_encap,
6318         .gpl_only       = false,
6319         .ret_type       = RET_INTEGER,
6320         .arg1_type      = ARG_PTR_TO_CTX,
6321         .arg2_type      = ARG_ANYTHING,
6322         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6323         .arg4_type      = ARG_CONST_SIZE
6324 };
6325
6326 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6327         .func           = bpf_lwt_xmit_push_encap,
6328         .gpl_only       = false,
6329         .ret_type       = RET_INTEGER,
6330         .arg1_type      = ARG_PTR_TO_CTX,
6331         .arg2_type      = ARG_ANYTHING,
6332         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6333         .arg4_type      = ARG_CONST_SIZE
6334 };
6335
6336 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6337 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6338            const void *, from, u32, len)
6339 {
6340         struct seg6_bpf_srh_state *srh_state =
6341                 this_cpu_ptr(&seg6_bpf_srh_states);
6342         struct ipv6_sr_hdr *srh = srh_state->srh;
6343         void *srh_tlvs, *srh_end, *ptr;
6344         int srhoff = 0;
6345
6346         if (srh == NULL)
6347                 return -EINVAL;
6348
6349         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6350         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6351
6352         ptr = skb->data + offset;
6353         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6354                 srh_state->valid = false;
6355         else if (ptr < (void *)&srh->flags ||
6356                  ptr + len > (void *)&srh->segments)
6357                 return -EFAULT;
6358
6359         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6360                 return -EFAULT;
6361         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6362                 return -EINVAL;
6363         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6364
6365         memcpy(skb->data + offset, from, len);
6366         return 0;
6367 }
6368
6369 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6370         .func           = bpf_lwt_seg6_store_bytes,
6371         .gpl_only       = false,
6372         .ret_type       = RET_INTEGER,
6373         .arg1_type      = ARG_PTR_TO_CTX,
6374         .arg2_type      = ARG_ANYTHING,
6375         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6376         .arg4_type      = ARG_CONST_SIZE
6377 };
6378
6379 static void bpf_update_srh_state(struct sk_buff *skb)
6380 {
6381         struct seg6_bpf_srh_state *srh_state =
6382                 this_cpu_ptr(&seg6_bpf_srh_states);
6383         int srhoff = 0;
6384
6385         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6386                 srh_state->srh = NULL;
6387         } else {
6388                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6389                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6390                 srh_state->valid = true;
6391         }
6392 }
6393
6394 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6395            u32, action, void *, param, u32, param_len)
6396 {
6397         struct seg6_bpf_srh_state *srh_state =
6398                 this_cpu_ptr(&seg6_bpf_srh_states);
6399         int hdroff = 0;
6400         int err;
6401
6402         switch (action) {
6403         case SEG6_LOCAL_ACTION_END_X:
6404                 if (!seg6_bpf_has_valid_srh(skb))
6405                         return -EBADMSG;
6406                 if (param_len != sizeof(struct in6_addr))
6407                         return -EINVAL;
6408                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6409         case SEG6_LOCAL_ACTION_END_T:
6410                 if (!seg6_bpf_has_valid_srh(skb))
6411                         return -EBADMSG;
6412                 if (param_len != sizeof(int))
6413                         return -EINVAL;
6414                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6415         case SEG6_LOCAL_ACTION_END_DT6:
6416                 if (!seg6_bpf_has_valid_srh(skb))
6417                         return -EBADMSG;
6418                 if (param_len != sizeof(int))
6419                         return -EINVAL;
6420
6421                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6422                         return -EBADMSG;
6423                 if (!pskb_pull(skb, hdroff))
6424                         return -EBADMSG;
6425
6426                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6427                 skb_reset_network_header(skb);
6428                 skb_reset_transport_header(skb);
6429                 skb->encapsulation = 0;
6430
6431                 bpf_compute_data_pointers(skb);
6432                 bpf_update_srh_state(skb);
6433                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6434         case SEG6_LOCAL_ACTION_END_B6:
6435                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6436                         return -EBADMSG;
6437                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6438                                           param, param_len);
6439                 if (!err)
6440                         bpf_update_srh_state(skb);
6441
6442                 return err;
6443         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6444                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6445                         return -EBADMSG;
6446                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6447                                           param, param_len);
6448                 if (!err)
6449                         bpf_update_srh_state(skb);
6450
6451                 return err;
6452         default:
6453                 return -EINVAL;
6454         }
6455 }
6456
6457 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6458         .func           = bpf_lwt_seg6_action,
6459         .gpl_only       = false,
6460         .ret_type       = RET_INTEGER,
6461         .arg1_type      = ARG_PTR_TO_CTX,
6462         .arg2_type      = ARG_ANYTHING,
6463         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6464         .arg4_type      = ARG_CONST_SIZE
6465 };
6466
6467 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6468            s32, len)
6469 {
6470         struct seg6_bpf_srh_state *srh_state =
6471                 this_cpu_ptr(&seg6_bpf_srh_states);
6472         struct ipv6_sr_hdr *srh = srh_state->srh;
6473         void *srh_end, *srh_tlvs, *ptr;
6474         struct ipv6hdr *hdr;
6475         int srhoff = 0;
6476         int ret;
6477
6478         if (unlikely(srh == NULL))
6479                 return -EINVAL;
6480
6481         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6482                         ((srh->first_segment + 1) << 4));
6483         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6484                         srh_state->hdrlen);
6485         ptr = skb->data + offset;
6486
6487         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6488                 return -EFAULT;
6489         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6490                 return -EFAULT;
6491
6492         if (len > 0) {
6493                 ret = skb_cow_head(skb, len);
6494                 if (unlikely(ret < 0))
6495                         return ret;
6496
6497                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6498         } else {
6499                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6500         }
6501
6502         bpf_compute_data_pointers(skb);
6503         if (unlikely(ret < 0))
6504                 return ret;
6505
6506         hdr = (struct ipv6hdr *)skb->data;
6507         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6508
6509         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6510                 return -EINVAL;
6511         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6512         srh_state->hdrlen += len;
6513         srh_state->valid = false;
6514         return 0;
6515 }
6516
6517 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6518         .func           = bpf_lwt_seg6_adjust_srh,
6519         .gpl_only       = false,
6520         .ret_type       = RET_INTEGER,
6521         .arg1_type      = ARG_PTR_TO_CTX,
6522         .arg2_type      = ARG_ANYTHING,
6523         .arg3_type      = ARG_ANYTHING,
6524 };
6525 #endif /* CONFIG_IPV6_SEG6_BPF */
6526
6527 #ifdef CONFIG_INET
6528 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6529                               int dif, int sdif, u8 family, u8 proto)
6530 {
6531         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6532         bool refcounted = false;
6533         struct sock *sk = NULL;
6534
6535         if (family == AF_INET) {
6536                 __be32 src4 = tuple->ipv4.saddr;
6537                 __be32 dst4 = tuple->ipv4.daddr;
6538
6539                 if (proto == IPPROTO_TCP)
6540                         sk = __inet_lookup(net, hinfo, NULL, 0,
6541                                            src4, tuple->ipv4.sport,
6542                                            dst4, tuple->ipv4.dport,
6543                                            dif, sdif, &refcounted);
6544                 else
6545                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6546                                                dst4, tuple->ipv4.dport,
6547                                                dif, sdif, net->ipv4.udp_table, NULL);
6548 #if IS_ENABLED(CONFIG_IPV6)
6549         } else {
6550                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6551                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6552
6553                 if (proto == IPPROTO_TCP)
6554                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6555                                             src6, tuple->ipv6.sport,
6556                                             dst6, ntohs(tuple->ipv6.dport),
6557                                             dif, sdif, &refcounted);
6558                 else if (likely(ipv6_bpf_stub))
6559                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6560                                                             src6, tuple->ipv6.sport,
6561                                                             dst6, tuple->ipv6.dport,
6562                                                             dif, sdif,
6563                                                             net->ipv4.udp_table, NULL);
6564 #endif
6565         }
6566
6567         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6568                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6569                 sk = NULL;
6570         }
6571         return sk;
6572 }
6573
6574 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6575  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6576  */
6577 static struct sock *
6578 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6579                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6580                  u64 flags, int sdif)
6581 {
6582         struct sock *sk = NULL;
6583         struct net *net;
6584         u8 family;
6585
6586         if (len == sizeof(tuple->ipv4))
6587                 family = AF_INET;
6588         else if (len == sizeof(tuple->ipv6))
6589                 family = AF_INET6;
6590         else
6591                 return NULL;
6592
6593         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6594                 goto out;
6595
6596         if (sdif < 0) {
6597                 if (family == AF_INET)
6598                         sdif = inet_sdif(skb);
6599                 else
6600                         sdif = inet6_sdif(skb);
6601         }
6602
6603         if ((s32)netns_id < 0) {
6604                 net = caller_net;
6605                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6606         } else {
6607                 net = get_net_ns_by_id(caller_net, netns_id);
6608                 if (unlikely(!net))
6609                         goto out;
6610                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6611                 put_net(net);
6612         }
6613
6614 out:
6615         return sk;
6616 }
6617
6618 static struct sock *
6619 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6620                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6621                 u64 flags, int sdif)
6622 {
6623         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6624                                            ifindex, proto, netns_id, flags,
6625                                            sdif);
6626
6627         if (sk) {
6628                 struct sock *sk2 = sk_to_full_sk(sk);
6629
6630                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6631                  * sock refcnt is decremented to prevent a request_sock leak.
6632                  */
6633                 if (!sk_fullsock(sk2))
6634                         sk2 = NULL;
6635                 if (sk2 != sk) {
6636                         sock_gen_put(sk);
6637                         /* Ensure there is no need to bump sk2 refcnt */
6638                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6639                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6640                                 return NULL;
6641                         }
6642                         sk = sk2;
6643                 }
6644         }
6645
6646         return sk;
6647 }
6648
6649 static struct sock *
6650 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6651                u8 proto, u64 netns_id, u64 flags)
6652 {
6653         struct net *caller_net;
6654         int ifindex;
6655
6656         if (skb->dev) {
6657                 caller_net = dev_net(skb->dev);
6658                 ifindex = skb->dev->ifindex;
6659         } else {
6660                 caller_net = sock_net(skb->sk);
6661                 ifindex = 0;
6662         }
6663
6664         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6665                                 netns_id, flags, -1);
6666 }
6667
6668 static struct sock *
6669 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6670               u8 proto, u64 netns_id, u64 flags)
6671 {
6672         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6673                                          flags);
6674
6675         if (sk) {
6676                 struct sock *sk2 = sk_to_full_sk(sk);
6677
6678                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6679                  * sock refcnt is decremented to prevent a request_sock leak.
6680                  */
6681                 if (!sk_fullsock(sk2))
6682                         sk2 = NULL;
6683                 if (sk2 != sk) {
6684                         sock_gen_put(sk);
6685                         /* Ensure there is no need to bump sk2 refcnt */
6686                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6687                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6688                                 return NULL;
6689                         }
6690                         sk = sk2;
6691                 }
6692         }
6693
6694         return sk;
6695 }
6696
6697 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6698            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6699 {
6700         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6701                                              netns_id, flags);
6702 }
6703
6704 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6705         .func           = bpf_skc_lookup_tcp,
6706         .gpl_only       = false,
6707         .pkt_access     = true,
6708         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6709         .arg1_type      = ARG_PTR_TO_CTX,
6710         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6711         .arg3_type      = ARG_CONST_SIZE,
6712         .arg4_type      = ARG_ANYTHING,
6713         .arg5_type      = ARG_ANYTHING,
6714 };
6715
6716 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6717            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6718 {
6719         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6720                                             netns_id, flags);
6721 }
6722
6723 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6724         .func           = bpf_sk_lookup_tcp,
6725         .gpl_only       = false,
6726         .pkt_access     = true,
6727         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6728         .arg1_type      = ARG_PTR_TO_CTX,
6729         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6730         .arg3_type      = ARG_CONST_SIZE,
6731         .arg4_type      = ARG_ANYTHING,
6732         .arg5_type      = ARG_ANYTHING,
6733 };
6734
6735 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6736            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6737 {
6738         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6739                                             netns_id, flags);
6740 }
6741
6742 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6743         .func           = bpf_sk_lookup_udp,
6744         .gpl_only       = false,
6745         .pkt_access     = true,
6746         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6747         .arg1_type      = ARG_PTR_TO_CTX,
6748         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6749         .arg3_type      = ARG_CONST_SIZE,
6750         .arg4_type      = ARG_ANYTHING,
6751         .arg5_type      = ARG_ANYTHING,
6752 };
6753
6754 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6755            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6756 {
6757         struct net_device *dev = skb->dev;
6758         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6759         struct net *caller_net = dev_net(dev);
6760
6761         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6762                                                ifindex, IPPROTO_TCP, netns_id,
6763                                                flags, sdif);
6764 }
6765
6766 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6767         .func           = bpf_tc_skc_lookup_tcp,
6768         .gpl_only       = false,
6769         .pkt_access     = true,
6770         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6771         .arg1_type      = ARG_PTR_TO_CTX,
6772         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6773         .arg3_type      = ARG_CONST_SIZE,
6774         .arg4_type      = ARG_ANYTHING,
6775         .arg5_type      = ARG_ANYTHING,
6776 };
6777
6778 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6779            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6780 {
6781         struct net_device *dev = skb->dev;
6782         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6783         struct net *caller_net = dev_net(dev);
6784
6785         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6786                                               ifindex, IPPROTO_TCP, netns_id,
6787                                               flags, sdif);
6788 }
6789
6790 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6791         .func           = bpf_tc_sk_lookup_tcp,
6792         .gpl_only       = false,
6793         .pkt_access     = true,
6794         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6795         .arg1_type      = ARG_PTR_TO_CTX,
6796         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6797         .arg3_type      = ARG_CONST_SIZE,
6798         .arg4_type      = ARG_ANYTHING,
6799         .arg5_type      = ARG_ANYTHING,
6800 };
6801
6802 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6803            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6804 {
6805         struct net_device *dev = skb->dev;
6806         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6807         struct net *caller_net = dev_net(dev);
6808
6809         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6810                                               ifindex, IPPROTO_UDP, netns_id,
6811                                               flags, sdif);
6812 }
6813
6814 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6815         .func           = bpf_tc_sk_lookup_udp,
6816         .gpl_only       = false,
6817         .pkt_access     = true,
6818         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6819         .arg1_type      = ARG_PTR_TO_CTX,
6820         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6821         .arg3_type      = ARG_CONST_SIZE,
6822         .arg4_type      = ARG_ANYTHING,
6823         .arg5_type      = ARG_ANYTHING,
6824 };
6825
6826 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6827 {
6828         if (sk && sk_is_refcounted(sk))
6829                 sock_gen_put(sk);
6830         return 0;
6831 }
6832
6833 static const struct bpf_func_proto bpf_sk_release_proto = {
6834         .func           = bpf_sk_release,
6835         .gpl_only       = false,
6836         .ret_type       = RET_INTEGER,
6837         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6838 };
6839
6840 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6841            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6842 {
6843         struct net_device *dev = ctx->rxq->dev;
6844         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6845         struct net *caller_net = dev_net(dev);
6846
6847         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6848                                               ifindex, IPPROTO_UDP, netns_id,
6849                                               flags, sdif);
6850 }
6851
6852 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6853         .func           = bpf_xdp_sk_lookup_udp,
6854         .gpl_only       = false,
6855         .pkt_access     = true,
6856         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6857         .arg1_type      = ARG_PTR_TO_CTX,
6858         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6859         .arg3_type      = ARG_CONST_SIZE,
6860         .arg4_type      = ARG_ANYTHING,
6861         .arg5_type      = ARG_ANYTHING,
6862 };
6863
6864 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6865            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6866 {
6867         struct net_device *dev = ctx->rxq->dev;
6868         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6869         struct net *caller_net = dev_net(dev);
6870
6871         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6872                                                ifindex, IPPROTO_TCP, netns_id,
6873                                                flags, sdif);
6874 }
6875
6876 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6877         .func           = bpf_xdp_skc_lookup_tcp,
6878         .gpl_only       = false,
6879         .pkt_access     = true,
6880         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6881         .arg1_type      = ARG_PTR_TO_CTX,
6882         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6883         .arg3_type      = ARG_CONST_SIZE,
6884         .arg4_type      = ARG_ANYTHING,
6885         .arg5_type      = ARG_ANYTHING,
6886 };
6887
6888 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6889            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6890 {
6891         struct net_device *dev = ctx->rxq->dev;
6892         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6893         struct net *caller_net = dev_net(dev);
6894
6895         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6896                                               ifindex, IPPROTO_TCP, netns_id,
6897                                               flags, sdif);
6898 }
6899
6900 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6901         .func           = bpf_xdp_sk_lookup_tcp,
6902         .gpl_only       = false,
6903         .pkt_access     = true,
6904         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6905         .arg1_type      = ARG_PTR_TO_CTX,
6906         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6907         .arg3_type      = ARG_CONST_SIZE,
6908         .arg4_type      = ARG_ANYTHING,
6909         .arg5_type      = ARG_ANYTHING,
6910 };
6911
6912 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6913            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6914 {
6915         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6916                                                sock_net(ctx->sk), 0,
6917                                                IPPROTO_TCP, netns_id, flags,
6918                                                -1);
6919 }
6920
6921 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6922         .func           = bpf_sock_addr_skc_lookup_tcp,
6923         .gpl_only       = false,
6924         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6925         .arg1_type      = ARG_PTR_TO_CTX,
6926         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6927         .arg3_type      = ARG_CONST_SIZE,
6928         .arg4_type      = ARG_ANYTHING,
6929         .arg5_type      = ARG_ANYTHING,
6930 };
6931
6932 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6933            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6934 {
6935         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6936                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6937                                               netns_id, flags, -1);
6938 }
6939
6940 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6941         .func           = bpf_sock_addr_sk_lookup_tcp,
6942         .gpl_only       = false,
6943         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6944         .arg1_type      = ARG_PTR_TO_CTX,
6945         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6946         .arg3_type      = ARG_CONST_SIZE,
6947         .arg4_type      = ARG_ANYTHING,
6948         .arg5_type      = ARG_ANYTHING,
6949 };
6950
6951 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6952            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6953 {
6954         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6955                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6956                                               netns_id, flags, -1);
6957 }
6958
6959 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6960         .func           = bpf_sock_addr_sk_lookup_udp,
6961         .gpl_only       = false,
6962         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6963         .arg1_type      = ARG_PTR_TO_CTX,
6964         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6965         .arg3_type      = ARG_CONST_SIZE,
6966         .arg4_type      = ARG_ANYTHING,
6967         .arg5_type      = ARG_ANYTHING,
6968 };
6969
6970 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6971                                   struct bpf_insn_access_aux *info)
6972 {
6973         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6974                                           icsk_retransmits))
6975                 return false;
6976
6977         if (off % size != 0)
6978                 return false;
6979
6980         switch (off) {
6981         case offsetof(struct bpf_tcp_sock, bytes_received):
6982         case offsetof(struct bpf_tcp_sock, bytes_acked):
6983                 return size == sizeof(__u64);
6984         default:
6985                 return size == sizeof(__u32);
6986         }
6987 }
6988
6989 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6990                                     const struct bpf_insn *si,
6991                                     struct bpf_insn *insn_buf,
6992                                     struct bpf_prog *prog, u32 *target_size)
6993 {
6994         struct bpf_insn *insn = insn_buf;
6995
6996 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6997         do {                                                            \
6998                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6999                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
7000                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
7001                                       si->dst_reg, si->src_reg,         \
7002                                       offsetof(struct tcp_sock, FIELD)); \
7003         } while (0)
7004
7005 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
7006         do {                                                            \
7007                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
7008                                           FIELD) >                      \
7009                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
7010                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
7011                                         struct inet_connection_sock,    \
7012                                         FIELD),                         \
7013                                       si->dst_reg, si->src_reg,         \
7014                                       offsetof(                         \
7015                                         struct inet_connection_sock,    \
7016                                         FIELD));                        \
7017         } while (0)
7018
7019         BTF_TYPE_EMIT(struct bpf_tcp_sock);
7020
7021         switch (si->off) {
7022         case offsetof(struct bpf_tcp_sock, rtt_min):
7023                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7024                              sizeof(struct minmax));
7025                 BUILD_BUG_ON(sizeof(struct minmax) <
7026                              sizeof(struct minmax_sample));
7027
7028                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7029                                       offsetof(struct tcp_sock, rtt_min) +
7030                                       offsetof(struct minmax_sample, v));
7031                 break;
7032         case offsetof(struct bpf_tcp_sock, snd_cwnd):
7033                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7034                 break;
7035         case offsetof(struct bpf_tcp_sock, srtt_us):
7036                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7037                 break;
7038         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7039                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7040                 break;
7041         case offsetof(struct bpf_tcp_sock, rcv_nxt):
7042                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7043                 break;
7044         case offsetof(struct bpf_tcp_sock, snd_nxt):
7045                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7046                 break;
7047         case offsetof(struct bpf_tcp_sock, snd_una):
7048                 BPF_TCP_SOCK_GET_COMMON(snd_una);
7049                 break;
7050         case offsetof(struct bpf_tcp_sock, mss_cache):
7051                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7052                 break;
7053         case offsetof(struct bpf_tcp_sock, ecn_flags):
7054                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7055                 break;
7056         case offsetof(struct bpf_tcp_sock, rate_delivered):
7057                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7058                 break;
7059         case offsetof(struct bpf_tcp_sock, rate_interval_us):
7060                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7061                 break;
7062         case offsetof(struct bpf_tcp_sock, packets_out):
7063                 BPF_TCP_SOCK_GET_COMMON(packets_out);
7064                 break;
7065         case offsetof(struct bpf_tcp_sock, retrans_out):
7066                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7067                 break;
7068         case offsetof(struct bpf_tcp_sock, total_retrans):
7069                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7070                 break;
7071         case offsetof(struct bpf_tcp_sock, segs_in):
7072                 BPF_TCP_SOCK_GET_COMMON(segs_in);
7073                 break;
7074         case offsetof(struct bpf_tcp_sock, data_segs_in):
7075                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7076                 break;
7077         case offsetof(struct bpf_tcp_sock, segs_out):
7078                 BPF_TCP_SOCK_GET_COMMON(segs_out);
7079                 break;
7080         case offsetof(struct bpf_tcp_sock, data_segs_out):
7081                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7082                 break;
7083         case offsetof(struct bpf_tcp_sock, lost_out):
7084                 BPF_TCP_SOCK_GET_COMMON(lost_out);
7085                 break;
7086         case offsetof(struct bpf_tcp_sock, sacked_out):
7087                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7088                 break;
7089         case offsetof(struct bpf_tcp_sock, bytes_received):
7090                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7091                 break;
7092         case offsetof(struct bpf_tcp_sock, bytes_acked):
7093                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7094                 break;
7095         case offsetof(struct bpf_tcp_sock, dsack_dups):
7096                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7097                 break;
7098         case offsetof(struct bpf_tcp_sock, delivered):
7099                 BPF_TCP_SOCK_GET_COMMON(delivered);
7100                 break;
7101         case offsetof(struct bpf_tcp_sock, delivered_ce):
7102                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7103                 break;
7104         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7105                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7106                 break;
7107         }
7108
7109         return insn - insn_buf;
7110 }
7111
7112 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7113 {
7114         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7115                 return (unsigned long)sk;
7116
7117         return (unsigned long)NULL;
7118 }
7119
7120 const struct bpf_func_proto bpf_tcp_sock_proto = {
7121         .func           = bpf_tcp_sock,
7122         .gpl_only       = false,
7123         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
7124         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7125 };
7126
7127 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7128 {
7129         sk = sk_to_full_sk(sk);
7130
7131         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7132                 return (unsigned long)sk;
7133
7134         return (unsigned long)NULL;
7135 }
7136
7137 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7138         .func           = bpf_get_listener_sock,
7139         .gpl_only       = false,
7140         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7141         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7142 };
7143
7144 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7145 {
7146         unsigned int iphdr_len;
7147
7148         switch (skb_protocol(skb, true)) {
7149         case cpu_to_be16(ETH_P_IP):
7150                 iphdr_len = sizeof(struct iphdr);
7151                 break;
7152         case cpu_to_be16(ETH_P_IPV6):
7153                 iphdr_len = sizeof(struct ipv6hdr);
7154                 break;
7155         default:
7156                 return 0;
7157         }
7158
7159         if (skb_headlen(skb) < iphdr_len)
7160                 return 0;
7161
7162         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7163                 return 0;
7164
7165         return INET_ECN_set_ce(skb);
7166 }
7167
7168 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7169                                   struct bpf_insn_access_aux *info)
7170 {
7171         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7172                 return false;
7173
7174         if (off % size != 0)
7175                 return false;
7176
7177         switch (off) {
7178         default:
7179                 return size == sizeof(__u32);
7180         }
7181 }
7182
7183 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7184                                     const struct bpf_insn *si,
7185                                     struct bpf_insn *insn_buf,
7186                                     struct bpf_prog *prog, u32 *target_size)
7187 {
7188         struct bpf_insn *insn = insn_buf;
7189
7190 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7191         do {                                                            \
7192                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7193                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7194                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7195                                       si->dst_reg, si->src_reg,         \
7196                                       offsetof(struct xdp_sock, FIELD)); \
7197         } while (0)
7198
7199         switch (si->off) {
7200         case offsetof(struct bpf_xdp_sock, queue_id):
7201                 BPF_XDP_SOCK_GET(queue_id);
7202                 break;
7203         }
7204
7205         return insn - insn_buf;
7206 }
7207
7208 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7209         .func           = bpf_skb_ecn_set_ce,
7210         .gpl_only       = false,
7211         .ret_type       = RET_INTEGER,
7212         .arg1_type      = ARG_PTR_TO_CTX,
7213 };
7214
7215 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7216            struct tcphdr *, th, u32, th_len)
7217 {
7218 #ifdef CONFIG_SYN_COOKIES
7219         u32 cookie;
7220         int ret;
7221
7222         if (unlikely(!sk || th_len < sizeof(*th)))
7223                 return -EINVAL;
7224
7225         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7226         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7227                 return -EINVAL;
7228
7229         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7230                 return -EINVAL;
7231
7232         if (!th->ack || th->rst || th->syn)
7233                 return -ENOENT;
7234
7235         if (unlikely(iph_len < sizeof(struct iphdr)))
7236                 return -EINVAL;
7237
7238         if (tcp_synq_no_recent_overflow(sk))
7239                 return -ENOENT;
7240
7241         cookie = ntohl(th->ack_seq) - 1;
7242
7243         /* Both struct iphdr and struct ipv6hdr have the version field at the
7244          * same offset so we can cast to the shorter header (struct iphdr).
