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                 return -EAGAIN;
2493         }
2494         return flags & BPF_F_NEIGH ?
2495                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2496                                     &ri->nh : NULL) :
2497                __bpf_redirect(skb, dev, flags);
2498 out_drop:
2499         kfree_skb(skb);
2500         return -EINVAL;
2501 }
2502
2503 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2504 {
2505         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2506
2507         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2508                 return TC_ACT_SHOT;
2509
2510         ri->flags = flags;
2511         ri->tgt_index = ifindex;
2512
2513         return TC_ACT_REDIRECT;
2514 }
2515
2516 static const struct bpf_func_proto bpf_redirect_proto = {
2517         .func           = bpf_redirect,
2518         .gpl_only       = false,
2519         .ret_type       = RET_INTEGER,
2520         .arg1_type      = ARG_ANYTHING,
2521         .arg2_type      = ARG_ANYTHING,
2522 };
2523
2524 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2525 {
2526         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2527
2528         if (unlikely(flags))
2529                 return TC_ACT_SHOT;
2530
2531         ri->flags = BPF_F_PEER;
2532         ri->tgt_index = ifindex;
2533
2534         return TC_ACT_REDIRECT;
2535 }
2536
2537 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2538         .func           = bpf_redirect_peer,
2539         .gpl_only       = false,
2540         .ret_type       = RET_INTEGER,
2541         .arg1_type      = ARG_ANYTHING,
2542         .arg2_type      = ARG_ANYTHING,
2543 };
2544
2545 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2546            int, plen, u64, flags)
2547 {
2548         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2549
2550         if (unlikely((plen && plen < sizeof(*params)) || flags))
2551                 return TC_ACT_SHOT;
2552
2553         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2554         ri->tgt_index = ifindex;
2555
2556         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2557         if (plen)
2558                 memcpy(&ri->nh, params, sizeof(ri->nh));
2559
2560         return TC_ACT_REDIRECT;
2561 }
2562
2563 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2564         .func           = bpf_redirect_neigh,
2565         .gpl_only       = false,
2566         .ret_type       = RET_INTEGER,
2567         .arg1_type      = ARG_ANYTHING,
2568         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2569         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2570         .arg4_type      = ARG_ANYTHING,
2571 };
2572
2573 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2574 {
2575         msg->apply_bytes = bytes;
2576         return 0;
2577 }
2578
2579 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2580         .func           = bpf_msg_apply_bytes,
2581         .gpl_only       = false,
2582         .ret_type       = RET_INTEGER,
2583         .arg1_type      = ARG_PTR_TO_CTX,
2584         .arg2_type      = ARG_ANYTHING,
2585 };
2586
2587 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2588 {
2589         msg->cork_bytes = bytes;
2590         return 0;
2591 }
2592
2593 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2594         .func           = bpf_msg_cork_bytes,
2595         .gpl_only       = false,
2596         .ret_type       = RET_INTEGER,
2597         .arg1_type      = ARG_PTR_TO_CTX,
2598         .arg2_type      = ARG_ANYTHING,
2599 };
2600
2601 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2602            u32, end, u64, flags)
2603 {
2604         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2605         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2606         struct scatterlist *sge;
2607         u8 *raw, *to, *from;
2608         struct page *page;
2609
2610         if (unlikely(flags || end <= start))
2611                 return -EINVAL;
2612
2613         /* First find the starting scatterlist element */
2614         i = msg->sg.start;
2615         do {
2616                 offset += len;
2617                 len = sk_msg_elem(msg, i)->length;
2618                 if (start < offset + len)
2619                         break;
2620                 sk_msg_iter_var_next(i);
2621         } while (i != msg->sg.end);
2622
2623         if (unlikely(start >= offset + len))
2624                 return -EINVAL;
2625
2626         first_sge = i;
2627         /* The start may point into the sg element so we need to also
2628          * account for the headroom.
2629          */
2630         bytes_sg_total = start - offset + bytes;
2631         if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2632                 goto out;
2633
2634         /* At this point we need to linearize multiple scatterlist
2635          * elements or a single shared page. Either way we need to
2636          * copy into a linear buffer exclusively owned by BPF. Then
2637          * place the buffer in the scatterlist and fixup the original
2638          * entries by removing the entries now in the linear buffer
2639          * and shifting the remaining entries. For now we do not try
2640          * to copy partial entries to avoid complexity of running out
2641          * of sg_entry slots. The downside is reading a single byte
2642          * will copy the entire sg entry.
2643          */
2644         do {
2645                 copy += sk_msg_elem(msg, i)->length;
2646                 sk_msg_iter_var_next(i);
2647                 if (bytes_sg_total <= copy)
2648                         break;
2649         } while (i != msg->sg.end);
2650         last_sge = i;
2651
2652         if (unlikely(bytes_sg_total > copy))
2653                 return -EINVAL;
2654
2655         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2656                            get_order(copy));
2657         if (unlikely(!page))
2658                 return -ENOMEM;
2659
2660         raw = page_address(page);
2661         i = first_sge;
2662         do {
2663                 sge = sk_msg_elem(msg, i);
2664                 from = sg_virt(sge);
2665                 len = sge->length;
2666                 to = raw + poffset;
2