7245          */
7246         switch (((struct iphdr *)iph)->version) {
7247         case 4:
7248                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7249                         return -EINVAL;
7250
7251                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7252                 break;
7253
7254 #if IS_BUILTIN(CONFIG_IPV6)
7255         case 6:
7256                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7257                         return -EINVAL;
7258
7259                 if (sk->sk_family != AF_INET6)
7260                         return -EINVAL;
7261
7262                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7263                 break;
7264 #endif /* CONFIG_IPV6 */
7265
7266         default:
7267                 return -EPROTONOSUPPORT;
7268         }
7269
7270         if (ret > 0)
7271                 return 0;
7272
7273         return -ENOENT;
7274 #else
7275         return -ENOTSUPP;
7276 #endif
7277 }
7278
7279 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7280         .func           = bpf_tcp_check_syncookie,
7281         .gpl_only       = true,
7282         .pkt_access     = true,
7283         .ret_type       = RET_INTEGER,
7284         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7285         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7286         .arg3_type      = ARG_CONST_SIZE,
7287         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7288         .arg5_type      = ARG_CONST_SIZE,
7289 };
7290
7291 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7292            struct tcphdr *, th, u32, th_len)
7293 {
7294 #ifdef CONFIG_SYN_COOKIES
7295         u32 cookie;
7296         u16 mss;
7297
7298         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7299                 return -EINVAL;
7300
7301         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7302                 return -EINVAL;
7303
7304         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7305                 return -ENOENT;
7306
7307         if (!th->syn || th->ack || th->fin || th->rst)
7308                 return -EINVAL;
7309
7310         if (unlikely(iph_len < sizeof(struct iphdr)))
7311                 return -EINVAL;
7312
7313         /* Both struct iphdr and struct ipv6hdr have the version field at the
7314          * same offset so we can cast to the shorter header (struct iphdr).
7315          */
7316         switch (((struct iphdr *)iph)->version) {
7317         case 4:
7318                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7319                         return -EINVAL;
7320
7321                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7322                 break;
7323
7324 #if IS_BUILTIN(CONFIG_IPV6)
7325         case 6:
7326                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7327                         return -EINVAL;
7328
7329                 if (sk->sk_family != AF_INET6)
7330                         return -EINVAL;
7331
7332                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7333                 break;
7334 #endif /* CONFIG_IPV6 */
7335
7336         default:
7337                 return -EPROTONOSUPPORT;
7338         }
7339         if (mss == 0)
7340                 return -ENOENT;
7341
7342         return cookie | ((u64)mss << 32);
7343 #else
7344         return -EOPNOTSUPP;
7345 #endif /* CONFIG_SYN_COOKIES */
7346 }
7347
7348 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7349         .func           = bpf_tcp_gen_syncookie,
7350         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7351         .pkt_access     = true,
7352         .ret_type       = RET_INTEGER,
7353         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7354         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7355         .arg3_type      = ARG_CONST_SIZE,
7356         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7357         .arg5_type      = ARG_CONST_SIZE,
7358 };
7359
7360 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7361 {
7362         if (!sk || flags != 0)
7363                 return -EINVAL;
7364         if (!skb_at_tc_ingress(skb))
7365                 return -EOPNOTSUPP;
7366         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7367                 return -ENETUNREACH;
7368         if (sk_unhashed(sk))
7369                 return -EOPNOTSUPP;
7370         if (sk_is_refcounted(sk) &&
7371             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7372                 return -ENOENT;
7373
7374         skb_orphan(skb);
7375         skb->sk = sk;
7376         skb->destructor = sock_pfree;
7377
7378         return 0;
7379 }
7380
7381 static const struct bpf_func_proto bpf_sk_assign_proto = {
7382         .func           = bpf_sk_assign,
7383         .gpl_only       = false,
7384         .ret_type       = RET_INTEGER,
7385         .arg1_type      = ARG_PTR_TO_CTX,
7386         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7387         .arg3_type      = ARG_ANYTHING,
7388 };
7389
7390 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7391                                     u8 search_kind, const u8 *magic,
7392                                     u8 magic_len, bool *eol)
7393 {
7394         u8 kind, kind_len;
7395
7396         *eol = false;
7397
7398         while (op < opend) {
7399                 kind = op[0];
7400
7401                 if (kind == TCPOPT_EOL) {
7402                         *eol = true;
7403                         return ERR_PTR(-ENOMSG);
7404                 } else if (kind == TCPOPT_NOP) {
7405                         op++;
7406                         continue;
7407                 }
7408
7409                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7410                         /* Something is wrong in the received header.
7411                          * Follow the TCP stack's tcp_parse_options()
7412                          * and just bail here.
7413                          */
7414                         return ERR_PTR(-EFAULT);
7415
7416                 kind_len = op[1];
7417                 if (search_kind == kind) {
7418                         if (!magic_len)
7419                                 return op;
7420
7421                         if (magic_len > kind_len - 2)
7422                                 return ERR_PTR(-ENOMSG);
7423
7424                         if (!memcmp(&op[2], magic, magic_len))
7425                                 return op;
7426                 }
7427
7428                 op += kind_len;
7429         }
7430
7431         return ERR_PTR(-ENOMSG);
7432 }
7433
7434 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7435            void *, search_res, u32, len, u64, flags)
7436 {
7437         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7438         const u8 *op, *opend, *magic, *search = search_res;
7439         u8 search_kind, search_len, copy_len, magic_len;
7440         int ret;
7441
7442         /* 2 byte is the minimal option len except TCPOPT_NOP and
7443          * TCPOPT_EOL which are useless for the bpf prog to learn
7444          * and this helper disallow loading them also.
7445          */
7446         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7447                 return -EINVAL;
7448
7449         search_kind = search[0];
7450         search_len = search[1];
7451
7452         if (search_len > len || search_kind == TCPOPT_NOP ||
7453             search_kind == TCPOPT_EOL)
7454                 return -EINVAL;
7455
7456         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7457                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7458                 if (search_len != 4 && search_len != 6)
7459                         return -EINVAL;
7460                 magic = &search[2];
7461                 magic_len = search_len - 2;
7462         } else {
7463                 if (search_len)
7464                         return -EINVAL;
7465                 magic = NULL;
7466                 magic_len = 0;
7467         }
7468
7469         if (load_syn) {
7470                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7471                 if (ret < 0)
7472                         return ret;
7473
7474                 opend = op + ret;
7475                 op += sizeof(struct tcphdr);
7476         } else {
7477                 if (!bpf_sock->skb ||
7478                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7479                         /* This bpf_sock->op cannot call this helper */
7480                         return -EPERM;
7481
7482                 opend = bpf_sock->skb_data_end;
7483                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7484         }
7485
7486         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7487                                 &eol);
7488         if (IS_ERR(op))
7489                 return PTR_ERR(op);
7490
7491         copy_len = op[1];
7492         ret = copy_len;
7493         if (copy_len > len) {
7494                 ret = -ENOSPC;
7495                 copy_len = len;
7496         }
7497
7498         memcpy(search_res, op, copy_len);
7499         return ret;
7500 }
7501
7502 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7503         .func           = bpf_sock_ops_load_hdr_opt,
7504         .gpl_only       = false,
7505         .ret_type       = RET_INTEGER,
7506         .arg1_type      = ARG_PTR_TO_CTX,
7507         .arg2_type      = ARG_PTR_TO_MEM,
7508         .arg3_type      = ARG_CONST_SIZE,
7509         .arg4_type      = ARG_ANYTHING,
7510 };
7511
7512 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7513            const void *, from, u32, len, u64, flags)
7514 {
7515         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7516         const u8 *op, *new_op, *magic = NULL;
7517         struct sk_buff *skb;
7518         bool eol;
7519
7520         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7521                 return -EPERM;
7522
7523         if (len < 2 || flags)
7524                 return -EINVAL;
7525
7526         new_op = from;
7527         new_kind = new_op[0];
7528         new_kind_len = new_op[1];
7529
7530         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7531             new_kind == TCPOPT_EOL)
7532                 return -EINVAL;
7533
7534         if (new_kind_len > bpf_sock->remaining_opt_len)
7535                 return -ENOSPC;
7536
7537         /* 253 is another experimental kind */
7538         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7539                 if (new_kind_len < 4)
7540                         return -EINVAL;
7541                 /* Match for the 2 byte magic also.
7542                  * RFC 6994: the magic could be 2 or 4 bytes.
7543                  * Hence, matching by 2 byte only is on the
7544                  * conservative side but it is the right
7545                  * thing to do for the 'search-for-duplication'
7546                  * purpose.
7547                  */
7548                 magic = &new_op[2];
7549                 magic_len = 2;
7550         }
7551
7552         /* Check for duplication */
7553         skb = bpf_sock->skb;
7554         op = skb->data + sizeof(struct tcphdr);
7555         opend = bpf_sock->skb_data_end;
7556
7557         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7558                                 &eol);
7559         if (!IS_ERR(op))
7560                 return -EEXIST;
7561
7562         if (PTR_ERR(op) != -ENOMSG)
7563                 return PTR_ERR(op);
7564
7565         if (eol)
7566                 /* The option has been ended.  Treat it as no more
7567                  * header option can be written.
7568                  */
7569                 return -ENOSPC;
7570
7571         /* No duplication found.  Store the header option. */
7572         memcpy(opend, from, new_kind_len);
7573
7574         bpf_sock->remaining_opt_len -= new_kind_len;
7575         bpf_sock->skb_data_end += new_kind_len;
7576
7577         return 0;
7578 }
7579
7580 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7581         .func           = bpf_sock_ops_store_hdr_opt,
7582         .gpl_only       = false,
7583         .ret_type       = RET_INTEGER,
7584         .arg1_type      = ARG_PTR_TO_CTX,
7585         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7586         .arg3_type      = ARG_CONST_SIZE,
7587         .arg4_type      = ARG_ANYTHING,
7588 };
7589
7590 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7591            u32, len, u64, flags)
7592 {
7593         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7594                 return -EPERM;
7595
7596         if (flags || len < 2)
7597                 return -EINVAL;
7598
7599         if (len > bpf_sock->remaining_opt_len)
7600                 return -ENOSPC;
7601
7602         bpf_sock->remaining_opt_len -= len;
7603
7604         return 0;
7605 }
7606
7607 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7608         .func           = bpf_sock_ops_reserve_hdr_opt,
7609         .gpl_only       = false,
7610         .ret_type       = RET_INTEGER,
7611         .arg1_type      = ARG_PTR_TO_CTX,
7612         .arg2_type      = ARG_ANYTHING,
7613         .arg3_type      = ARG_ANYTHING,
7614 };
7615
7616 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7617            u64, tstamp, u32, tstamp_type)
7618 {
7619         /* skb_clear_delivery_time() is done for inet protocol */
7620         if (skb->protocol != htons(ETH_P_IP) &&
7621             skb->protocol != htons(ETH_P_IPV6))
7622                 return -EOPNOTSUPP;
7623
7624         switch (tstamp_type) {
7625         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7626                 if (!tstamp)
7627                         return -EINVAL;
7628                 skb->tstamp = tstamp;
7629                 skb->mono_delivery_time = 1;
7630                 break;
7631         case BPF_SKB_TSTAMP_UNSPEC:
7632                 if (tstamp)
7633                         return -EINVAL;
7634                 skb->tstamp = 0;
7635                 skb->mono_delivery_time = 0;
7636                 break;
7637         default:
7638                 return -EINVAL;
7639         }
7640
7641         return 0;
7642 }
7643
7644 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7645         .func           = bpf_skb_set_tstamp,
7646         .gpl_only       = false,
7647         .ret_type       = RET_INTEGER,
7648         .arg1_type      = ARG_PTR_TO_CTX,
7649         .arg2_type      = ARG_ANYTHING,
7650         .arg3_type      = ARG_ANYTHING,
7651 };
7652
7653 #ifdef CONFIG_SYN_COOKIES
7654 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7655            struct tcphdr *, th, u32, th_len)
7656 {
7657         u32 cookie;
7658         u16 mss;
7659
7660         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7661                 return -EINVAL;
7662
7663         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7664         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7665
7666         return cookie | ((u64)mss << 32);
7667 }
7668
7669 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7670         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7671         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7672         .pkt_access     = true,
7673         .ret_type       = RET_INTEGER,
7674         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7675         .arg1_size      = sizeof(struct iphdr),
7676         .arg2_type      = ARG_PTR_TO_MEM,
7677         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7678 };
7679
7680 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7681            struct tcphdr *, th, u32, th_len)
7682 {
7683 #if IS_BUILTIN(CONFIG_IPV6)
7684         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7685                 sizeof(struct ipv6hdr);
7686         u32 cookie;
7687         u16 mss;
7688
7689         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7690                 return -EINVAL;
7691
7692         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7693         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7694
7695         return cookie | ((u64)mss << 32);
7696 #else
7697         return -EPROTONOSUPPORT;
7698 #endif
7699 }
7700
7701 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7702         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7703         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7704         .pkt_access     = true,
7705         .ret_type       = RET_INTEGER,
7706         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7707         .arg1_size      = sizeof(struct ipv6hdr),
7708         .arg2_type      = ARG_PTR_TO_MEM,
7709         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7710 };
7711
7712 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7713            struct tcphdr *, th)
7714 {
7715         u32 cookie = ntohl(th->ack_seq) - 1;
7716
7717         if (__cookie_v4_check(iph, th, cookie) > 0)
7718                 return 0;
7719
7720         return -EACCES;
7721 }
7722
7723 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7724         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7725         .gpl_only       = true, /* __cookie_v4_check is GPL */
7726         .pkt_access     = true,
7727         .ret_type       = RET_INTEGER,
7728         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7729         .arg1_size      = sizeof(struct iphdr),
7730         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7731         .arg2_size      = sizeof(struct tcphdr),
7732 };
7733
7734 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7735            struct tcphdr *, th)
7736 {
7737 #if IS_BUILTIN(CONFIG_IPV6)
7738         u32 cookie = ntohl(th->ack_seq) - 1;
7739
7740         if (__cookie_v6_check(iph, th, cookie) > 0)
7741                 return 0;
7742
7743         return -EACCES;
7744 #else
7745         return -EPROTONOSUPPORT;
7746 #endif
7747 }
7748
7749 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7750         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7751         .gpl_only       = true, /* __cookie_v6_check is GPL */
7752         .pkt_access     = true,
7753         .ret_type       = RET_INTEGER,
7754         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7755         .arg1_size      = sizeof(struct ipv6hdr),
7756         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7757         .arg2_size      = sizeof(struct tcphdr),
7758 };
7759 #endif /* CONFIG_SYN_COOKIES */
7760
7761 #endif /* CONFIG_INET */
7762
7763 bool bpf_helper_changes_pkt_data(void *func)
7764 {
7765         if (func == bpf_skb_vlan_push ||
7766             func == bpf_skb_vlan_pop ||
7767             func == bpf_skb_store_bytes ||
7768             func == bpf_skb_change_proto ||
7769             func == bpf_skb_change_head ||
7770             func == sk_skb_change_head ||
7771             func == bpf_skb_change_tail ||
7772             func == sk_skb_change_tail ||
7773             func == bpf_skb_adjust_room ||
7774             func == sk_skb_adjust_room ||
7775             func == bpf_skb_pull_data ||
7776             func == sk_skb_pull_data ||
7777             func == bpf_clone_redirect ||
7778             func == bpf_l3_csum_replace ||
7779             func == bpf_l4_csum_replace ||
7780             func == bpf_xdp_adjust_head ||
7781             func == bpf_xdp_adjust_meta ||
7782             func == bpf_msg_pull_data ||
7783             func == bpf_msg_push_data ||
7784             func == bpf_msg_pop_data ||
7785             func == bpf_xdp_adjust_tail ||
7786 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7787             func == bpf_lwt_seg6_store_bytes ||
7788             func == bpf_lwt_seg6_adjust_srh ||
7789             func == bpf_lwt_seg6_action ||
7790 #endif
7791 #ifdef CONFIG_INET
7792             func == bpf_sock_ops_store_hdr_opt ||
7793 #endif
7794             func == bpf_lwt_in_push_encap ||
7795             func == bpf_lwt_xmit_push_encap)
7796                 return true;
7797
7798         return false;
7799 }
7800
7801 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7802 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7803
7804 static const struct bpf_func_proto *
7805 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7806 {
7807         const struct bpf_func_proto *func_proto;
7808
7809         func_proto = cgroup_common_func_proto(func_id, prog);
7810         if (func_proto)
7811                 return func_proto;
7812
7813         func_proto = cgroup_current_func_proto(func_id, prog);
7814         if (func_proto)
7815                 return func_proto;
7816
7817         switch (func_id) {
7818         case BPF_FUNC_get_socket_cookie:
7819                 return &bpf_get_socket_cookie_sock_proto;
7820         case BPF_FUNC_get_netns_cookie:
7821                 return &bpf_get_netns_cookie_sock_proto;
7822         case BPF_FUNC_perf_event_output:
7823                 return &bpf_event_output_data_proto;
7824         case BPF_FUNC_sk_storage_get:
7825                 return &bpf_sk_storage_get_cg_sock_proto;
7826         case BPF_FUNC_ktime_get_coarse_ns:
7827                 return &bpf_ktime_get_coarse_ns_proto;
7828         default:
7829                 return bpf_base_func_proto(func_id);
7830         }
7831 }
7832
7833 static const struct bpf_func_proto *
7834 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7835 {
7836         const struct bpf_func_proto *func_proto;
7837
7838         func_proto = cgroup_common_func_proto(func_id, prog);
7839         if (func_proto)
7840                 return func_proto;
7841
7842         func_proto = cgroup_current_func_proto(func_id, prog);
7843         if (func_proto)
7844                 return func_proto;
7845
7846         switch (func_id) {
7847         case BPF_FUNC_bind:
7848                 switch (prog->expected_attach_type) {
7849                 case BPF_CGROUP_INET4_CONNECT:
7850                 case BPF_CGROUP_INET6_CONNECT:
7851                         return &bpf_bind_proto;
7852                 default:
7853                         return NULL;
7854                 }
7855         case BPF_FUNC_get_socket_cookie:
7856                 return &bpf_get_socket_cookie_sock_addr_proto;
7857         case BPF_FUNC_get_netns_cookie:
7858                 return &bpf_get_netns_cookie_sock_addr_proto;
7859         case BPF_FUNC_perf_event_output:
7860                 return &bpf_event_output_data_proto;
7861 #ifdef CONFIG_INET
7862         case BPF_FUNC_sk_lookup_tcp:
7863                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7864         case BPF_FUNC_sk_lookup_udp:
7865                 return &bpf_sock_addr_sk_lookup_udp_proto;
7866         case BPF_FUNC_sk_release:
7867                 return &bpf_sk_release_proto;
7868         case BPF_FUNC_skc_lookup_tcp:
7869                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7870 #endif /* CONFIG_INET */
7871         case BPF_FUNC_sk_storage_get:
7872                 return &bpf_sk_storage_get_proto;
7873         case BPF_FUNC_sk_storage_delete:
7874                 return &bpf_sk_storage_delete_proto;
7875         case BPF_FUNC_setsockopt:
7876                 switch (prog->expected_attach_type) {
7877                 case BPF_CGROUP_INET4_BIND:
7878                 case BPF_CGROUP_INET6_BIND:
7879                 case BPF_CGROUP_INET4_CONNECT:
7880                 case BPF_CGROUP_INET6_CONNECT:
7881                 case BPF_CGROUP_UDP4_RECVMSG:
7882                 case BPF_CGROUP_UDP6_RECVMSG:
7883                 case BPF_CGROUP_UDP4_SENDMSG:
7884                 case BPF_CGROUP_UDP6_SENDMSG:
7885                 case BPF_CGROUP_INET4_GETPEERNAME:
7886                 case BPF_CGROUP_INET6_GETPEERNAME:
7887                 case BPF_CGROUP_INET4_GETSOCKNAME:
7888                 case BPF_CGROUP_INET6_GETSOCKNAME:
7889                         return &bpf_sock_addr_setsockopt_proto;
7890                 default:
7891                         return NULL;
7892                 }
7893         case BPF_FUNC_getsockopt:
7894                 switch (prog->expected_attach_type) {
7895                 case BPF_CGROUP_INET4_BIND:
7896                 case BPF_CGROUP_INET6_BIND:
7897                 case BPF_CGROUP_INET4_CONNECT:
7898                 case BPF_CGROUP_INET6_CONNECT:
7899                 case BPF_CGROUP_UDP4_RECVMSG:
7900                 case BPF_CGROUP_UDP6_RECVMSG:
7901                 case BPF_CGROUP_UDP4_SENDMSG:
7902                 case BPF_CGROUP_UDP6_SENDMSG:
7903                 case BPF_CGROUP_INET4_GETPEERNAME:
7904                 case BPF_CGROUP_INET6_GETPEERNAME:
7905                 case BPF_CGROUP_INET4_GETSOCKNAME:
7906                 case BPF_CGROUP_INET6_GETSOCKNAME:
7907                         return &bpf_sock_addr_getsockopt_proto;
7908                 default:
7909                         return NULL;
7910                 }
7911         default:
7912                 return bpf_sk_base_func_proto(func_id);
7913         }
7914 }
7915
7916 static const struct bpf_func_proto *
7917 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7918 {
7919         switch (func_id) {
7920         case BPF_FUNC_skb_load_bytes:
7921                 return &bpf_skb_load_bytes_proto;
7922         case BPF_FUNC_skb_load_bytes_relative:
7923                 return &bpf_skb_load_bytes_relative_proto;
7924         case BPF_FUNC_get_socket_cookie:
7925                 return &bpf_get_socket_cookie_proto;
7926         case BPF_FUNC_get_socket_uid:
7927                 return &bpf_get_socket_uid_proto;
7928         case BPF_FUNC_perf_event_output:
7929                 return &bpf_skb_event_output_proto;
7930         default:
7931                 return bpf_sk_base_func_proto(func_id);
7932         }
7933 }
7934
7935 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7936 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7937
7938 static const struct bpf_func_proto *
7939 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7940 {
7941         const struct bpf_func_proto *func_proto;
7942
7943         func_proto = cgroup_common_func_proto(func_id, prog);
7944         if (func_proto)
7945                 return func_proto;
7946
7947         switch (func_id) {
7948         case BPF_FUNC_sk_fullsock:
7949                 return &bpf_sk_fullsock_proto;
7950         case BPF_FUNC_sk_storage_get:
7951                 return &bpf_sk_storage_get_proto;
7952         case BPF_FUNC_sk_storage_delete:
7953                 return &bpf_sk_storage_delete_proto;
7954         case BPF_FUNC_perf_event_output:
7955                 return &bpf_skb_event_output_proto;
7956 #ifdef CONFIG_SOCK_CGROUP_DATA
7957         case BPF_FUNC_skb_cgroup_id:
7958                 return &bpf_skb_cgroup_id_proto;
7959         case BPF_FUNC_skb_ancestor_cgroup_id:
7960                 return &bpf_skb_ancestor_cgroup_id_proto;
7961         case BPF_FUNC_sk_cgroup_id:
7962                 return &bpf_sk_cgroup_id_proto;
7963         case BPF_FUNC_sk_ancestor_cgroup_id:
7964                 return &bpf_sk_ancestor_cgroup_id_proto;
7965 #endif
7966 #ifdef CONFIG_INET
7967         case BPF_FUNC_sk_lookup_tcp:
7968                 return &bpf_sk_lookup_tcp_proto;
7969         case BPF_FUNC_sk_lookup_udp:
7970                 return &bpf_sk_lookup_udp_proto;
7971         case BPF_FUNC_sk_release:
7972                 return &bpf_sk_release_proto;
7973         case BPF_FUNC_skc_lookup_tcp:
7974                 return &bpf_skc_lookup_tcp_proto;
7975         case BPF_FUNC_tcp_sock:
7976                 return &bpf_tcp_sock_proto;
7977         case BPF_FUNC_get_listener_sock:
7978                 return &bpf_get_listener_sock_proto;
7979         case BPF_FUNC_skb_ecn_set_ce:
7980                 return &bpf_skb_ecn_set_ce_proto;
7981 #endif
7982         default:
7983                 return sk_filter_func_proto(func_id, prog);
7984         }
7985 }
7986
7987 static const struct bpf_func_proto *
7988 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7989 {
7990         switch (func_id) {
7991         case BPF_FUNC_skb_store_bytes:
7992                 return &bpf_skb_store_bytes_proto;
7993         case BPF_FUNC_skb_load_bytes:
7994                 return &bpf_skb_load_bytes_proto;
7995         case BPF_FUNC_skb_load_bytes_relative:
7996                 return &bpf_skb_load_bytes_relative_proto;
7997         case BPF_FUNC_skb_pull_data:
7998                 return &bpf_skb_pull_data_proto;
7999         case BPF_FUNC_csum_diff:
8000                 return &bpf_csum_diff_proto;
8001         case BPF_FUNC_csum_update:
8002                 return &bpf_csum_update_proto;
8003         case BPF_FUNC_csum_level:
8004                 return &bpf_csum_level_proto;
8005         case BPF_FUNC_l3_csum_replace:
8006                 return &bpf_l3_csum_replace_proto;
8007         case BPF_FUNC_l4_csum_replace:
8008                 return &bpf_l4_csum_replace_proto;
8009         case BPF_FUNC_clone_redirect:
8010                 return &bpf_clone_redirect_proto;
8011         case BPF_FUNC_get_cgroup_classid:
8012                 return &bpf_get_cgroup_classid_proto;
8013         case BPF_FUNC_skb_vlan_push:
8014                 return &bpf_skb_vlan_push_proto;
8015         case BPF_FUNC_skb_vlan_pop:
8016                 return &bpf_skb_vlan_pop_proto;
8017         case BPF_FUNC_skb_change_proto:
8018                 return &bpf_skb_change_proto_proto;
8019         case BPF_FUNC_skb_change_type:
8020                 return &bpf_skb_change_type_proto;
8021         case BPF_FUNC_skb_adjust_room:
8022                 return &bpf_skb_adjust_room_proto;
8023         case BPF_FUNC_skb_change_tail:
8024                 return &bpf_skb_change_tail_proto;
8025         case BPF_FUNC_skb_change_head:
8026                 return &bpf_skb_change_head_proto;
8027         case BPF_FUNC_skb_get_tunnel_key:
8028                 return &bpf_skb_get_tunnel_key_proto;
8029         case BPF_FUNC_skb_set_tunnel_key:
8030                 return bpf_get_skb_set_tunnel_proto(func_id);
8031         case BPF_FUNC_skb_get_tunnel_opt:
8032                 return &bpf_skb_get_tunnel_opt_proto;
8033         case BPF_FUNC_skb_set_tunnel_opt:
8034                 return bpf_get_skb_set_tunnel_proto(func_id);
8035         case BPF_FUNC_redirect:
8036                 return &bpf_redirect_proto;
8037         case BPF_FUNC_redirect_neigh:
8038                 return &bpf_redirect_neigh_proto;
8039         case BPF_FUNC_redirect_peer:
8040                 return &bpf_redirect_peer_proto;
8041         case BPF_FUNC_get_route_realm:
8042                 return &bpf_get_route_realm_proto;
8043         case BPF_FUNC_get_hash_recalc:
8044                 return &bpf_get_hash_recalc_proto;
8045         case BPF_FUNC_set_hash_invalid:
8046                 return &bpf_set_hash_invalid_proto;
8047         case BPF_FUNC_set_hash:
8048                 return &bpf_set_hash_proto;
8049         case BPF_FUNC_perf_event_output:
8050                 return &bpf_skb_event_output_proto;
8051         case BPF_FUNC_get_smp_processor_id:
8052                 return &bpf_get_smp_processor_id_proto;
8053         case BPF_FUNC_skb_under_cgroup:
8054                 return &bpf_skb_under_cgroup_proto;
8055         case BPF_FUNC_get_socket_cookie:
8056                 return &bpf_get_socket_cookie_proto;
8057         case BPF_FUNC_get_socket_uid:
8058                 return &bpf_get_socket_uid_proto;
8059         case BPF_FUNC_fib_lookup:
8060                 return &bpf_skb_fib_lookup_proto;
8061         case BPF_FUNC_check_mtu:
8062                 return &bpf_skb_check_mtu_proto;
8063         case BPF_FUNC_sk_fullsock:
8064                 return &bpf_sk_fullsock_proto;
8065         case BPF_FUNC_sk_storage_get:
8066                 return &bpf_sk_storage_get_proto;
8067         case BPF_FUNC_sk_storage_delete:
8068                 return &bpf_sk_storage_delete_proto;
8069 #ifdef CONFIG_XFRM
8070         case BPF_FUNC_skb_get_xfrm_state:
8071                 return &bpf_skb_get_xfrm_state_proto;
8072 #endif
8073 #ifdef CONFIG_CGROUP_NET_CLASSID
8074         case BPF_FUNC_skb_cgroup_classid:
8075                 return &bpf_skb_cgroup_classid_proto;
8076 #endif
8077 #ifdef CONFIG_SOCK_CGROUP_DATA
8078         case BPF_FUNC_skb_cgroup_id:
8079                 return &bpf_skb_cgroup_id_proto;
8080         case BPF_FUNC_skb_ancestor_cgroup_id:
8081                 return &bpf_skb_ancestor_cgroup_id_proto;
8082 #endif
8083 #ifdef CONFIG_INET
8084         case BPF_FUNC_sk_lookup_tcp:
8085                 return &bpf_tc_sk_lookup_tcp_proto;
8086         case BPF_FUNC_sk_lookup_udp:
8087                 return &bpf_tc_sk_lookup_udp_proto;
8088         case BPF_FUNC_sk_release:
8089                 return &bpf_sk_release_proto;
8090         case BPF_FUNC_tcp_sock:
8091                 return &bpf_tcp_sock_proto;
8092         case BPF_FUNC_get_listener_sock:
8093                 return &bpf_get_listener_sock_proto;
8094         case BPF_FUNC_skc_lookup_tcp:
8095                 return &bpf_tc_skc_lookup_tcp_proto;
8096         case BPF_FUNC_tcp_check_syncookie:
8097                 return &bpf_tcp_check_syncookie_proto;
8098         case BPF_FUNC_skb_ecn_set_ce:
8099                 return &bpf_skb_ecn_set_ce_proto;
8100         case BPF_FUNC_tcp_gen_syncookie:
8101                 return &bpf_tcp_gen_syncookie_proto;
8102         case BPF_FUNC_sk_assign:
8103                 return &bpf_sk_assign_proto;
8104         case BPF_FUNC_skb_set_tstamp:
8105                 return &bpf_skb_set_tstamp_proto;
8106 #ifdef CONFIG_SYN_COOKIES
8107         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8108                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8109         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8110                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8111         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8112                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8113         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8114                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8115 #endif
8116 #endif
8117         default:
8118                 return bpf_sk_base_func_proto(func_id);
8119         }
8120 }
8121
8122 static const struct bpf_func_proto *
8123 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8124 {
8125         switch (func_id) {
8126         case BPF_FUNC_perf_event_output:
8127                 return &bpf_xdp_event_output_proto;
8128         case BPF_FUNC_get_smp_processor_id:
8129                 return &bpf_get_smp_processor_id_proto;
8130         case BPF_FUNC_csum_diff:
8131                 return &bpf_csum_diff_proto;
8132         case BPF_FUNC_xdp_adjust_head:
8133                 return &bpf_xdp_adjust_head_proto;
8134         case BPF_FUNC_xdp_adjust_meta:
8135                 return &bpf_xdp_adjust_meta_proto;
8136         case BPF_FUNC_redirect:
8137                 return &bpf_xdp_redirect_proto;
8138         case BPF_FUNC_redirect_map:
8139                 return &bpf_xdp_redirect_map_proto;
8140         case BPF_FUNC_xdp_adjust_tail:
8141                 return &bpf_xdp_adjust_tail_proto;
8142         case BPF_FUNC_xdp_get_buff_len:
8143                 return &bpf_xdp_get_buff_len_proto;
8144         case BPF_FUNC_xdp_load_bytes:
8145                 return &bpf_xdp_load_bytes_proto;
8146         case BPF_FUNC_xdp_store_bytes:
8147                 return &bpf_xdp_store_bytes_proto;
8148         case BPF_FUNC_fib_lookup:
8149                 return &bpf_xdp_fib_lookup_proto;
8150         case BPF_FUNC_check_mtu:
8151                 return &bpf_xdp_check_mtu_proto;
8152 #ifdef CONFIG_INET
8153         case BPF_FUNC_sk_lookup_udp:
8154                 return &bpf_xdp_sk_lookup_udp_proto;
8155         case BPF_FUNC_sk_lookup_tcp:
8156                 return &bpf_xdp_sk_lookup_tcp_proto;
8157         case BPF_FUNC_sk_release:
8158                 return &bpf_sk_release_proto;
8159         case BPF_FUNC_skc_lookup_tcp:
8160                 return &bpf_xdp_skc_lookup_tcp_proto;
8161         case BPF_FUNC_tcp_check_syncookie:
8162                 return &bpf_tcp_check_syncookie_proto;
8163         case BPF_FUNC_tcp_gen_syncookie:
8164                 return &bpf_tcp_gen_syncookie_proto;
8165 #ifdef CONFIG_SYN_COOKIES
8166         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8167                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8168         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8169                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8170         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8171                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8172         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8173                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8174 #endif
8175 #endif
8176         default:
8177                 return bpf_sk_base_func_proto(func_id);
8178         }
8179
8180 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8181         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8182          * kfuncs are defined in two different modules, and we want to be able
8183          * to use them interchangably with the same BTF type ID. Because modules
8184          * can't de-duplicate BTF IDs between each other, we need the type to be
8185          * referenced in the vmlinux BTF or the verifier will get confused about
8186          * the different types. So we add this dummy type reference which will
8187          * be included in vmlinux BTF, allowing both modules to refer to the
8188          * same type ID.
8189          */
8190         BTF_TYPE_EMIT(struct nf_conn___init);
8191 #endif
8192 }
8193
8194 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8195 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8196
8197 static const struct bpf_func_proto *
8198 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8199 {
8200         const struct bpf_func_proto *func_proto;
8201
8202         func_proto = cgroup_common_func_proto(func_id, prog);
8203         if (func_proto)
8204                 return func_proto;
8205
8206         switch (func_id) {
8207         case BPF_FUNC_setsockopt:
8208                 return &bpf_sock_ops_setsockopt_proto;
8209         case BPF_FUNC_getsockopt:
8210                 return &bpf_sock_ops_getsockopt_proto;
8211         case BPF_FUNC_sock_ops_cb_flags_set:
8212                 return &bpf_sock_ops_cb_flags_set_proto;
8213         case BPF_FUNC_sock_map_update:
8214                 return &bpf_sock_map_update_proto;
8215         case BPF_FUNC_sock_hash_update:
8216                 return &bpf_sock_hash_update_proto;
8217         case BPF_FUNC_get_socket_cookie:
8218                 return &bpf_get_socket_cookie_sock_ops_proto;
8219         case BPF_FUNC_perf_event_output:
8220                 return &bpf_event_output_data_proto;
8221         case BPF_FUNC_sk_storage_get:
8222                 return &bpf_sk_storage_get_proto;
8223         case BPF_FUNC_sk_storage_delete:
8224                 return &bpf_sk_storage_delete_proto;
8225         case BPF_FUNC_get_netns_cookie:
8226                 return &bpf_get_netns_cookie_sock_ops_proto;
8227 #ifdef CONFIG_INET
8228         case BPF_FUNC_load_hdr_opt:
8229                 return &bpf_sock_ops_load_hdr_opt_proto;
8230         case BPF_FUNC_store_hdr_opt:
8231                 return &bpf_sock_ops_store_hdr_opt_proto;
8232         case BPF_FUNC_reserve_hdr_opt:
8233                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8234         case BPF_FUNC_tcp_sock:
8235                 return &bpf_tcp_sock_proto;
8236 #endif /* CONFIG_INET */
8237         default:
8238                 return bpf_sk_base_func_proto(func_id);
8239         }
8240 }
8241
8242 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8243 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8244
8245 static const struct bpf_func_proto *
8246 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8247 {
8248         switch (func_id) {
8249         case BPF_FUNC_msg_redirect_map:
8250                 return &bpf_msg_redirect_map_proto;
8251         case BPF_FUNC_msg_redirect_hash:
8252                 return &bpf_msg_redirect_hash_proto;
8253         case BPF_FUNC_msg_apply_bytes:
8254                 return &bpf_msg_apply_bytes_proto;
8255         case BPF_FUNC_msg_cork_bytes:
8256                 return &bpf_msg_cork_bytes_proto;
8257         case BPF_FUNC_msg_pull_data:
8258                 return &bpf_msg_pull_data_proto;
8259         case BPF_FUNC_msg_push_data:
8260                 return &bpf_msg_push_data_proto;
8261         case BPF_FUNC_msg_pop_data:
8262                 return &bpf_msg_pop_data_proto;
8263         case BPF_FUNC_perf_event_output:
8264                 return &bpf_event_output_data_proto;
8265         case BPF_FUNC_get_current_uid_gid:
8266                 return &bpf_get_current_uid_gid_proto;
8267         case BPF_FUNC_get_current_pid_tgid:
8268                 return &bpf_get_current_pid_tgid_proto;
8269         case BPF_FUNC_sk_storage_get:
8270                 return &bpf_sk_storage_get_proto;
8271         case BPF_FUNC_sk_storage_delete:
8272                 return &bpf_sk_storage_delete_proto;
8273         case BPF_FUNC_get_netns_cookie:
8274                 return &bpf_get_netns_cookie_sk_msg_proto;
8275 #ifdef CONFIG_CGROUP_NET_CLASSID
8276         case BPF_FUNC_get_cgroup_classid:
8277                 return &bpf_get_cgroup_classid_curr_proto;
8278 #endif
8279         default:
8280                 return bpf_sk_base_func_proto(func_id);
8281         }
8282 }
8283
8284 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8285 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8286
8287 static const struct bpf_func_proto *
8288 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8289 {
8290         switch (func_id) {
8291         case BPF_FUNC_skb_store_bytes:
8292                 return &bpf_skb_store_bytes_proto;
8293         case BPF_FUNC_skb_load_bytes:
8294                 return &bpf_skb_load_bytes_proto;
8295         case BPF_FUNC_skb_pull_data:
8296                 return &sk_skb_pull_data_proto;
8297         case BPF_FUNC_skb_change_tail:
8298                 return &sk_skb_change_tail_proto;
8299         case BPF_FUNC_skb_change_head:
8300                 return &sk_skb_change_head_proto;
8301         case BPF_FUNC_skb_adjust_room:
8302                 return &sk_skb_adjust_room_proto;
8303         case BPF_FUNC_get_socket_cookie:
8304                 return &bpf_get_socket_cookie_proto;
8305         case BPF_FUNC_get_socket_uid:
8306                 return &bpf_get_socket_uid_proto;
8307         case BPF_FUNC_sk_redirect_map:
8308                 return &bpf_sk_redirect_map_proto;
8309         case BPF_FUNC_sk_redirect_hash:
8310                 return &bpf_sk_redirect_hash_proto;
8311         case BPF_FUNC_perf_event_output:
8312                 return &bpf_skb_event_output_proto;
8313 #ifdef CONFIG_INET
8314         case BPF_FUNC_sk_lookup_tcp:
8315                 return &bpf_sk_lookup_tcp_proto;
8316         case BPF_FUNC_sk_lookup_udp:
8317                 return &bpf_sk_lookup_udp_proto;
8318         case BPF_FUNC_sk_release:
8319                 return &bpf_sk_release_proto;
8320         case BPF_FUNC_skc_lookup_tcp:
8321                 return &bpf_skc_lookup_tcp_proto;
8322 #endif
8323         default:
8324                 return bpf_sk_base_func_proto(func_id);
8325         }
8326 }
8327
8328 static const struct bpf_func_proto *
8329 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8330 {
8331         switch (func_id) {
8332         case BPF_FUNC_skb_load_bytes:
8333                 return &bpf_flow_dissector_load_bytes_proto;
8334         default:
8335                 return bpf_sk_base_func_proto(func_id);
8336         }
8337 }
8338
8339 static const struct bpf_func_proto *
8340 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8341 {
8342         switch (func_id) {
8343         case BPF_FUNC_skb_load_bytes:
8344                 return &bpf_skb_load_bytes_proto;
8345         case BPF_FUNC_skb_pull_data:
8346                 return &bpf_skb_pull_data_proto;
8347         case BPF_FUNC_csum_diff:
8348                 return &bpf_csum_diff_proto;
8349         case BPF_FUNC_get_cgroup_classid:
8350                 return &bpf_get_cgroup_classid_proto;
8351         case BPF_FUNC_get_route_realm:
8352                 return &bpf_get_route_realm_proto;
8353         case BPF_FUNC_get_hash_recalc:
8354                 return &bpf_get_hash_recalc_proto;
8355         case BPF_FUNC_perf_event_output:
8356                 return &bpf_skb_event_output_proto;
8357         case BPF_FUNC_get_smp_processor_id:
8358                 return &bpf_get_smp_processor_id_proto;
8359         case BPF_FUNC_skb_under_cgroup:
8360                 return &bpf_skb_under_cgroup_proto;
8361         default:
8362                 return bpf_sk_base_func_proto(func_id);
8363         }
8364 }
8365
8366 static const struct bpf_func_proto *
8367 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8368 {
8369         switch (func_id) {
8370         case BPF_FUNC_lwt_push_encap:
8371                 return &bpf_lwt_in_push_encap_proto;
8372         default:
8373                 return lwt_out_func_proto(func_id, prog);
8374         }
8375 }
8376
8377 static const struct bpf_func_proto *
8378 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8379 {
8380         switch (func_id) {
8381         case BPF_FUNC_skb_get_tunnel_key:
8382                 return &bpf_skb_get_tunnel_key_proto;
8383         case BPF_FUNC_skb_set_tunnel_key:
8384                 return bpf_get_skb_set_tunnel_proto(func_id);
8385         case BPF_FUNC_skb_get_tunnel_opt:
8386                 return &bpf_skb_get_tunnel_opt_proto;
8387         case BPF_FUNC_skb_set_tunnel_opt:
8388                 return bpf_get_skb_set_tunnel_proto(func_id);
8389         case BPF_FUNC_redirect:
8390                 return &bpf_redirect_proto;
8391         case BPF_FUNC_clone_redirect:
8392                 return &bpf_clone_redirect_proto;
8393         case BPF_FUNC_skb_change_tail:
8394                 return &bpf_skb_change_tail_proto;
8395         case BPF_FUNC_skb_change_head:
8396                 return &bpf_skb_change_head_proto;
8397         case BPF_FUNC_skb_store_bytes:
8398                 return &bpf_skb_store_bytes_proto;
8399         case BPF_FUNC_csum_update:
8400                 return &bpf_csum_update_proto;
8401         case BPF_FUNC_csum_level:
8402                 return &bpf_csum_level_proto;
8403         case BPF_FUNC_l3_csum_replace:
8404                 return &bpf_l3_csum_replace_proto;
8405         case BPF_FUNC_l4_csum_replace:
8406                 return &bpf_l4_csum_replace_proto;
8407         case BPF_FUNC_set_hash_invalid:
8408                 return &bpf_set_hash_invalid_proto;
8409         case BPF_FUNC_lwt_push_encap:
8410                 return &bpf_lwt_xmit_push_encap_proto;
8411         default:
8412                 return lwt_out_func_proto(func_id, prog);
8413         }
8414 }
8415
8416 static const struct bpf_func_proto *
8417 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8418 {
8419         switch (func_id) {
8420 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8421         case BPF_FUNC_lwt_seg6_store_bytes:
8422                 return &bpf_lwt_seg6_store_bytes_proto;
8423         case BPF_FUNC_lwt_seg6_action:
8424                 return &bpf_lwt_seg6_action_proto;
8425         case BPF_FUNC_lwt_seg6_adjust_srh:
8426                 return &bpf_lwt_seg6_adjust_srh_proto;
8427 #endif
8428         default:
8429                 return lwt_out_func_proto(func_id, prog);
8430         }
8431 }
8432
8433 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8434                                     const struct bpf_prog *prog,
8435                                     struct bpf_insn_access_aux *info)
8436 {
8437         const int size_default = sizeof(__u32);
8438
8439         if (off < 0 || off >= sizeof(struct __sk_buff))
8440                 return false;
8441
8442         /* The verifier guarantees that size > 0. */
8443         if (off % size != 0)
8444                 return false;
8445
8446         switch (off) {
8447         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8448                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8449                         return false;
8450                 break;
8451         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8452         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8453         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8454         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8455         case bpf_ctx_range(struct __sk_buff, data):
8456         case bpf_ctx_range(struct __sk_buff, data_meta):
8457         case bpf_ctx_range(struct __sk_buff, data_end):
8458                 if (size != size_default)
8459                         return false;
8460                 break;
8461         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8462                 return false;
8463         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8464                 if (type == BPF_WRITE || size != sizeof(__u64))
8465                         return false;
8466                 break;
8467         case bpf_ctx_range(struct __sk_buff, tstamp):
8468                 if (size != sizeof(__u64))
8469                         return false;
8470                 break;
8471         case offsetof(struct __sk_buff, sk):
8472                 if (type == BPF_WRITE || size != sizeof(__u64))
8473                         return false;
8474                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8475                 break;
8476         case offsetof(struct __sk_buff, tstamp_type):
8477                 return false;
8478         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8479                 /* Explicitly prohibit access to padding in __sk_buff. */
8480                 return false;
8481         default:
8482                 /* Only narrow read access allowed for now. */
8483                 if (type == BPF_WRITE) {
8484                         if (size != size_default)
8485                                 return false;
8486                 } else {
8487                         bpf_ctx_record_field_size(info, size_default);
8488                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8489                                 return false;
8490                 }
8491         }
8492
8493         return true;
8494 }
8495
8496 static bool sk_filter_is_valid_access(int off, int size,
8497                                       enum bpf_access_type type,
8498                                       const struct bpf_prog *prog,
8499                                       struct bpf_insn_access_aux *info)
8500 {
8501         switch (off) {
8502         case bpf_ctx_range(struct __sk_buff, tc_classid):
8503         case bpf_ctx_range(struct __sk_buff, data):
8504         case bpf_ctx_range(struct __sk_buff, data_meta):
8505         case bpf_ctx_range(struct __sk_buff, data_end):
8506         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8507         case bpf_ctx_range(struct __sk_buff, tstamp):
8508         case bpf_ctx_range(struct __sk_buff, wire_len):
8509         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8510                 return false;
8511         }
8512
8513         if (type == BPF_WRITE) {
8514                 switch (off) {
8515                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8516                         break;
8517                 default:
8518                         return false;
8519                 }
8520         }
8521
8522         return bpf_skb_is_valid_access(off, size, type, prog, info);
8523 }
8524
8525 static bool cg_skb_is_valid_access(int off, int size,
8526                                    enum bpf_access_type type,
8527                                    const struct bpf_prog *prog,
8528                                    struct bpf_insn_access_aux *info)
8529 {
8530         switch (off) {
8531         case bpf_ctx_range(struct __sk_buff, tc_classid):
8532         case bpf_ctx_range(struct __sk_buff, data_meta):
8533         case bpf_ctx_range(struct __sk_buff, wire_len):
8534                 return false;
8535         case bpf_ctx_range(struct __sk_buff, data):
8536         case bpf_ctx_range(struct __sk_buff, data_end):
8537                 if (!bpf_capable())
8538                         return false;
8539                 break;
8540         }
8541
8542         if (type == BPF_WRITE) {
8543                 switch (off) {
8544                 case bpf_ctx_range(struct __sk_buff, mark):
8545                 case bpf_ctx_range(struct __sk_buff, priority):
8546                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8547                         break;
8548                 case bpf_ctx_range(struct __sk_buff, tstamp):
8549                         if (!bpf_capable())
8550                                 return false;
8551                         break;
8552                 default:
8553                         return false;
8554                 }
8555         }
8556
8557         switch (off) {
8558         case bpf_ctx_range(struct __sk_buff, data):
8559                 info->reg_type = PTR_TO_PACKET;
8560                 break;
8561         case bpf_ctx_range(struct __sk_buff, data_end):
8562                 info->reg_type = PTR_TO_PACKET_END;
8563                 break;
8564         }
8565
8566         return bpf_skb_is_valid_access(off, size, type, prog, info);
8567 }
8568
8569 static bool lwt_is_valid_access(int off, int size,
8570                                 enum bpf_access_type type,
8571                                 const struct bpf_prog *prog,
8572                                 struct bpf_insn_access_aux *info)
8573 {
8574         switch (off) {
8575         case bpf_ctx_range(struct __sk_buff, tc_classid):
8576         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8577         case bpf_ctx_range(struct __sk_buff, data_meta):
8578         case bpf_ctx_range(struct __sk_buff, tstamp):
8579         case bpf_ctx_range(struct __sk_buff, wire_len):
8580         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8581                 return false;
8582         }
8583
8584         if (type == BPF_WRITE) {
8585                 switch (off) {
8586                 case bpf_ctx_range(struct __sk_buff, mark):
8587                 case bpf_ctx_range(struct __sk_buff, priority):
8588                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8589                         break;
8590                 default:
8591                         return false;
8592                 }
8593         }
8594
8595         switch (off) {
8596         case bpf_ctx_range(struct __sk_buff, data):
8597                 info->reg_type = PTR_TO_PACKET;
8598                 break;
8599         case bpf_ctx_range(struct __sk_buff, data_end):
8600                 info->reg_type = PTR_TO_PACKET_END;
8601                 break;
8602         }
8603
8604         return bpf_skb_is_valid_access(off, size, type, prog, info);
8605 }
8606
8607 /* Attach type specific accesses */
8608 static bool __sock_filter_check_attach_type(int off,
8609                                             enum bpf_access_type access_type,
8610                                             enum bpf_attach_type attach_type)
8611 {
8612         switch (off) {
8613         case offsetof(struct bpf_sock, bound_dev_if):
8614         case offsetof(struct bpf_sock, mark):
8615         case offsetof(struct bpf_sock, priority):
8616                 switch (attach_type) {
8617                 case BPF_CGROUP_INET_SOCK_CREATE:
8618                 case BPF_CGROUP_INET_SOCK_RELEASE:
8619                         goto full_access;
8620                 default:
8621                         return false;
8622                 }
8623         case bpf_ctx_range(struct bpf_sock, src_ip4):
8624                 switch (attach_type) {
8625                 case BPF_CGROUP_INET4_POST_BIND:
8626                         goto read_only;
8627                 default:
8628                         return false;
8629                 }
8630         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8631                 switch (attach_type) {
8632                 case BPF_CGROUP_INET6_POST_BIND:
8633                         goto read_only;
8634                 default:
8635                         return false;
8636                 }
8637         case bpf_ctx_range(struct bpf_sock, src_port):
8638                 switch (attach_type) {
8639                 case BPF_CGROUP_INET4_POST_BIND:
8640                 case BPF_CGROUP_INET6_POST_BIND:
8641                         goto read_only;
8642                 default:
8643                         return false;
8644                 }
8645         }
8646 read_only:
8647         return access_type == BPF_READ;
8648 full_access:
8649         return true;
8650 }
8651
8652 bool bpf_sock_common_is_valid_access(int off, int size,
8653                                      enum bpf_access_type type,
8654                                      struct bpf_insn_access_aux *info)
8655 {
8656         switch (off) {
8657         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8658                 return false;
8659         default:
8660                 return bpf_sock_is_valid_access(off, size, type, info);
8661         }
8662 }
8663
8664 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8665                               struct bpf_insn_access_aux *info)
8666 {
8667         const int size_default = sizeof(__u32);
8668         int field_size;
8669
8670         if (off < 0 || off >= sizeof(struct bpf_sock))
8671                 return false;
8672         if (off % size != 0)
8673                 return false;
8674
8675         switch (off) {
8676         case offsetof(struct bpf_sock, state):
8677         case offsetof(struct bpf_sock, family):
8678         case offsetof(struct bpf_sock, type):
8679         case offsetof(struct bpf_sock, protocol):
8680         case offsetof(struct bpf_sock, src_port):
8681         case offsetof(struct bpf_sock, rx_queue_mapping):
8682         case bpf_ctx_range(struct bpf_sock, src_ip4):
8683         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8684         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8685         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8686                 bpf_ctx_record_field_size(info, size_default);
8687                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8688         case bpf_ctx_range(struct bpf_sock, dst_port):
8689                 field_size = size == size_default ?
8690                         size_default : sizeof_field(struct bpf_sock, dst_port);
8691                 bpf_ctx_record_field_size(info, field_size);
8692                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8693         case offsetofend(struct bpf_sock, dst_port) ...
8694              offsetof(struct bpf_sock, dst_ip4) - 1:
8695                 return false;
8696         }
8697
8698         return size == size_default;
8699 }
8700
8701 static bool sock_filter_is_valid_access(int off, int size,
8702                                         enum bpf_access_type type,
8703                                         const struct bpf_prog *prog,
8704                                         struct bpf_insn_access_aux *info)
8705 {
8706         if (!bpf_sock_is_valid_access(off, size, type, info))
8707                 return false;
8708         return __sock_filter_check_attach_type(off, type,
8709                                                prog->expected_attach_type);
8710 }
8711
8712 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8713                              const struct bpf_prog *prog)
8714 {
8715         /* Neither direct read nor direct write requires any preliminary
8716          * action.
8717          */
8718         return 0;
8719 }
8720
8721 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8722                                 const struct bpf_prog *prog, int drop_verdict)
8723 {
8724         struct bpf_insn *insn = insn_buf;
8725
8726         if (!direct_write)
8727                 return 0;
8728
8729         /* if (!skb->cloned)
8730          *       goto start;
8731          *
8732          * (Fast-path, otherwise approximation that we might be
8733          *  a clone, do the rest in helper.)
8734          */
8735         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8736         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8737         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8738
8739         /* ret = bpf_skb_pull_data(skb, 0); */
8740         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8741         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8742         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8743                                BPF_FUNC_skb_pull_data);
8744         /* if (!ret)
8745          *      goto restore;
8746          * return TC_ACT_SHOT;
8747          */
8748         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8749         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8750         *insn++ = BPF_EXIT_INSN();
8751
8752         /* restore: */
8753         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8754         /* start: */
8755         *insn++ = prog->insnsi[0];
8756
8757         return insn - insn_buf;
8758 }
8759
8760 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8761                           struct bpf_insn *insn_buf)
8762 {
8763         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8764         struct bpf_insn *insn = insn_buf;
8765
8766         if (!indirect) {
8767                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8768         } else {
8769                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8770                 if (orig->imm)
8771                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8772         }
8773         /* We're guaranteed here that CTX is in R6. */
8774         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8775
8776         switch (BPF_SIZE(orig->code)) {
8777         case BPF_B:
8778                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8779                 break;
8780         case BPF_H:
8781                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8782                 break;
8783         case BPF_W:
8784                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8785                 break;
8786         }
8787
8788         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8789         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8790         *insn++ = BPF_EXIT_INSN();
8791
8792         return insn - insn_buf;
8793 }
8794
8795 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8796                                const struct bpf_prog *prog)
8797 {
8798         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8799 }
8800
8801 static bool tc_cls_act_is_valid_access(int off, int size,
8802                                        enum bpf_access_type type,
8803                                        const struct bpf_prog *prog,
8804                                        struct bpf_insn_access_aux *info)
8805 {
8806         if (type == BPF_WRITE) {
8807                 switch (off) {
8808                 case bpf_ctx_range(struct __sk_buff, mark):
8809                 case bpf_ctx_range(struct __sk_buff, tc_index):
8810                 case bpf_ctx_range(struct __sk_buff, priority):
8811                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8812                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8813                 case bpf_ctx_range(struct __sk_buff, tstamp):
8814                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8815                         break;
8816                 default:
8817                         return false;
8818                 }
8819         }
8820
8821         switch (off) {
8822         case bpf_ctx_range(struct __sk_buff, data):
8823                 info->reg_type = PTR_TO_PACKET;
8824                 break;
8825         case bpf_ctx_range(struct __sk_buff, data_meta):
8826                 info->reg_type = PTR_TO_PACKET_META;
8827                 break;
8828         case bpf_ctx_range(struct __sk_buff, data_end):
8829                 info->reg_type = PTR_TO_PACKET_END;
8830                 break;
8831         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8832                 return false;
8833         case offsetof(struct __sk_buff, tstamp_type):
8834                 /* The convert_ctx_access() on reading and writing
8835                  * __sk_buff->tstamp depends on whether the bpf prog
8836                  * has used __sk_buff->tstamp_type or not.
8837                  * Thus, we need to set prog->tstamp_type_access
8838                  * earlier during is_valid_access() here.
8839                  */
8840                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8841                 return size == sizeof(__u8);
8842         }
8843
8844         return bpf_skb_is_valid_access(off, size, type, prog, info);
8845 }
8846
8847 DEFINE_MUTEX(nf_conn_btf_access_lock);
8848 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8849
8850 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8851                               const struct bpf_reg_state *reg,
8852                               int off, int size);
8853 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8854
8855 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8856                                         const struct bpf_reg_state *reg,
8857                                         int off, int size)
8858 {
8859         int ret = -EACCES;
8860
8861         mutex_lock(&nf_conn_btf_access_lock);
8862         if (nfct_btf_struct_access)
8863                 ret = nfct_btf_struct_access(log, reg, off, size);
8864         mutex_unlock(&nf_conn_btf_access_lock);
8865
8866         return ret;
8867 }
8868
8869 static bool __is_valid_xdp_access(int off, int size)
8870 {
8871         if (off < 0 || off >= sizeof(struct xdp_md))
8872                 return false;
8873         if (off % size != 0)
8874                 return false;
8875         if (size != sizeof(__u32))
8876                 return false;
8877
8878         return true;
8879 }
8880
8881 static bool xdp_is_valid_access(int off, int size,
8882                                 enum bpf_access_type type,
8883                                 const struct bpf_prog *prog,
8884                                 struct bpf_insn_access_aux *info)
8885 {
8886         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8887                 switch (off) {
8888                 case offsetof(struct xdp_md, egress_ifindex):
8889                         return false;
8890                 }
8891         }
8892
8893         if (type == BPF_WRITE) {
8894                 if (bpf_prog_is_offloaded(prog->aux)) {
8895                         switch (off) {
8896                         case offsetof(struct xdp_md, rx_queue_index):
8897                                 return __is_valid_xdp_access(off, size);
8898                         }
8899                 }
8900                 return false;
8901         }
8902
8903         switch (off) {
8904         case offsetof(struct xdp_md, data):
8905                 info->reg_type = PTR_TO_PACKET;
8906                 break;
8907         case offsetof(struct xdp_md, data_meta):
8908                 info->reg_type = PTR_TO_PACKET_META;
8909                 break;
8910         case offsetof(struct xdp_md, data_end):
8911                 info->reg_type = PTR_TO_PACKET_END;
8912                 break;
8913         }
8914
8915         return __is_valid_xdp_access(off, size);
8916 }
8917
8918 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8919 {
8920         const u32 act_max = XDP_REDIRECT;
8921
8922         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8923                      act > act_max ? "Illegal" : "Driver unsupported",
8924                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8925 }
8926 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8927
8928 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8929                                  const struct bpf_reg_state *reg,
8930                                  int off, int size)
8931 {
8932         int ret = -EACCES;
8933
8934         mutex_lock(&nf_conn_btf_access_lock);
8935         if (nfct_btf_struct_access)
8936                 ret = nfct_btf_struct_access(log, reg, off, size);
8937         mutex_unlock(&nf_conn_btf_access_lock);
8938
8939         return ret;
8940 }
8941
8942 static bool sock_addr_is_valid_access(int off, int size,
8943                                       enum bpf_access_type type,
8944                                       const struct bpf_prog *prog,
8945                                       struct bpf_insn_access_aux *info)
8946 {
8947         const int size_default = sizeof(__u32);
8948
8949         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8950                 return false;
8951         if (off % size != 0)
8952                 return false;
8953
8954         /* Disallow access to IPv6 fields from IPv4 contex and vise
8955          * versa.
8956          */
8957         switch (off) {
8958         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8959                 switch (prog->expected_attach_type) {
8960                 case BPF_CGROUP_INET4_BIND:
8961                 case BPF_CGROUP_INET4_CONNECT:
8962                 case BPF_CGROUP_INET4_GETPEERNAME:
8963                 case BPF_CGROUP_INET4_GETSOCKNAME:
8964                 case BPF_CGROUP_UDP4_SENDMSG:
8965                 case BPF_CGROUP_UDP4_RECVMSG:
8966                         break;
8967                 default:
8968                         return false;
8969                 }
8970                 break;
8971         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8972                 switch (prog->expected_attach_type) {
8973                 case BPF_CGROUP_INET6_BIND:
8974                 case BPF_CGROUP_INET6_CONNECT:
8975                 case BPF_CGROUP_INET6_GETPEERNAME:
8976                 case BPF_CGROUP_INET6_GETSOCKNAME:
8977                 case BPF_CGROUP_UDP6_SENDMSG:
8978                 case BPF_CGROUP_UDP6_RECVMSG:
8979                         break;
8980                 default:
8981                         return false;
8982                 }
8983                 break;
8984         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8985                 switch (prog->expected_attach_type) {
8986                 case BPF_CGROUP_UDP4_SENDMSG:
8987                         break;
8988                 default:
8989                         return false;
8990                 }
8991                 break;
8992         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8993                                 msg_src_ip6[3]):
8994                 switch (prog->expected_attach_type) {
8995                 case BPF_CGROUP_UDP6_SENDMSG:
8996                         break;
8997                 default:
8998                         return false;
8999                 }
9000                 break;
9001         }
9002
9003         switch (off) {
9004         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
9005         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9006         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
9007         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9008                                 msg_src_ip6[3]):
9009         case bpf_ctx_range(struct bpf_sock_addr, user_port):
9010                 if (type == BPF_READ) {
9011                         bpf_ctx_record_field_size(info, size_default);
9012
9013                         if (bpf_ctx_wide_access_ok(off, size,
9014                                                    struct bpf_sock_addr,
9015                                                    user_ip6))
9016                                 return true;
9017
9018                         if (bpf_ctx_wide_access_ok(off, size,
9019                                                    struct bpf_sock_addr,
9020                                                    msg_src_ip6))
9021                                 return true;
9022
9023                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9024                                 return false;
9025                 } else {
9026                         if (bpf_ctx_wide_access_ok(off, size,
9027                                                    struct bpf_sock_addr,
9028                                                    user_ip6))
9029                                 return true;
9030
9031                         if (bpf_ctx_wide_access_ok(off, size,
9032                                                    struct bpf_sock_addr,
9033                                                    msg_src_ip6))
9034                                 return true;
9035
9036                         if (size != size_default)
9037                                 return false;
9038                 }
9039                 break;
9040         case offsetof(struct bpf_sock_addr, sk):
9041                 if (type != BPF_READ)
9042                         return false;
9043                 if (size != sizeof(__u64))
9044                         return false;
9045                 info->reg_type = PTR_TO_SOCKET;
9046                 break;
9047         default:
9048                 if (type == BPF_READ) {
9049                         if (size != size_default)
9050                                 return false;
9051                 } else {
9052                         return false;
9053                 }
9054         }
9055
9056         return true;
9057 }
9058
9059 static bool sock_ops_is_valid_access(int off, int size,
9060                                      enum bpf_access_type type,
9061                                      const struct bpf_prog *prog,
9062                                      struct bpf_insn_access_aux *info)
9063 {
9064         const int size_default = sizeof(__u32);
9065
9066         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9067                 return false;
9068
9069         /* The verifier guarantees that size > 0. */
9070         if (off % size != 0)
9071                 return false;
9072
9073         if (type == BPF_WRITE) {
9074                 switch (off) {
9075                 case offsetof(struct bpf_sock_ops, reply):
9076                 case offsetof(struct bpf_sock_ops, sk_txhash):
9077                         if (size != size_default)
9078                                 return false;
9079                         break;
9080                 default:
9081                         return false;
9082                 }
9083         } else {
9084                 switch (off) {
9085                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9086                                         bytes_acked):
9087                         if (size != sizeof(__u64))
9088                                 return false;
9089                         break;
9090                 case offsetof(struct bpf_sock_ops, sk):
9091                         if (size != sizeof(__u64))
9092                                 return false;
9093                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
9094                         break;
9095                 case offsetof(struct bpf_sock_ops, skb_data):
9096                         if (size != sizeof(__u64))
9097                                 return false;
9098                         info->reg_type = PTR_TO_PACKET;
9099                         break;
9100                 case offsetof(struct bpf_sock_ops, skb_data_end):
9101                         if (size != sizeof(__u64))
9102                                 return false;
9103                         info->reg_type = PTR_TO_PACKET_END;
9104                         break;
9105                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9106                         bpf_ctx_record_field_size(info, size_default);
9107                         return bpf_ctx_narrow_access_ok(off, size,
9108                                                         size_default);
9109                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9110                         if (size != sizeof(__u64))
9111                                 return false;
9112                         break;
9113                 default:
9114                         if (size != size_default)
9115                                 return false;
9116                         break;
9117                 }
9118         }
9119
9120         return true;
9121 }
9122
9123 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9124                            const struct bpf_prog *prog)
9125 {
9126         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9127 }
9128
9129 static bool sk_skb_is_valid_access(int off, int size,
9130                                    enum bpf_access_type type,
9131                                    const struct bpf_prog *prog,
9132                                    struct bpf_insn_access_aux *info)
9133 {
9134         switch (off) {
9135         case bpf_ctx_range(struct __sk_buff, tc_classid):
9136         case bpf_ctx_range(struct __sk_buff, data_meta):
9137         case bpf_ctx_range(struct __sk_buff, tstamp):
9138         case bpf_ctx_range(struct __sk_buff, wire_len):
9139         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9140                 return false;
9141         }
9142
9143         if (type == BPF_WRITE) {
9144                 switch (off) {
9145                 case bpf_ctx_range(struct __sk_buff, tc_index):
9146                 case bpf_ctx_range(struct __sk_buff, priority):
9147                         break;
9148                 default:
9149                         return false;
9150                 }
9151         }
9152
9153         switch (off) {
9154         case bpf_ctx_range(struct __sk_buff, mark):
9155                 return false;
9156         case bpf_ctx_range(struct __sk_buff, data):
9157                 info->reg_type = PTR_TO_PACKET;
9158                 break;
9159         case bpf_ctx_range(struct __sk_buff, data_end):
9160                 info->reg_type = PTR_TO_PACKET_END;
9161                 break;
9162         }
9163
9164         return bpf_skb_is_valid_access(off, size, type, prog, info);
9165 }
9166
9167 static bool sk_msg_is_valid_access(int off, int size,
9168                                    enum bpf_access_type type,
9169                                    const struct bpf_prog *prog,
9170                                    struct bpf_insn_access_aux *info)
9171 {
9172         if (type == BPF_WRITE)
9173                 return false;
9174
9175         if (off % size != 0)
9176                 return false;
9177
9178         switch (off) {
9179         case offsetof(struct sk_msg_md, data):
9180                 info->reg_type = PTR_TO_PACKET;
9181                 if (size != sizeof(__u64))
9182                         return false;
9183                 break;
9184         case offsetof(struct sk_msg_md, data_end):
9185                 info->reg_type = PTR_TO_PACKET_END;
9186                 if (size != sizeof(__u64))
9187                         return false;
9188                 break;
9189         case offsetof(struct sk_msg_md, sk):
9190                 if (size != sizeof(__u64))
9191                         return false;
9192                 info->reg_type = PTR_TO_SOCKET;
9193                 break;
9194         case bpf_ctx_range(struct sk_msg_md, family):
9195         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9196         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9197         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9198         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9199         case bpf_ctx_range(struct sk_msg_md, remote_port):
9200         case bpf_ctx_range(struct sk_msg_md, local_port):
9201         case bpf_ctx_range(struct sk_msg_md, size):
9202                 if (size != sizeof(__u32))
9203                         return false;
9204                 break;
9205         default:
9206                 return false;
9207         }
9208         return true;
9209 }
9210
9211 static bool flow_dissector_is_valid_access(int off, int size,
9212                                            enum bpf_access_type type,
9213                                            const struct bpf_prog *prog,
9214                                            struct bpf_insn_access_aux *info)
9215 {
9216         const int size_default = sizeof(__u32);
9217
9218         if (off < 0 || off >= sizeof(struct __sk_buff))
9219                 return false;
9220
9221         if (type == BPF_WRITE)
9222                 return false;
9223
9224         switch (off) {
9225         case bpf_ctx_range(struct __sk_buff, data):
9226                 if (size != size_default)
9227                         return false;
9228                 info->reg_type = PTR_TO_PACKET;
9229                 return true;
9230         case bpf_ctx_range(struct __sk_buff, data_end):
9231                 if (size != size_default)
9232                         return false;
9233                 info->reg_type = PTR_TO_PACKET_END;
9234                 return true;
9235         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9236                 if (size != sizeof(__u64))
9237                         return false;
9238                 info->reg_type = PTR_TO_FLOW_KEYS;
9239                 return true;
9240         default:
9241                 return false;
9242         }
9243 }
9244
9245 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9246                                              const struct bpf_insn *si,
9247                                              struct bpf_insn *insn_buf,
9248                                              struct bpf_prog *prog,
9249                                              u32 *target_size)
9250
9251 {
9252         struct bpf_insn *insn = insn_buf;
9253
9254         switch (si->off) {
9255         case offsetof(struct __sk_buff, data):
9256                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9257                                       si->dst_reg, si->src_reg,
9258                                       offsetof(struct bpf_flow_dissector, data));
9259                 break;
9260
9261         case offsetof(struct __sk_buff, data_end):
9262                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9263                                       si->dst_reg, si->src_reg,
9264                                       offsetof(struct bpf_flow_dissector, data_end));
9265                 break;
9266
9267         case offsetof(struct __sk_buff, flow_keys):
9268                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9269                                       si->dst_reg, si->src_reg,
9270                                       offsetof(struct bpf_flow_dissector, flow_keys));
9271                 break;
9272         }
9273
9274         return insn - insn_buf;
9275 }
9276
9277 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9278                                                      struct bpf_insn *insn)
9279 {
9280         __u8 value_reg = si->dst_reg;
9281         __u8 skb_reg = si->src_reg;
9282         /* AX is needed because src_reg and dst_reg could be the same */
9283         __u8 tmp_reg = BPF_REG_AX;
9284
9285         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9286                               SKB_BF_MONO_TC_OFFSET);
9287         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9288                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9289         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9290         *insn++ = BPF_JMP_A(1);
9291         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9292
9293         return insn;
9294 }
9295
9296 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9297                                                   struct bpf_insn *insn)
9298 {
9299         /* si->dst_reg = skb_shinfo(SKB); */
9300 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9301         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9302                               BPF_REG_AX, skb_reg,
9303                               offsetof(struct sk_buff, end));
9304         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9305                               dst_reg, skb_reg,
9306                               offsetof(struct sk_buff, head));
9307         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9308 #else
9309         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9310                               dst_reg, skb_reg,
9311                               offsetof(struct sk_buff, end));
9312 #endif
9313
9314         return insn;
9315 }
9316
9317 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9318                                                 const struct bpf_insn *si,
9319                                                 struct bpf_insn *insn)
9320 {
9321         __u8 value_reg = si->dst_reg;
9322         __u8 skb_reg = si->src_reg;
9323
9324 #ifdef CONFIG_NET_XGRESS
9325         /* If the tstamp_type is read,
9326          * the bpf prog is aware the tstamp could have delivery time.
9327          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9328          */
9329         if (!prog->tstamp_type_access) {
9330                 /* AX is needed because src_reg and dst_reg could be the same */
9331                 __u8 tmp_reg = BPF_REG_AX;
9332
9333                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9334                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9335                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9336                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9337                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9338                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9339                  * read 0 as the (rcv) timestamp.
9340                  */
9341                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9342                 *insn++ = BPF_JMP_A(1);
9343         }
9344 #endif
9345
9346         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9347                               offsetof(struct sk_buff, tstamp));
9348         return insn;
9349 }
9350
9351 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9352                                                  const struct bpf_insn *si,
9353                                                  struct bpf_insn *insn)
9354 {
9355         __u8 value_reg = si->src_reg;
9356         __u8 skb_reg = si->dst_reg;
9357
9358 #ifdef CONFIG_NET_XGRESS
9359         /* If the tstamp_type is read,
9360          * the bpf prog is aware the tstamp could have delivery time.
9361          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9362          * Otherwise, writing at ingress will have to clear the
9363          * mono_delivery_time bit also.
9364          */
9365         if (!prog->tstamp_type_access) {
9366                 __u8 tmp_reg = BPF_REG_AX;
9367
9368                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9369                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9370                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9371                 /* goto <store> */
9372                 *insn++ = BPF_JMP_A(2);
9373                 /* <clear>: mono_delivery_time */
9374                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9375                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9376         }
9377 #endif
9378
9379         /* <store>: skb->tstamp = tstamp */
9380         *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9381                                skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9382         return insn;
9383 }
9384
9385 #define BPF_EMIT_STORE(size, si, off)                                   \
9386         BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM,          \
9387                      (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9388
9389 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9390                                   const struct bpf_insn *si,
9391                                   struct bpf_insn *insn_buf,
9392                                   struct bpf_prog *prog, u32 *target_size)
9393 {
9394         struct bpf_insn *insn = insn_buf;
9395         int off;
9396
9397         switch (si->off) {
9398         case offsetof(struct __sk_buff, len):
9399                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9400                                       bpf_target_off(struct sk_buff, len, 4,
9401                                                      target_size));
9402                 break;
9403
9404         case offsetof(struct __sk_buff, protocol):
9405                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9406                                       bpf_target_off(struct sk_buff, protocol, 2,
9407                                                      target_size));
9408                 break;
9409
9410         case offsetof(struct __sk_buff, vlan_proto):
9411                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9412                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9413                                                      target_size));
9414                 break;
9415
9416         case offsetof(struct __sk_buff, priority):
9417                 if (type == BPF_WRITE)
9418                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9419                                                  bpf_target_off(struct sk_buff, priority, 4,
9420                                                                 target_size));
9421                 else
9422                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9423                                               bpf_target_off(struct sk_buff, priority, 4,
9424                                                              target_size));
9425                 break;
9426
9427         case offsetof(struct __sk_buff, ingress_ifindex):
9428                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9429                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9430                                                      target_size));
9431                 break;
9432
9433         case offsetof(struct __sk_buff, ifindex):
9434                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9435                                       si->dst_reg, si->src_reg,
9436                                       offsetof(struct sk_buff, dev));
9437                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9438                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9439                                       bpf_target_off(struct net_device, ifindex, 4,
9440                                                      target_size));
9441                 break;
9442
9443         case offsetof(struct __sk_buff, hash):
9444                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9445                                       bpf_target_off(struct sk_buff, hash, 4,
9446                                                      target_size));
9447                 break;
9448
9449         case offsetof(struct __sk_buff, mark):
9450                 if (type == BPF_WRITE)
9451                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9452                                                  bpf_target_off(struct sk_buff, mark, 4,
9453                                                                 target_size));
9454                 else
9455                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9456                                               bpf_target_off(struct sk_buff, mark, 4,
9457                                                              target_size));
9458                 break;
9459
9460         case offsetof(struct __sk_buff, pkt_type):
9461                 *target_size = 1;
9462                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9463                                       PKT_TYPE_OFFSET);
9464                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9465 #ifdef __BIG_ENDIAN_BITFIELD
9466                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9467 #endif
9468                 break;
9469
9470         case offsetof(struct __sk_buff, queue_mapping):
9471                 if (type == BPF_WRITE) {
9472                         u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9473
9474                         if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9475                                 *insn++ = BPF_JMP_A(0); /* noop */
9476                                 break;
9477                         }
9478
9479                         if (BPF_CLASS(si->code) == BPF_STX)
9480                                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9481                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9482                 } else {
9483                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9484                                               bpf_target_off(struct sk_buff,
9485                                                              queue_mapping,
9486                                                              2, target_size));
9487                 }
9488                 break;
9489
9490         case offsetof(struct __sk_buff, vlan_present):
9491                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9492                                       bpf_target_off(struct sk_buff,
9493                                                      vlan_all, 4, target_size));
9494                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9495                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9496                 break;
9497
9498         case offsetof(struct __sk_buff, vlan_tci):
9499                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9500                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9501                                                      target_size));
9502                 break;
9503
9504         case offsetof(struct __sk_buff, cb[0]) ...
9505              offsetofend(struct __sk_buff, cb[4]) - 1:
9506                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9507                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9508                               offsetof(struct qdisc_skb_cb, data)) %
9509                              sizeof(__u64));
9510
9511                 prog->cb_access = 1;
9512                 off  = si->off;
9513                 off -= offsetof(struct __sk_buff, cb[0]);
9514                 off += offsetof(struct sk_buff, cb);
9515                 off += offsetof(struct qdisc_skb_cb, data);
9516                 if (type == BPF_WRITE)
9517                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9518                 else
9519                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9520                                               si->src_reg, off);
9521                 break;
9522
9523         case offsetof(struct __sk_buff, tc_classid):
9524                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9525
9526                 off  = si->off;
9527                 off -= offsetof(struct __sk_buff, tc_classid);
9528                 off += offsetof(struct sk_buff, cb);
9529                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9530                 *target_size = 2;
9531                 if (type == BPF_WRITE)
9532                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9533                 else
9534                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9535                                               si->src_reg, off);
9536                 break;
9537
9538         case offsetof(struct __sk_buff, data):
9539                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9540                                       si->dst_reg, si->src_reg,
9541                                       offsetof(struct sk_buff, data));
9542                 break;
9543
9544         case offsetof(struct __sk_buff, data_meta):
9545                 off  = si->off;
9546                 off -= offsetof(struct __sk_buff, data_meta);
9547                 off += offsetof(struct sk_buff, cb);
9548                 off += offsetof(struct bpf_skb_data_end, data_meta);
9549                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9550                                       si->src_reg, off);
9551                 break;
9552
9553         case offsetof(struct __sk_buff, data_end):
9554                 off  = si->off;
9555                 off -= offsetof(struct __sk_buff, data_end);
9556                 off += offsetof(struct sk_buff, cb);
9557                 off += offsetof(struct bpf_skb_data_end, data_end);
9558                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9559                                       si->src_reg, off);
9560                 break;
9561
9562         case offsetof(struct __sk_buff, tc_index):
9563 #ifdef CONFIG_NET_SCHED
9564                 if (type == BPF_WRITE)
9565                         *insn++ = BPF_EMIT_STORE(BPF_H, si,
9566                                                  bpf_target_off(struct sk_buff, tc_index, 2,
9567                                                                 target_size));
9568                 else
9569                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9570                                               bpf_target_off(struct sk_buff, tc_index, 2,
9571                                                              target_size));
9572 #else
9573                 *target_size = 2;
9574                 if (type == BPF_WRITE)
9575                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9576                 else
9577                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9578 #endif
9579                 break;
9580
9581         case offsetof(struct __sk_buff, napi_id):
9582 #if defined(CONFIG_NET_RX_BUSY_POLL)
9583                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9584                                       bpf_target_off(struct sk_buff, napi_id, 4,
9585                                                      target_size));
9586                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9587                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9588 #else
9589                 *target_size = 4;
9590                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9591 #endif
9592                 break;
9593         case offsetof(struct __sk_buff, family):
9594                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9595
9596                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9597                                       si->dst_reg, si->src_reg,
9598                                       offsetof(struct sk_buff, sk));
9599                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9600                                       bpf_target_off(struct sock_common,
9601                                                      skc_family,
9602                                                      2, target_size));
9603                 break;
9604         case offsetof(struct __sk_buff, remote_ip4):
9605                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9606
9607                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9608                                       si->dst_reg, si->src_reg,
9609                                       offsetof(struct sk_buff, sk));
9610                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9611                                       bpf_target_off(struct sock_common,
9612                                                      skc_daddr,
9613                                                      4, target_size));
9614                 break;
9615         case offsetof(struct __sk_buff, local_ip4):
9616                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9617                                           skc_rcv_saddr) != 4);
9618
9619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9620                                       si->dst_reg, si->src_reg,
9621                                       offsetof(struct sk_buff, sk));
9622                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9623                                       bpf_target_off(struct sock_common,
9624                                                      skc_rcv_saddr,
9625                                                      4, target_size));
9626                 break;
9627         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9628              offsetof(struct __sk_buff, remote_ip6[3]):
9629 #if IS_ENABLED(CONFIG_IPV6)
9630                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9631                                           skc_v6_daddr.s6_addr32[0]) != 4);
9632
9633                 off = si->off;
9634                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9635
9636                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9637                                       si->dst_reg, si->src_reg,
9638                                       offsetof(struct sk_buff, sk));
9639                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9640                                       offsetof(struct sock_common,
9641                                                skc_v6_daddr.s6_addr32[0]) +
9642                                       off);
9643 #else
9644                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9645 #endif
9646                 break;
9647         case offsetof(struct __sk_buff, local_ip6[0]) ...
9648              offsetof(struct __sk_buff, local_ip6[3]):
9649 #if IS_ENABLED(CONFIG_IPV6)
9650                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9651                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9652
9653                 off = si->off;
9654                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9655
9656                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9657                                       si->dst_reg, si->src_reg,
9658                                       offsetof(struct sk_buff, sk));
9659                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9660                                       offsetof(struct sock_common,
9661                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9662                                       off);
9663 #else
9664                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9665 #endif
9666                 break;
9667
9668         case offsetof(struct __sk_buff, remote_port):
9669                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9670
9671                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9672                                       si->dst_reg, si->src_reg,
9673                                       offsetof(struct sk_buff, sk));
9674                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9675                                       bpf_target_off(struct sock_common,
9676                                                      skc_dport,
9677                                                      2, target_size));
9678 #ifndef __BIG_ENDIAN_BITFIELD
9679                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9680 #endif
9681                 break;
9682
9683         case offsetof(struct __sk_buff, local_port):
9684                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9685
9686                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9687                                       si->dst_reg, si->src_reg,
9688                                       offsetof(struct sk_buff, sk));
9689                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9690                                       bpf_target_off(struct sock_common,
9691                                                      skc_num, 2, target_size));
9692                 break;
9693
9694         case offsetof(struct __sk_buff, tstamp):
9695                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9696
9697                 if (type == BPF_WRITE)
9698                         insn = bpf_convert_tstamp_write(prog, si, insn);
9699                 else
9700                         insn = bpf_convert_tstamp_read(prog, si, insn);
9701                 break;
9702
9703         case offsetof(struct __sk_buff, tstamp_type):
9704                 insn = bpf_convert_tstamp_type_read(si, insn);
9705                 break;
9706
9707         case offsetof(struct __sk_buff, gso_segs):
9708                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9709                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9710                                       si->dst_reg, si->dst_reg,
9711                                       bpf_target_off(struct skb_shared_info,
9712                                                      gso_segs, 2,
9713                                                      target_size));
9714                 break;
9715         case offsetof(struct __sk_buff, gso_size):
9716                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9717                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9718                                       si->dst_reg, si->dst_reg,
9719                                       bpf_target_off(struct skb_shared_info,
9720                                                      gso_size, 2,
9721                                                      target_size));
9722                 break;
9723         case offsetof(struct __sk_buff, wire_len):
9724                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9725
9726                 off = si->off;
9727                 off -= offsetof(struct __sk_buff, wire_len);
9728                 off += offsetof(struct sk_buff, cb);
9729                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9730                 *target_size = 4;
9731                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9732                 break;
9733
9734         case offsetof(struct __sk_buff, sk):
9735                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9736                                       si->dst_reg, si->src_reg,
9737                                       offsetof(struct sk_buff, sk));
9738                 break;
9739         case offsetof(struct __sk_buff, hwtstamp):
9740                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9741                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9742
9743                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9744                 *insn++ = BPF_LDX_MEM(BPF_DW,
9745                                       si->dst_reg, si->dst_reg,
9746                                       bpf_target_off(struct skb_shared_info,
9747                                                      hwtstamps, 8,
9748                                                      target_size));
9749                 break;
9750         }
9751
9752         return insn - insn_buf;
9753 }
9754
9755 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9756                                 const struct bpf_insn *si,
9757                                 struct bpf_insn *insn_buf,
9758                                 struct bpf_prog *prog, u32 *target_size)
9759 {
9760         struct bpf_insn *insn = insn_buf;
9761         int off;
9762
9763         switch (si->off) {
9764         case offsetof(struct bpf_sock, bound_dev_if):
9765                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9766
9767                 if (type == BPF_WRITE)
9768                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9769                                                  offsetof(struct sock, sk_bound_dev_if));
9770                 else
9771                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9772                                       offsetof(struct sock, sk_bound_dev_if));
9773                 break;
9774
9775         case offsetof(struct bpf_sock, mark):
9776                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9777
9778                 if (type == BPF_WRITE)
9779                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9780                                                  offsetof(struct sock, sk_mark));
9781                 else
9782                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9783                                       offsetof(struct sock, sk_mark));
9784                 break;
9785
9786         case offsetof(struct bpf_sock, priority):
9787                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9788
9789                 if (type == BPF_WRITE)
9790                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9791                                                  offsetof(struct sock, sk_priority));
9792                 else
9793                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9794                                       offsetof(struct sock, sk_priority));
9795                 break;
9796
9797         case offsetof(struct bpf_sock, family):
9798                 *insn++ = BPF_LDX_MEM(
9799                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9800                         si->dst_reg, si->src_reg,
9801                         bpf_target_off(struct sock_common,
9802                                        skc_family,
9803                                        sizeof_field(struct sock_common,
9804                                                     skc_family),
9805                                        target_size));
9806                 break;
9807
9808         case offsetof(struct bpf_sock, type):
9809                 *insn++ = BPF_LDX_MEM(
9810                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9811                         si->dst_reg, si->src_reg,
9812                         bpf_target_off(struct sock, sk_type,
9813                                        sizeof_field(struct sock, sk_type),
9814                                        target_size));
9815                 break;
9816
9817         case offsetof(struct bpf_sock, protocol):
9818                 *insn++ = BPF_LDX_MEM(
9819                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9820                         si->dst_reg, si->src_reg,
9821                         bpf_target_off(struct sock, sk_protocol,
9822                                        sizeof_field(struct sock, sk_protocol),
9823                                        target_size));
9824                 break;
9825
9826         case offsetof(struct bpf_sock, src_ip4):
9827                 *insn++ = BPF_LDX_MEM(
9828                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9829                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9830                                        sizeof_field(struct sock_common,
9831                                                     skc_rcv_saddr),
9832                                        target_size));
9833                 break;
9834
9835         case offsetof(struct bpf_sock, dst_ip4):
9836                 *insn++ = BPF_LDX_MEM(
9837                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9838                         bpf_target_off(struct sock_common, skc_daddr,
9839                                        sizeof_field(struct sock_common,
9840                                                     skc_daddr),
9841                                        target_size));
9842                 break;
9843
9844         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9845 #if IS_ENABLED(CONFIG_IPV6)
9846                 off = si->off;
9847                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9848                 *insn++ = BPF_LDX_MEM(
9849                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9850                         bpf_target_off(
9851                                 struct sock_common,
9852                                 skc_v6_rcv_saddr.s6_addr32[0],
9853                                 sizeof_field(struct sock_common,
9854                                              skc_v6_rcv_saddr.s6_addr32[0]),
9855                                 target_size) + off);
9856 #else
9857                 (void)off;
9858                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9859 #endif
9860                 break;
9861
9862         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9863 #if IS_ENABLED(CONFIG_IPV6)
9864                 off = si->off;
9865                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9866                 *insn++ = BPF_LDX_MEM(
9867                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9868                         bpf_target_off(struct sock_common,
9869                                        skc_v6_daddr.s6_addr32[0],
9870                                        sizeof_field(struct sock_common,
9871                                                     skc_v6_daddr.s6_addr32[0]),
9872                                        target_size) + off);
9873 #else
9874                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9875                 *target_size = 4;
9876 #endif
9877                 break;
9878
9879         case offsetof(struct bpf_sock, src_port):
9880                 *insn++ = BPF_LDX_MEM(
9881                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9882                         si->dst_reg, si->src_reg,
9883                         bpf_target_off(struct sock_common, skc_num,
9884                                        sizeof_field(struct sock_common,
9885                                                     skc_num),
9886                                        target_size));
9887                 break;
9888
9889         case offsetof(struct bpf_sock, dst_port):
9890                 *insn++ = BPF_LDX_MEM(
9891                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9892                         si->dst_reg, si->src_reg,
9893                         bpf_target_off(struct sock_common, skc_dport,
9894                                        sizeof_field(struct sock_common,
9895                                                     skc_dport),
9896                                        target_size));
9897                 break;
9898
9899         case offsetof(struct bpf_sock, state):
9900                 *insn++ = BPF_LDX_MEM(
9901                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9902                         si->dst_reg, si->src_reg,
9903                         bpf_target_off(struct sock_common, skc_state,
9904                                        sizeof_field(struct sock_common,
9905                                                     skc_state),
9906                                        target_size));
9907                 break;
9908         case offsetof(struct bpf_sock, rx_queue_mapping):
9909 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9910                 *insn++ = BPF_LDX_MEM(
9911                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9912                         si->dst_reg, si->src_reg,
9913                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9914                                        sizeof_field(struct sock,
9915                                                     sk_rx_queue_mapping),
9916                                        target_size));
9917                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9918                                       1);
9919                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9920 #else
9921                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9922                 *target_size = 2;
9923 #endif
9924                 break;
9925         }
9926
9927         return insn - insn_buf;
9928 }
9929
9930 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9931                                          const struct bpf_insn *si,
9932                                          struct bpf_insn *insn_buf,
9933                                          struct bpf_prog *prog, u32 *target_size)
9934 {
9935         struct bpf_insn *insn = insn_buf;
9936
9937         switch (si->off) {
9938         case offsetof(struct __sk_buff, ifindex):
9939                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9940                                       si->dst_reg, si->src_reg,
9941                                       offsetof(struct sk_buff, dev));
9942                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9943                                       bpf_target_off(struct net_device, ifindex, 4,
9944                                                      target_size));
9945                 break;
9946         default:
9947                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9948                                               target_size);
9949         }
9950
9951         return insn - insn_buf;
9952 }
9953
9954 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9955                                   const struct bpf_insn *si,
9956                                   struct bpf_insn *insn_buf,
9957                                   struct bpf_prog *prog, u32 *target_size)
9958 {
9959         struct bpf_insn *insn = insn_buf;
9960
9961         switch (si->off) {
9962         case offsetof(struct xdp_md, data):
9963                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9964                                       si->dst_reg, si->src_reg,
9965                                       offsetof(struct xdp_buff, data));
9966                 break;
9967         case offsetof(struct xdp_md, data_meta):
9968                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9969                                       si->dst_reg, si->src_reg,
9970                                       offsetof(struct xdp_buff, data_meta));
9971                 break;
9972         case offsetof(struct xdp_md, data_end):
9973                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9974                                       si->dst_reg, si->src_reg,
9975                                       offsetof(struct xdp_buff, data_end));
9976                 break;
9977         case offsetof(struct xdp_md, ingress_ifindex):
9978                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9979                                       si->dst_reg, si->src_reg,
9980                                       offsetof(struct xdp_buff, rxq));
9981                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9982                                       si->dst_reg, si->dst_reg,
9983                                       offsetof(struct xdp_rxq_info, dev));
9984                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9985                                       offsetof(struct net_device, ifindex));
9986                 break;
9987         case offsetof(struct xdp_md, rx_queue_index):
9988                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9989                                       si->dst_reg, si->src_reg,
9990                                       offsetof(struct xdp_buff, rxq));
9991                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9992                                       offsetof(struct xdp_rxq_info,
9993                                                queue_index));
9994                 break;
9995         case offsetof(struct xdp_md, egress_ifindex):
9996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9997                                       si->dst_reg, si->src_reg,
9998                                       offsetof(struct xdp_buff, txq));
9999                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
10000                                       si->dst_reg, si->dst_reg,
10001                                       offsetof(struct xdp_txq_info, dev));
10002                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10003                                       offsetof(struct net_device, ifindex));
10004                 break;
10005         }
10006
10007         return insn - insn_buf;
10008 }
10009
10010 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10011  * context Structure, F is Field in context structure that contains a pointer
10012  * to Nested Structure of type NS that has the field NF.
10013  *
10014  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10015  * sure that SIZE is not greater than actual size of S.F.NF.
10016  *
10017  * If offset OFF is provided, the load happens from that offset relative to
10018  * offset of NF.
10019  */
10020 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
10021         do {                                                                   \
10022                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
10023                                       si->src_reg, offsetof(S, F));            \
10024                 *insn++ = BPF_LDX_MEM(                                         \
10025                         SIZE, si->dst_reg, si->dst_reg,                        \
10026                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10027                                        target_size)                            \
10028                                 + OFF);                                        \
10029         } while (0)
10030
10031 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
10032         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
10033                                              BPF_FIELD_SIZEOF(NS, NF), 0)
10034
10035 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10036  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10037  *
10038  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10039  * "register" since two registers available in convert_ctx_access are not
10040  * enough: we can't override neither SRC, since it contains value to store, nor
10041  * DST since it contains pointer to context that may be used by later
10042  * instructions. But we need a temporary place to save pointer to nested
10043  * structure whose field we want to store to.
10044  */
10045 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
10046         do {                                                                   \
10047                 int tmp_reg = BPF_REG_9;                                       \
10048                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10049                         --tmp_reg;                                             \
10050                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10051                         --tmp_reg;                                             \
10052                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
10053                                       offsetof(S, TF));                        \
10054                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
10055                                       si->dst_reg, offsetof(S, F));            \
10056                 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code),   \
10057                                        tmp_reg, si->src_reg,                   \
10058                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10059                                        target_size)                            \
10060                                        + OFF,                                  \
10061                                        si->imm);                               \
10062                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
10063                                       offsetof(S, TF));                        \
10064         } while (0)
10065
10066 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10067                                                       TF)                      \
10068         do {                                                                   \
10069                 if (type == BPF_WRITE) {                                       \
10070                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
10071                                                          OFF, TF);             \
10072                 } else {                                                       \
10073                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
10074                                 S, NS, F, NF, SIZE, OFF);  \
10075                 }                                                              \
10076         } while (0)
10077
10078 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
10079         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
10080                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10081
10082 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10083                                         const struct bpf_insn *si,
10084                                         struct bpf_insn *insn_buf,
10085                                         struct bpf_prog *prog, u32 *target_size)
10086 {
10087         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10088         struct bpf_insn *insn = insn_buf;
10089
10090         switch (si->off) {
10091         case offsetof(struct bpf_sock_addr, user_family):
10092                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10093                                             struct sockaddr, uaddr, sa_family);
10094                 break;
10095
10096         case offsetof(struct bpf_sock_addr, user_ip4):
10097                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10098                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10099                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10100                 break;
10101
10102         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10103                 off = si->off;
10104                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10105                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10106                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10107                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10108                         tmp_reg);
10109                 break;
10110
10111         case offsetof(struct bpf_sock_addr, user_port):
10112                 /* To get port we need to know sa_family first and then treat
10113                  * sockaddr as either sockaddr_in or sockaddr_in6.
10114                  * Though we can simplify since port field has same offset and
10115                  * size in both structures.
10116                  * Here we check this invariant and use just one of the
10117                  * structures if it's true.
10118                  */
10119                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10120                              offsetof(struct sockaddr_in6, sin6_port));
10121                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10122                              sizeof_field(struct sockaddr_in6, sin6_port));
10123                 /* Account for sin6_port being smaller than user_port. */
10124                 port_size = min(port_size, BPF_LDST_BYTES(si));
10125                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10126                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10127                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10128                 break;
10129
10130         case offsetof(struct bpf_sock_addr, family):
10131                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10132                                             struct sock, sk, sk_family);
10133                 break;
10134
10135         case offsetof(struct bpf_sock_addr, type):
10136                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10137                                             struct sock, sk, sk_type);
10138                 break;
10139
10140         case offsetof(struct bpf_sock_addr, protocol):
10141                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10142                                             struct sock, sk, sk_protocol);
10143                 break;
10144
10145         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10146                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10147                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10148                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10149                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10150                 break;
10151
10152         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10153                                 msg_src_ip6[3]):
10154                 off = si->off;
10155                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10156                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10157                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10158                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10159                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10160                 break;
10161         case offsetof(struct bpf_sock_addr, sk):
10162                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10163                                       si->dst_reg, si->src_reg,
10164                                       offsetof(struct bpf_sock_addr_kern, sk));
10165                 break;
10166         }
10167
10168         return insn - insn_buf;
10169 }
10170
10171 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10172                                        const struct bpf_insn *si,
10173                                        struct bpf_insn *insn_buf,
10174                                        struct bpf_prog *prog,
10175                                        u32 *target_size)
10176 {
10177         struct bpf_insn *insn = insn_buf;
10178         int off;
10179
10180 /* Helper macro for adding read access to tcp_sock or sock fields. */
10181 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10182         do {                                                                  \
10183                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10184                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10185                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10186                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10187                         reg--;                                                \
10188                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10189                         reg--;                                                \
10190                 if (si->dst_reg == si->src_reg) {                             \
10191                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10192                                           offsetof(struct bpf_sock_ops_kern,  \
10193                                           temp));                             \
10194                         fullsock_reg = reg;                                   \
10195                         jmp += 2;                                             \
10196                 }                                                             \
10197                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10198                                                 struct bpf_sock_ops_kern,     \
10199                                                 is_fullsock),                 \
10200                                       fullsock_reg, si->src_reg,              \
10201                                       offsetof(struct bpf_sock_ops_kern,      \
10202                                                is_fullsock));                 \
10203                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10204                 if (si->dst_reg == si->src_reg)                               \
10205                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10206                                       offsetof(struct bpf_sock_ops_kern,      \
10207                                       temp));                                 \
10208                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10209                                                 struct bpf_sock_ops_kern, sk),\
10210                                       si->dst_reg, si->src_reg,               \
10211                                       offsetof(struct bpf_sock_ops_kern, sk));\
10212                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10213                                                        OBJ_FIELD),            \
10214                                       si->dst_reg, si->dst_reg,               \
10215                                       offsetof(OBJ, OBJ_FIELD));              \
10216                 if (si->dst_reg == si->src_reg) {                             \
10217                         *insn++ = BPF_JMP_A(1);                               \
10218                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10219                                       offsetof(struct bpf_sock_ops_kern,      \
10220                                       temp));                                 \
10221                 }                                                             \
10222         } while (0)
10223
10224 #define SOCK_OPS_GET_SK()                                                             \
10225         do {                                                                  \
10226                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10227                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10228                         reg--;                                                \
10229                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10230                         reg--;                                                \
10231                 if (si->dst_reg == si->src_reg) {                             \
10232                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10233                                           offsetof(struct bpf_sock_ops_kern,  \
10234                                           temp));                             \
10235                         fullsock_reg = reg;                                   \
10236                         jmp += 2;                                             \
10237                 }                                                             \
10238                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10239                                                 struct bpf_sock_ops_kern,     \
10240                                                 is_fullsock),                 \
10241                                       fullsock_reg, si->src_reg,              \
10242                                       offsetof(struct bpf_sock_ops_kern,      \
10243                                                is_fullsock));                 \
10244                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10245                 if (si->dst_reg == si->src_reg)                               \
10246                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10247                                       offsetof(struct bpf_sock_ops_kern,      \
10248                                       temp));                                 \
10249                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10250                                                 struct bpf_sock_ops_kern, sk),\
10251                                       si->dst_reg, si->src_reg,               \
10252                                       offsetof(struct bpf_sock_ops_kern, sk));\
10253                 if (si->dst_reg == si->src_reg) {                             \
10254                         *insn++ = BPF_JMP_A(1);                               \
10255                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10256                                       offsetof(struct bpf_sock_ops_kern,      \
10257                                       temp));                                 \
10258                 }                                                             \
10259         } while (0)
10260
10261 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10262                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10263
10264 /* Helper macro for adding write access to tcp_sock or sock fields.
10265  * The macro is called with two registers, dst_reg which contains a pointer
10266  * to ctx (context) and src_reg which contains the value that should be
10267  * stored. However, we need an additional register since we cannot overwrite
10268  * dst_reg because it may be used later in the program.
10269  * Instead we "borrow" one of the other register. We first save its value
10270  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10271  * it at the end of the macro.
10272  */
10273 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10274         do {                                                                  \
10275                 int reg = BPF_REG_9;                                          \
10276                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10277                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10278                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10279                         reg--;                                                \
10280                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10281                         reg--;                                                \
10282                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10283                                       offsetof(struct bpf_sock_ops_kern,      \
10284                                                temp));                        \
10285                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10286                                                 struct bpf_sock_ops_kern,     \
10287                                                 is_fullsock),                 \
10288                                       reg, si->dst_reg,                       \
10289                                       offsetof(struct bpf_sock_ops_kern,      \
10290                                                is_fullsock));                 \
10291                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10292                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10293                                                 struct bpf_sock_ops_kern, sk),\
10294                                       reg, si->dst_reg,                       \
10295                                       offsetof(struct bpf_sock_ops_kern, sk));\
10296                 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) |     \
10297                                        BPF_MEM | BPF_CLASS(si->code),         \
10298                                        reg, si->src_reg,                      \
10299                                        offsetof(OBJ, OBJ_FIELD),              \
10300                                        si->imm);                              \
10301                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10302                                       offsetof(struct bpf_sock_ops_kern,      \
10303                                                temp));                        \
10304         } while (0)
10305
10306 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10307         do {                                                                  \
10308                 if (TYPE == BPF_WRITE)                                        \
10309                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10310                 else                                                          \
10311                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10312         } while (0)
10313
10314         switch (si->off) {
10315         case offsetof(struct bpf_sock_ops, op):
10316                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10317                                                        op),
10318                                       si->dst_reg, si->src_reg,
10319                                       offsetof(struct bpf_sock_ops_kern, op));
10320                 break;
10321
10322         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10323              offsetof(struct bpf_sock_ops, replylong[3]):
10324                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10325                              sizeof_field(struct bpf_sock_ops_kern, reply));
10326                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10327                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10328                 off = si->off;
10329                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10330                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10331                 if (type == BPF_WRITE)
10332                         *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10333                 else
10334                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10335                                               off);
10336                 break;
10337
10338         case offsetof(struct bpf_sock_ops, family):
10339                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10340
10341                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10342                                               struct bpf_sock_ops_kern, sk),
10343                                       si->dst_reg, si->src_reg,
10344                                       offsetof(struct bpf_sock_ops_kern, sk));
10345                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10346                                       offsetof(struct sock_common, skc_family));
10347                 break;
10348
10349         case offsetof(struct bpf_sock_ops, remote_ip4):
10350                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10351
10352                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10353                                                 struct bpf_sock_ops_kern, sk),
10354                                       si->dst_reg, si->src_reg,
10355                                       offsetof(struct bpf_sock_ops_kern, sk));
10356                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10357                                       offsetof(struct sock_common, skc_daddr));
10358                 break;
10359
10360         case offsetof(struct bpf_sock_ops, local_ip4):
10361                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10362                                           skc_rcv_saddr) != 4);
10363
10364                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10365                                               struct bpf_sock_ops_kern, sk),
10366                                       si->dst_reg, si->src_reg,
10367                                       offsetof(struct bpf_sock_ops_kern, sk));
10368                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10369                                       offsetof(struct sock_common,
10370                                                skc_rcv_saddr));
10371                 break;
10372
10373         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10374              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10375 #if IS_ENABLED(CONFIG_IPV6)
10376                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10377                                           skc_v6_daddr.s6_addr32[0]) != 4);
10378
10379                 off = si->off;
10380                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10381                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10382                                                 struct bpf_sock_ops_kern, sk),
10383                                       si->dst_reg, si->src_reg,
10384                                       offsetof(struct bpf_sock_ops_kern, sk));
10385                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10386                                       offsetof(struct sock_common,
10387                                                skc_v6_daddr.s6_addr32[0]) +
10388                                       off);
10389 #else
10390                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10391 #endif
10392                 break;
10393
10394         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10395              offsetof(struct bpf_sock_ops, local_ip6[3]):
10396 #if IS_ENABLED(CONFIG_IPV6)
10397                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10398                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10399
10400                 off = si->off;
10401                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10402                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10403                                                 struct bpf_sock_ops_kern, sk),
10404                                       si->dst_reg, si->src_reg,
10405                                       offsetof(struct bpf_sock_ops_kern, sk));
10406                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10407                                       offsetof(struct sock_common,
10408                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10409                                       off);
10410 #else
10411                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10412 #endif
10413                 break;
10414
10415         case offsetof(struct bpf_sock_ops, remote_port):
10416                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10417
10418                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10419                                                 struct bpf_sock_ops_kern, sk),
10420                                       si->dst_reg, si->src_reg,
10421                                       offsetof(struct bpf_sock_ops_kern, sk));
10422                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10423                                       offsetof(struct sock_common, skc_dport));
10424 #ifndef __BIG_ENDIAN_BITFIELD
10425                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10426 #endif
10427                 break;
10428
10429         case offsetof(struct bpf_sock_ops, local_port):
10430                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10431
10432                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10433                                                 struct bpf_sock_ops_kern, sk),
10434                                       si->dst_reg, si->src_reg,
10435                                       offsetof(struct bpf_sock_ops_kern, sk));
10436                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10437                                       offsetof(struct sock_common, skc_num));
10438                 break;
10439
10440         case offsetof(struct bpf_sock_ops, is_fullsock):
10441                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10442                                                 struct bpf_sock_ops_kern,
10443                                                 is_fullsock),
10444                                       si->dst_reg, si->src_reg,
10445                                       offsetof(struct bpf_sock_ops_kern,
10446                                                is_fullsock));
10447                 break;
10448
10449         case offsetof(struct bpf_sock_ops, state):
10450                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10451
10452                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10453                                                 struct bpf_sock_ops_kern, sk),
10454                                       si->dst_reg, si->src_reg,
10455                                       offsetof(struct bpf_sock_ops_kern, sk));
10456                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10457                                       offsetof(struct sock_common, skc_state));
10458                 break;
10459
10460         case offsetof(struct bpf_sock_ops, rtt_min):
10461                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10462                              sizeof(struct minmax));
10463                 BUILD_BUG_ON(sizeof(struct minmax) <
10464                              sizeof(struct minmax_sample));
10465
10466                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10467                                                 struct bpf_sock_ops_kern, sk),
10468                                       si->dst_reg, si->src_reg,
10469                                       offsetof(struct bpf_sock_ops_kern, sk));
10470                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10471                                       offsetof(struct tcp_sock, rtt_min) +
10472                                       sizeof_field(struct minmax_sample, t));
10473                 break;
10474
10475         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10476                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10477                                    struct tcp_sock);
10478                 break;
10479
10480         case offsetof(struct bpf_sock_ops, sk_txhash):
10481                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10482                                           struct sock, type);
10483                 break;
10484         case offsetof(struct bpf_sock_ops, snd_cwnd):
10485                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10486                 break;
10487         case offsetof(struct bpf_sock_ops, srtt_us):
10488                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10489                 break;
10490         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10491                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10492                 break;
10493         case offsetof(struct bpf_sock_ops, rcv_nxt):
10494                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10495                 break;
10496         case offsetof(struct bpf_sock_ops, snd_nxt):
10497                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10498                 break;
10499         case offsetof(struct bpf_sock_ops, snd_una):
10500                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10501                 break;
10502         case offsetof(struct bpf_sock_ops, mss_cache):
10503                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10504                 break;
10505         case offsetof(struct bpf_sock_ops, ecn_flags):
10506                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10507                 break;
10508         case offsetof(struct bpf_sock_ops, rate_delivered):
10509                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10510                 break;
10511         case offsetof(struct bpf_sock_ops, rate_interval_us):
10512                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10513                 break;
10514         case offsetof(struct bpf_sock_ops, packets_out):
10515                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10516                 break;
10517         case offsetof(struct bpf_sock_ops, retrans_out):
10518                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10519                 break;
10520         case offsetof(struct bpf_sock_ops, total_retrans):
10521                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10522                 break;
10523         case offsetof(struct bpf_sock_ops, segs_in):
10524                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10525                 break;
10526         case offsetof(struct bpf_sock_ops, data_segs_in):
10527                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10528                 break;
10529         case offsetof(struct bpf_sock_ops, segs_out):
10530                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10531                 break;
10532         case offsetof(struct bpf_sock_ops, data_segs_out):
10533                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10534                 break;
10535         case offsetof(struct bpf_sock_ops, lost_out):
10536                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10537                 break;
10538         case offsetof(struct bpf_sock_ops, sacked_out):
10539                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10540                 break;
10541         case offsetof(struct bpf_sock_ops, bytes_received):
10542                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10543                 break;
10544         case offsetof(struct bpf_sock_ops, bytes_acked):
10545                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10546                 break;
10547         case offsetof(struct bpf_sock_ops, sk):
10548                 SOCK_OPS_GET_SK();
10549                 break;
10550         case offsetof(struct bpf_sock_ops, skb_data_end):
10551                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10552                                                        skb_data_end),
10553                                       si->dst_reg, si->src_reg,
10554                                       offsetof(struct bpf_sock_ops_kern,
10555                                                skb_data_end));
10556                 break;
10557         case offsetof(struct bpf_sock_ops, skb_data):
10558                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10559                                                        skb),
10560                                       si->dst_reg, si->src_reg,
10561                                       offsetof(struct bpf_sock_ops_kern,
10562                                                skb));
10563                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10564                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10565                                       si->dst_reg, si->dst_reg,
10566                                       offsetof(struct sk_buff, data));
10567                 break;
10568         case offsetof(struct bpf_sock_ops, skb_len):
10569                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10570                                                        skb),
10571                                       si->dst_reg, si->src_reg,
10572                                       offsetof(struct bpf_sock_ops_kern,
10573                                                skb));
10574                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10575                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10576                                       si->dst_reg, si->dst_reg,
10577                                       offsetof(struct sk_buff, len));
10578                 break;
10579         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10580                 off = offsetof(struct sk_buff, cb);
10581                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10582                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10583                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10584                                                        skb),
10585                                       si->dst_reg, si->src_reg,
10586                                       offsetof(struct bpf_sock_ops_kern,
10587                                                skb));
10588                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10589                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10590                                                        tcp_flags),
10591                                       si->dst_reg, si->dst_reg, off);
10592                 break;
10593         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10594                 struct bpf_insn *jmp_on_null_skb;
10595
10596                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10597                                                        skb),
10598                                       si->dst_reg, si->src_reg,
10599                                       offsetof(struct bpf_sock_ops_kern,
10600                                                skb));
10601                 /* Reserve one insn to test skb == NULL */
10602                 jmp_on_null_skb = insn++;
10603                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10604                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10605                                       bpf_target_off(struct skb_shared_info,
10606                                                      hwtstamps, 8,
10607                                                      target_size));
10608                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10609                                                insn - jmp_on_null_skb - 1);
10610                 break;
10611         }
10612         }
10613         return insn - insn_buf;
10614 }
10615
10616 /* data_end = skb->data + skb_headlen() */
10617 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10618                                                     struct bpf_insn *insn)
10619 {
10620         int reg;
10621         int temp_reg_off = offsetof(struct sk_buff, cb) +
10622                            offsetof(struct sk_skb_cb, temp_reg);
10623
10624         if (si->src_reg == si->dst_reg) {
10625                 /* We need an extra register, choose and save a register. */
10626                 reg = BPF_REG_9;
10627                 if (si->src_reg == reg || si->dst_reg == reg)
10628                         reg--;
10629                 if (si->src_reg == reg || si->dst_reg == reg)
10630                         reg--;
10631                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10632         } else {
10633                 reg = si->dst_reg;
10634         }
10635
10636         /* reg = skb->data */
10637         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10638                               reg, si->src_reg,
10639                               offsetof(struct sk_buff, data));
10640         /* AX = skb->len */
10641         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10642                               BPF_REG_AX, si->src_reg,
10643                               offsetof(struct sk_buff, len));
10644         /* reg = skb->data + skb->len */
10645         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10646         /* AX = skb->data_len */
10647         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10648                               BPF_REG_AX, si->src_reg,
10649                               offsetof(struct sk_buff, data_len));
10650
10651         /* reg = skb->data + skb->len - skb->data_len */
10652         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10653
10654         if (si->src_reg == si->dst_reg) {
10655                 /* Restore the saved register */
10656                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10657                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10658                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10659         }
10660
10661         return insn;
10662 }
10663
10664 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10665                                      const struct bpf_insn *si,
10666                                      struct bpf_insn *insn_buf,
10667                                      struct bpf_prog *prog, u32 *target_size)
10668 {
10669         struct bpf_insn *insn = insn_buf;
10670         int off;
10671
10672         switch (si->off) {
10673         case offsetof(struct __sk_buff, data_end):
10674                 insn = bpf_convert_data_end_access(si, insn);
10675                 break;
10676         case offsetof(struct __sk_buff, cb[0]) ...
10677              offsetofend(struct __sk_buff, cb[4]) - 1:
10678                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10679                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10680                               offsetof(struct sk_skb_cb, data)) %
10681                              sizeof(__u64));
10682
10683                 prog->cb_access = 1;
10684                 off  = si->off;
10685                 off -= offsetof(struct __sk_buff, cb[0]);
10686                 off += offsetof(struct sk_buff, cb);
10687                 off += offsetof(struct sk_skb_cb, data);
10688                 if (type == BPF_WRITE)
10689                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10690                 else
10691                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10692                                               si->src_reg, off);
10693                 break;
10694
10695
10696         default:
10697                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10698                                               target_size);
10699         }
10700
10701         return insn - insn_buf;
10702 }
10703
10704 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10705                                      const struct bpf_insn *si,
10706                                      struct bpf_insn *insn_buf,
10707                                      struct bpf_prog *prog, u32 *target_size)
10708 {
10709         struct bpf_insn *insn = insn_buf;
10710 #if IS_ENABLED(CONFIG_IPV6)
10711         int off;
10712 #endif
10713
10714         /* convert ctx uses the fact sg element is first in struct */
10715         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10716
10717         switch (si->off) {
10718         case offsetof(struct sk_msg_md, data):
10719                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10720                                       si->dst_reg, si->src_reg,
10721                                       offsetof(struct sk_msg, data));
10722                 break;
10723         case offsetof(struct sk_msg_md, data_end):
10724                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10725                                       si->dst_reg, si->src_reg,
10726                                       offsetof(struct sk_msg, data_end));
10727                 break;
10728         case offsetof(struct sk_msg_md, family):
10729                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10730
10731                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10732                                               struct sk_msg, sk),
10733                                       si->dst_reg, si->src_reg,
10734                                       offsetof(struct sk_msg, sk));
10735                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10736                                       offsetof(struct sock_common, skc_family));
10737                 break;
10738
10739         case offsetof(struct sk_msg_md, remote_ip4):
10740                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10741
10742                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10743                                                 struct sk_msg, sk),
10744                                       si->dst_reg, si->src_reg,
10745                                       offsetof(struct sk_msg, sk));
10746                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10747                                       offsetof(struct sock_common, skc_daddr));
10748                 break;
10749
10750         case offsetof(struct sk_msg_md, local_ip4):
10751                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10752                                           skc_rcv_saddr) != 4);
10753
10754                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10755                                               struct sk_msg, sk),
10756                                       si->dst_reg, si->src_reg,
10757                                       offsetof(struct sk_msg, sk));
10758                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10759                                       offsetof(struct sock_common,
10760                                                skc_rcv_saddr));
10761                 break;
10762
10763         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10764              offsetof(struct sk_msg_md, remote_ip6[3]):
10765 #if IS_ENABLED(CONFIG_IPV6)
10766                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10767                                           skc_v6_daddr.s6_addr32[0]) != 4);
10768
10769                 off = si->off;
10770                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10771                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10772                                                 struct sk_msg, sk),
10773                                       si->dst_reg, si->src_reg,
10774                                       offsetof(struct sk_msg, sk));
10775                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10776                                       offsetof(struct sock_common,
10777                                                skc_v6_daddr.s6_addr32[0]) +
10778                                       off);
10779 #else
10780                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10781 #endif
10782                 break;
10783
10784         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10785              offsetof(struct sk_msg_md, local_ip6[3]):
10786 #if IS_ENABLED(CONFIG_IPV6)
10787                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10788                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10789
10790                 off = si->off;
10791                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10792                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10793                                                 struct sk_msg, sk),
10794                                       si->dst_reg, si->src_reg,
10795                                       offsetof(struct sk_msg, sk));
10796                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10797                                       offsetof(struct sock_common,
10798                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10799                                       off);
10800 #else
10801                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10802 #endif
10803                 break;
10804
10805         case offsetof(struct sk_msg_md, remote_port):
10806                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10807
10808                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10809                                                 struct sk_msg, sk),
10810                                       si->dst_reg, si->src_reg,
10811                                       offsetof(struct sk_msg, sk));
10812                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10813                                       offsetof(struct sock_common, skc_dport));
10814 #ifndef __BIG_ENDIAN_BITFIELD
10815                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10816 #endif
10817                 break;
10818
10819         case offsetof(struct sk_msg_md, local_port):
10820                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10821
10822                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10823                                                 struct sk_msg, sk),
10824                                       si->dst_reg, si->src_reg,
10825                                       offsetof(struct sk_msg, sk));
10826                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10827                                       offsetof(struct sock_common, skc_num));
10828                 break;
10829
10830         case offsetof(struct sk_msg_md, size):
10831                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10832                                       si->dst_reg, si->src_reg,
10833                                       offsetof(struct sk_msg_sg, size));
10834                 break;
10835
10836         case offsetof(struct sk_msg_md, sk):
10837                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10838                                       si->dst_reg, si->src_reg,
10839                                       offsetof(struct sk_msg, sk));
10840                 break;
10841         }
10842
10843         return insn - insn_buf;
10844 }
10845
10846 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10847         .get_func_proto         = sk_filter_func_proto,
10848         .is_valid_access        = sk_filter_is_valid_access,
10849         .convert_ctx_access     = bpf_convert_ctx_access,
10850         .gen_ld_abs             = bpf_gen_ld_abs,
10851 };
10852
10853 const struct bpf_prog_ops sk_filter_prog_ops = {
10854         .test_run               = bpf_prog_test_run_skb,
10855 };
10856
10857 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10858         .get_func_proto         = tc_cls_act_func_proto,
10859         .is_valid_access        = tc_cls_act_is_valid_access,
10860         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10861         .gen_prologue           = tc_cls_act_prologue,
10862         .gen_ld_abs             = bpf_gen_ld_abs,
10863         .btf_struct_access      = tc_cls_act_btf_struct_access,
10864 };
10865
10866 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10867         .test_run               = bpf_prog_test_run_skb,
10868 };
10869
10870 const struct bpf_verifier_ops xdp_verifier_ops = {
10871         .get_func_proto         = xdp_func_proto,
10872         .is_valid_access        = xdp_is_valid_access,
10873         .convert_ctx_access     = xdp_convert_ctx_access,
10874         .gen_prologue           = bpf_noop_prologue,
10875         .btf_struct_access      = xdp_btf_struct_access,
10876 };
10877
10878 const struct bpf_prog_ops xdp_prog_ops = {
10879         .test_run               = bpf_prog_test_run_xdp,
10880 };
10881
10882 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10883         .get_func_proto         = cg_skb_func_proto,
10884         .is_valid_access        = cg_skb_is_valid_access,
10885         .convert_ctx_access     = bpf_convert_ctx_access,
10886 };
10887
10888 const struct bpf_prog_ops cg_skb_prog_ops = {
10889         .test_run               = bpf_prog_test_run_skb,
10890 };
10891
10892 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10893         .get_func_proto         = lwt_in_func_proto,
10894         .is_valid_access        = lwt_is_valid_access,
10895         .convert_ctx_access     = bpf_convert_ctx_access,
10896 };
10897
10898 const struct bpf_prog_ops lwt_in_prog_ops = {
10899         .test_run               = bpf_prog_test_run_skb,
10900 };
10901
10902 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10903         .get_func_proto         = lwt_out_func_proto,
10904         .is_valid_access        = lwt_is_valid_access,
10905         .convert_ctx_access     = bpf_convert_ctx_access,
10906 };
10907
10908 const struct bpf_prog_ops lwt_out_prog_ops = {
10909         .test_run               = bpf_prog_test_run_skb,
10910 };
10911
10912 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10913         .get_func_proto         = lwt_xmit_func_proto,
10914         .is_valid_access        = lwt_is_valid_access,
10915         .convert_ctx_access     = bpf_convert_ctx_access,
10916         .gen_prologue           = tc_cls_act_prologue,
10917 };
10918
10919 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10920         .test_run               = bpf_prog_test_run_skb,
10921 };
10922
10923 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10924         .get_func_proto         = lwt_seg6local_func_proto,
10925         .is_valid_access        = lwt_is_valid_access,
10926         .convert_ctx_access     = bpf_convert_ctx_access,
10927 };
10928
10929 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10930         .test_run               = bpf_prog_test_run_skb,
10931 };
10932
10933 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10934         .get_func_proto         = sock_filter_func_proto,
10935         .is_valid_access        = sock_filter_is_valid_access,
10936         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10937 };
10938
10939 const struct bpf_prog_ops cg_sock_prog_ops = {
10940 };
10941
10942 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10943         .get_func_proto         = sock_addr_func_proto,
10944         .is_valid_access        = sock_addr_is_valid_access,
10945         .convert_ctx_access     = sock_addr_convert_ctx_access,
10946 };
10947
10948 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10949 };
10950
10951 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10952         .get_func_proto         = sock_ops_func_proto,
10953         .is_valid_access        = sock_ops_is_valid_access,
10954         .convert_ctx_access     = sock_ops_convert_ctx_access,
10955 };
10956
10957 const struct bpf_prog_ops sock_ops_prog_ops = {
10958 };
10959
10960 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10961         .get_func_proto         = sk_skb_func_proto,
10962         .is_valid_access        = sk_skb_is_valid_access,
10963         .convert_ctx_access     = sk_skb_convert_ctx_access,
10964         .gen_prologue           = sk_skb_prologue,
10965 };
10966
10967 const struct bpf_prog_ops sk_skb_prog_ops = {
10968 };
10969
10970 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10971         .get_func_proto         = sk_msg_func_proto,
10972         .is_valid_access        = sk_msg_is_valid_access,
10973         .convert_ctx_access     = sk_msg_convert_ctx_access,
10974         .gen_prologue           = bpf_noop_prologue,
10975 };
10976
10977 const struct bpf_prog_ops sk_msg_prog_ops = {
10978 };
10979
10980 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10981         .get_func_proto         = flow_dissector_func_proto,
10982         .is_valid_access        = flow_dissector_is_valid_access,
10983         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10984 };
10985
10986 const struct bpf_prog_ops flow_dissector_prog_ops = {
10987         .test_run               = bpf_prog_test_run_flow_dissector,
10988 };
10989
10990 int sk_detach_filter(struct sock *sk)
10991 {
10992         int ret = -ENOENT;
10993         struct sk_filter *filter;
10994
10995         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10996                 return -EPERM;
10997
10998         filter = rcu_dereference_protected(sk->sk_filter,
10999                                            lockdep_sock_is_held(sk));
11000         if (filter) {
11001                 RCU_INIT_POINTER(sk->sk_filter, NULL);
11002                 sk_filter_uncharge(sk, filter);
11003                 ret = 0;
11004         }
11005
11006         return ret;
11007 }
11008 EXPORT_SYMBOL_GPL(sk_detach_filter);
11009
11010 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11011 {
11012         struct sock_fprog_kern *fprog;
11013         struct sk_filter *filter;
11014         int ret = 0;
11015
11016         sockopt_lock_sock(sk);
11017         filter = rcu_dereference_protected(sk->sk_filter,
11018                                            lockdep_sock_is_held(sk));
11019         if (!filter)
11020                 goto out;
11021
11022         /* We're copying the filter that has been originally attached,
11023          * so no conversion/decode needed anymore. eBPF programs that
11024          * have no original program cannot be dumped through this.
11025          */
11026         ret = -EACCES;
11027         fprog = filter->prog->orig_prog;
11028         if (!fprog)
11029                 goto out;
11030
11031         ret = fprog->len;
11032         if (!len)
11033                 /* User space only enquires number of filter blocks. */
11034                 goto out;
11035
11036         ret = -EINVAL;
11037         if (len < fprog->len)
11038                 goto out;
11039
11040         ret = -EFAULT;
11041         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11042                 goto out;
11043
11044         /* Instead of bytes, the API requests to return the number
11045          * of filter blocks.
11046          */
11047         ret = fprog->len;
11048 out:
11049         sockopt_release_sock(sk);
11050         return ret;
11051 }
11052
11053 #ifdef CONFIG_INET
11054 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11055                                     struct sock_reuseport *reuse,
11056                                     struct sock *sk, struct sk_buff *skb,
11057                                     struct sock *migrating_sk,
11058                                     u32 hash)
11059 {
11060         reuse_kern->skb = skb;
11061         reuse_kern->sk = sk;
11062         reuse_kern->selected_sk = NULL;
11063         reuse_kern->migrating_sk = migrating_sk;
11064         reuse_kern->data_end = skb->data + skb_headlen(skb);
11065         reuse_kern->hash = hash;
11066         reuse_kern->reuseport_id = reuse->reuseport_id;
11067         reuse_kern->bind_inany = reuse->bind_inany;
11068 }
11069
11070 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11071                                   struct bpf_prog *prog, struct sk_buff *skb,
11072                                   struct sock *migrating_sk,
11073                                   u32 hash)
11074 {
11075         struct sk_reuseport_kern reuse_kern;
11076         enum sk_action action;
11077
11078         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11079         action = bpf_prog_run(prog, &reuse_kern);
11080
11081         if (action == SK_PASS)
11082                 return reuse_kern.selected_sk;
11083         else
11084                 return ERR_PTR(-ECONNREFUSED);
11085 }
11086
11087 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11088            struct bpf_map *, map, void *, key, u32, flags)
11089 {
11090         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11091         struct sock_reuseport *reuse;
11092         struct sock *selected_sk;
11093
11094         selected_sk = map->ops->map_lookup_elem(map, key);
11095         if (!selected_sk)
11096                 return -ENOENT;
11097
11098         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11099         if (!reuse) {
11100                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11101                 if (sk_is_refcounted(selected_sk))
11102                         sock_put(selected_sk);
11103
11104                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11105                  * The only (!reuse) case here is - the sk has already been
11106                  * unhashed (e.g. by close()), so treat it as -ENOENT.
11107                  *
11108                  * Other maps (e.g. sock_map) do not provide this guarantee and
11109                  * the sk may never be in the reuseport group to begin with.
11110                  */
11111                 return is_sockarray ? -ENOENT : -EINVAL;
11112         }
11113
11114         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11115                 struct sock *sk = reuse_kern->sk;
11116
11117                 if (sk->sk_protocol != selected_sk->sk_protocol)
11118                         return -EPROTOTYPE;
11119                 else if (sk->sk_family != selected_sk->sk_family)
11120                         return -EAFNOSUPPORT;
11121
11122                 /* Catch all. Likely bound to a different sockaddr. */
11123                 return -EBADFD;
11124         }
11125
11126         reuse_kern->selected_sk = selected_sk;
11127
11128         return 0;
11129 }
11130
11131 static const struct bpf_func_proto sk_select_reuseport_proto = {
11132         .func           = sk_select_reuseport,
11133         .gpl_only       = false,
11134         .ret_type       = RET_INTEGER,
11135         .arg1_type      = ARG_PTR_TO_CTX,
11136         .arg2_type      = ARG_CONST_MAP_PTR,
11137         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11138         .arg4_type      = ARG_ANYTHING,
11139 };
11140
11141 BPF_CALL_4(sk_reuseport_load_bytes,
11142            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11143            void *, to, u32, len)
11144 {
11145         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11146 }
11147
11148 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11149         .func           = sk_reuseport_load_bytes,
11150         .gpl_only       = false,
11151         .ret_type       = RET_INTEGER,
11152         .arg1_type      = ARG_PTR_TO_CTX,
11153         .arg2_type      = ARG_ANYTHING,
11154         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11155         .arg4_type      = ARG_CONST_SIZE,
11156 };
11157
11158 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11159            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11160            void *, to, u32, len, u32, start_header)
11161 {
11162         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11163                                                len, start_header);
11164 }
11165
11166 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11167         .func           = sk_reuseport_load_bytes_relative,
11168         .gpl_only       = false,
11169         .ret_type       = RET_INTEGER,
11170         .arg1_type      = ARG_PTR_TO_CTX,
11171         .arg2_type      = ARG_ANYTHING,
11172         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11173         .arg4_type      = ARG_CONST_SIZE,
11174         .arg5_type      = ARG_ANYTHING,
11175 };
11176
11177 static const struct bpf_func_proto *
11178 sk_reuseport_func_proto(enum bpf_func_id func_id,
11179                         const struct bpf_prog *prog)
11180 {
11181         switch (func_id) {
11182         case BPF_FUNC_sk_select_reuseport:
11183                 return &sk_select_reuseport_proto;
11184         case BPF_FUNC_skb_load_bytes:
11185                 return &sk_reuseport_load_bytes_proto;
11186         case BPF_FUNC_skb_load_bytes_relative:
11187                 return &sk_reuseport_load_bytes_relative_proto;
11188         case BPF_FUNC_get_socket_cookie:
11189                 return &bpf_get_socket_ptr_cookie_proto;
11190         case BPF_FUNC_ktime_get_coarse_ns:
11191                 return &bpf_ktime_get_coarse_ns_proto;
11192         default:
11193                 return bpf_base_func_proto(func_id);
11194         }
11195 }
11196
11197 static bool
11198 sk_reuseport_is_valid_access(int off, int size,
11199                              enum bpf_access_type type,
11200                              const struct bpf_prog *prog,
11201                              struct bpf_insn_access_aux *info)
11202 {
11203         const u32 size_default = sizeof(__u32);
11204
11205         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11206             off % size || type != BPF_READ)
11207                 return false;
11208
11209         switch (off) {
11210         case offsetof(struct sk_reuseport_md, data):
11211                 info->reg_type = PTR_TO_PACKET;
11212                 return size == sizeof(__u64);
11213
11214         case offsetof(struct sk_reuseport_md, data_end):
11215                 info->reg_type = PTR_TO_PACKET_END;
11216                 return size == sizeof(__u64);
11217
11218         case offsetof(struct sk_reuseport_md, hash):
11219                 return size == size_default;
11220
11221         case offsetof(struct sk_reuseport_md, sk):
11222                 info->reg_type = PTR_TO_SOCKET;
11223                 return size == sizeof(__u64);
11224
11225         case offsetof(struct sk_reuseport_md, migrating_sk):
11226                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11227                 return size == sizeof(__u64);
11228
11229         /* Fields that allow narrowing */
11230         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11231                 if (size < sizeof_field(struct sk_buff, protocol))
11232                         return false;
11233                 fallthrough;
11234         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11235         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11236         case bpf_ctx_range(struct sk_reuseport_md, len):
11237                 bpf_ctx_record_field_size(info, size_default);
11238                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11239
11240         default:
11241                 return false;
11242         }
11243 }
11244
11245 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11246         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11247                               si->dst_reg, si->src_reg,                 \
11248                               bpf_target_off(struct sk_reuseport_kern, F, \
11249                                              sizeof_field(struct sk_reuseport_kern, F), \
11250                                              target_size));             \
11251         })
11252
11253 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11254         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11255                                     struct sk_buff,                     \
11256                                     skb,                                \
11257                                     SKB_FIELD)
11258
11259 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11260         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11261                                     struct sock,                        \
11262                                     sk,                                 \
11263                                     SK_FIELD)
11264
11265 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11266                                            const struct bpf_insn *si,
11267                                            struct bpf_insn *insn_buf,
11268                                            struct bpf_prog *prog,
11269                                            u32 *target_size)
11270 {
11271         struct bpf_insn *insn = insn_buf;
11272
11273         switch (si->off) {
11274         case offsetof(struct sk_reuseport_md, data):
11275                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11276                 break;
11277
11278         case offsetof(struct sk_reuseport_md, len):
11279                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11280                 break;
11281
11282         case offsetof(struct sk_reuseport_md, eth_protocol):
11283                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11284                 break;
11285
11286         case offsetof(struct sk_reuseport_md, ip_protocol):
11287                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11288                 break;
11289
11290         case offsetof(struct sk_reuseport_md, data_end):
11291                 SK_REUSEPORT_LOAD_FIELD(data_end);
11292                 break;
11293
11294         case offsetof(struct sk_reuseport_md, hash):
11295                 SK_REUSEPORT_LOAD_FIELD(hash);
11296                 break;
11297
11298         case offsetof(struct sk_reuseport_md, bind_inany):
11299                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11300                 break;
11301
11302         case offsetof(struct sk_reuseport_md, sk):
11303                 SK_REUSEPORT_LOAD_FIELD(sk);
11304                 break;
11305
11306         case offsetof(struct sk_reuseport_md, migrating_sk):
11307                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11308                 break;
11309         }
11310
11311         return insn - insn_buf;
11312 }
11313
11314 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11315         .get_func_proto         = sk_reuseport_func_proto,
11316         .is_valid_access        = sk_reuseport_is_valid_access,
11317         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11318 };
11319
11320 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11321 };
11322
11323 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11324 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11325
11326 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11327            struct sock *, sk, u64, flags)
11328 {
11329         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11330                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11331                 return -EINVAL;
11332         if (unlikely(sk && sk_is_refcounted(sk)))
11333                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11334         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11335                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11336         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11337                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11338
11339         /* Check if socket is suitable for packet L3/L4 protocol */
11340         if (sk && sk->sk_protocol != ctx->protocol)
11341                 return -EPROTOTYPE;
11342         if (sk && sk->sk_family != ctx->family &&
11343             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11344                 return -EAFNOSUPPORT;
11345
11346         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11347                 return -EEXIST;
11348
11349         /* Select socket as lookup result */
11350         ctx->selected_sk = sk;
11351         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11352         return 0;
11353 }
11354
11355 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11356         .func           = bpf_sk_lookup_assign,
11357         .gpl_only       = false,
11358         .ret_type       = RET_INTEGER,
11359         .arg1_type      = ARG_PTR_TO_CTX,
11360         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11361         .arg3_type      = ARG_ANYTHING,
11362 };
11363
11364 static const struct bpf_func_proto *
11365 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11366 {
11367         switch (func_id) {
11368         case BPF_FUNC_perf_event_output:
11369                 return &bpf_event_output_data_proto;
11370         case BPF_FUNC_sk_assign:
11371                 return &bpf_sk_lookup_assign_proto;
11372         case BPF_FUNC_sk_release:
11373                 return &bpf_sk_release_proto;
11374         default:
11375                 return bpf_sk_base_func_proto(func_id);
11376         }
11377 }
11378
11379 static bool sk_lookup_is_valid_access(int off, int size,
11380                                       enum bpf_access_type type,
11381                                       const struct bpf_prog *prog,
11382                                       struct bpf_insn_access_aux *info)
11383 {
11384         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11385                 return false;
11386         if (off % size != 0)
11387                 return false;
11388         if (type != BPF_READ)
11389                 return false;
11390
11391         switch (off) {
11392         case offsetof(struct bpf_sk_lookup, sk):
11393                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11394                 return size == sizeof(__u64);
11395
11396         case bpf_ctx_range(struct bpf_sk_lookup, family):
11397         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11398         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11399         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11400         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11401         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11402         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11403         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11404                 bpf_ctx_record_field_size(info, sizeof(__u32));
11405                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11406
11407         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11408                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11409                 if (size == sizeof(__u32))
11410                         return true;
11411                 bpf_ctx_record_field_size(info, sizeof(__be16));
11412                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11413
11414         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11415              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11416                 /* Allow access to zero padding for backward compatibility */
11417                 bpf_ctx_record_field_size(info, sizeof(__u16));
11418                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11419
11420         default:
11421                 return false;
11422         }
11423 }
11424
11425 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11426                                         const struct bpf_insn *si,
11427                                         struct bpf_insn *insn_buf,
11428                                         struct bpf_prog *prog,
11429                                         u32 *target_size)
11430 {
11431         struct bpf_insn *insn = insn_buf;
11432
11433         switch (si->off) {
11434         case offsetof(struct bpf_sk_lookup, sk):
11435                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11436                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11437                 break;
11438
11439         case offsetof(struct bpf_sk_lookup, family):
11440                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11441                                       bpf_target_off(struct bpf_sk_lookup_kern,
11442                                                      family, 2, target_size));
11443                 break;
11444
11445         case offsetof(struct bpf_sk_lookup, protocol):
11446                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11447                                       bpf_target_off(struct bpf_sk_lookup_kern,
11448                                                      protocol, 2, target_size));
11449                 break;
11450
11451         case offsetof(struct bpf_sk_lookup, remote_ip4):
11452                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11453                                       bpf_target_off(struct bpf_sk_lookup_kern,
11454                                                      v4.saddr, 4, target_size));
11455                 break;
11456
11457         case offsetof(struct bpf_sk_lookup, local_ip4):
11458                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11459                                       bpf_target_off(struct bpf_sk_lookup_kern,
11460                                                      v4.daddr, 4, target_size));
11461                 break;
11462
11463         case bpf_ctx_range_till(struct bpf_sk_lookup,
11464                                 remote_ip6[0], remote_ip6[3]): {
11465 #if IS_ENABLED(CONFIG_IPV6)
11466                 int off = si->off;
11467
11468                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11469                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11470                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11471                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11472                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11473                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11474 #else
11475                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11476 #endif
11477                 break;
11478         }
11479         case bpf_ctx_range_till(struct bpf_sk_lookup,
11480                                 local_ip6[0], local_ip6[3]): {
11481 #if IS_ENABLED(CONFIG_IPV6)
11482                 int off = si->off;
11483
11484                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11485                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11486                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11487                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11488                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11489                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11490 #else
11491                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11492 #endif
11493                 break;
11494         }
11495         case offsetof(struct bpf_sk_lookup, remote_port):
11496                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11497                                       bpf_target_off(struct bpf_sk_lookup_kern,
11498                                                      sport, 2, target_size));
11499                 break;
11500
11501         case offsetofend(struct bpf_sk_lookup, remote_port):
11502                 *target_size = 2;
11503                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11504                 break;
11505
11506         case offsetof(struct bpf_sk_lookup, local_port):
11507                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11508                                       bpf_target_off(struct bpf_sk_lookup_kern,
11509                                                      dport, 2, target_size));
11510                 break;
11511
11512         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11513                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11514                                       bpf_target_off(struct bpf_sk_lookup_kern,
11515                                                      ingress_ifindex, 4, target_size));
11516                 break;
11517         }
11518
11519         return insn - insn_buf;
11520 }
11521
11522 const struct bpf_prog_ops sk_lookup_prog_ops = {
11523         .test_run = bpf_prog_test_run_sk_lookup,
11524 };
11525
11526 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11527         .get_func_proto         = sk_lookup_func_proto,
11528         .is_valid_access        = sk_lookup_is_valid_access,
11529         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11530 };
11531
11532 #endif /* CONFIG_INET */
11533
11534 DEFINE_BPF_DISPATCHER(xdp)
11535
11536 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11537 {
11538         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11539 }
11540
11541 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11542 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11543 BTF_SOCK_TYPE_xxx
11544 #undef BTF_SOCK_TYPE
11545
11546 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11547 {
11548         /* tcp6_sock type is not generated in dwarf and hence btf,
11549          * trigger an explicit type generation here.
11550          */
11551         BTF_TYPE_EMIT(struct tcp6_sock);
11552         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11553             sk->sk_family == AF_INET6)
11554                 return (unsigned long)sk;
11555
11556         return (unsigned long)NULL;
11557 }
11558
11559 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11560         .func                   = bpf_skc_to_tcp6_sock,
11561         .gpl_only               = false,
11562         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11563         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11564         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11565 };
11566
11567 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11568 {
11569         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11570                 return (unsigned long)sk;
11571
11572         return (unsigned long)NULL;
11573 }
11574
11575 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11576         .func                   = bpf_skc_to_tcp_sock,
11577         .gpl_only               = false,
11578         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11579         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11580         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11581 };
11582
11583 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11584 {
11585         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11586          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11587          */
11588         BTF_TYPE_EMIT(struct inet_timewait_sock);
11589         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11590
11591 #ifdef CONFIG_INET
11592         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11593                 return (unsigned long)sk;
11594 #endif
11595
11596 #if IS_BUILTIN(CONFIG_IPV6)
11597         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11598                 return (unsigned long)sk;
11599 #endif
11600
11601         return (unsigned long)NULL;
11602 }
11603
11604 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11605         .func                   = bpf_skc_to_tcp_timewait_sock,
11606         .gpl_only               = false,
11607         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11608         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11609         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11610 };
11611
11612 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11613 {
11614 #ifdef CONFIG_INET
11615         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11616                 return (unsigned long)sk;
11617 #endif
11618
11619 #if IS_BUILTIN(CONFIG_IPV6)
11620         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11621                 return (unsigned long)sk;
11622 #endif
11623
11624         return (unsigned long)NULL;
11625 }
11626
11627 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11628         .func                   = bpf_skc_to_tcp_request_sock,
11629         .gpl_only               = false,
11630         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11631         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11632         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11633 };
11634
11635 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11636 {
11637         /* udp6_sock type is not generated in dwarf and hence btf,
11638          * trigger an explicit type generation here.
11639          */
11640         BTF_TYPE_EMIT(struct udp6_sock);
11641         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11642             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11643                 return (unsigned long)sk;
11644
11645         return (unsigned long)NULL;
11646 }
11647
11648 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11649         .func                   = bpf_skc_to_udp6_sock,
11650         .gpl_only               = false,
11651         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11652         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11653         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11654 };
11655
11656 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11657 {
11658         /* unix_sock type is not generated in dwarf and hence btf,
11659          * trigger an explicit type generation here.
11660          */
11661         BTF_TYPE_EMIT(struct unix_sock);
11662         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11663                 return (unsigned long)sk;
11664
11665         return (unsigned long)NULL;
11666 }
11667
11668 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11669         .func                   = bpf_skc_to_unix_sock,
11670         .gpl_only               = false,
11671         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11672         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11673         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11674 };
11675
11676 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11677 {
11678         BTF_TYPE_EMIT(struct mptcp_sock);
11679         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11680 }
11681
11682 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11683         .func           = bpf_skc_to_mptcp_sock,
11684         .gpl_only       = false,
11685         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11686         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11687         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11688 };
11689
11690 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11691 {
11692         return (unsigned long)sock_from_file(file);
11693 }
11694
11695 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11696 BTF_ID(struct, socket)
11697 BTF_ID(struct, file)
11698
11699 const struct bpf_func_proto bpf_sock_from_file_proto = {
11700         .func           = bpf_sock_from_file,
11701         .gpl_only       = false,
11702         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11703         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11704         .arg1_type      = ARG_PTR_TO_BTF_ID,
11705         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11706 };
11707
11708 static const struct bpf_func_proto *
11709 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11710 {
11711         const struct bpf_func_proto *func;
11712
11713         switch (func_id) {
11714         case BPF_FUNC_skc_to_tcp6_sock:
11715                 func = &bpf_skc_to_tcp6_sock_proto;
11716                 break;
11717         case BPF_FUNC_skc_to_tcp_sock:
11718                 func = &bpf_skc_to_tcp_sock_proto;
11719                 break;
11720         case BPF_FUNC_skc_to_tcp_timewait_sock:
11721                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11722                 break;
11723         case BPF_FUNC_skc_to_tcp_request_sock:
11724                 func = &bpf_skc_to_tcp_request_sock_proto;
11725                 break;
11726         case BPF_FUNC_skc_to_udp6_sock:
11727                 func = &bpf_skc_to_udp6_sock_proto;
11728                 break;
11729         case BPF_FUNC_skc_to_unix_sock:
11730                 func = &bpf_skc_to_unix_sock_proto;
11731                 break;
11732         case BPF_FUNC_skc_to_mptcp_sock:
11733                 func = &bpf_skc_to_mptcp_sock_proto;
11734                 break;
11735         case BPF_FUNC_ktime_get_coarse_ns:
11736                 return &bpf_ktime_get_coarse_ns_proto;
11737         default:
11738                 return bpf_base_func_proto(func_id);
11739         }
11740
11741         if (!perfmon_capable())
11742                 return NULL;
11743
11744         return func;
11745 }
11746
11747 __diag_push();
11748 __diag_ignore_all("-Wmissing-prototypes",
11749                   "Global functions as their definitions will be in vmlinux BTF");
11750 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11751                                     struct bpf_dynptr_kern *ptr__uninit)
11752 {
11753         if (flags) {
11754                 bpf_dynptr_set_null(ptr__uninit);
11755                 return -EINVAL;
11756         }
11757
11758         bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11759
11760         return 0;
11761 }
11762
11763 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11764                                     struct bpf_dynptr_kern *ptr__uninit)
11765 {
11766         if (flags) {
11767                 bpf_dynptr_set_null(ptr__uninit);
11768                 return -EINVAL;
11769         }
11770
11771         bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11772
11773         return 0;
11774 }
11775 __diag_pop();
11776
11777 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11778                                struct bpf_dynptr_kern *ptr__uninit)
11779 {
11780         int err;
11781
11782         err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11783         if (err)
11784                 return err;
11785
11786         bpf_dynptr_set_rdonly(ptr__uninit);
11787
11788         return 0;
11789 }
11790
11791 BTF_SET8_START(bpf_kfunc_check_set_skb)
11792 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11793 BTF_SET8_END(bpf_kfunc_check_set_skb)
11794
11795 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11796 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11797 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11798
11799 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11800         .owner = THIS_MODULE,
11801         .set = &bpf_kfunc_check_set_skb,
11802 };
11803
11804 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11805         .owner = THIS_MODULE,
11806         .set = &bpf_kfunc_check_set_xdp,
11807 };
11808
11809 static int __init bpf_kfunc_init(void)
11810 {
11811         int ret;
11812
11813         ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11814         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11815         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11816         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11817         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11818         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11819         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11820         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11821         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11822         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11823         return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11824 }
11825 late_initcall(bpf_kfunc_init);
11826
11827 /* Disables missing prototype warnings */
11828 __diag_push();
11829 __diag_ignore_all("-Wmissing-prototypes",
11830                   "Global functions as their definitions will be in vmlinux BTF");
11831
11832 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11833  *
11834  * The function expects a non-NULL pointer to a socket, and invokes the
11835  * protocol specific socket destroy handlers.
11836  *
11837  * The helper can only be called from BPF contexts that have acquired the socket
11838  * locks.
11839  *
11840  * Parameters:
11841  * @sock: Pointer to socket to be destroyed
11842  *
11843  * Return:
11844  * On error, may return EPROTONOSUPPORT, EINVAL.
11845  * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11846  * 0 otherwise
11847  */
11848 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11849 {
11850         struct sock *sk = (struct sock *)sock;
11851
11852         /* The locking semantics that allow for synchronous execution of the
11853          * destroy handlers are only supported for TCP and UDP.
11854          * Supporting protocols will need to acquire sock lock in the BPF context
11855          * prior to invoking this kfunc.
11856          */
11857         if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11858                                            sk->sk_protocol != IPPROTO_UDP))
11859                 return -EOPNOTSUPP;
11860
11861         return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11862 }
11863
11864 __diag_pop()
11865
11866 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11867 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11868 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11869
11870 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11871 {
11872         if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11873             prog->expected_attach_type != BPF_TRACE_ITER)
11874                 return -EACCES;
11875         return 0;
11876 }
11877
11878 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11879         .owner = THIS_MODULE,
11880         .set   = &bpf_sk_iter_kfunc_ids,
11881         .filter = tracing_iter_filter,
11882 };
11883
11884 static int init_subsystem(void)
11885 {
11886         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11887 }
11888 late_initcall(init_subsystem);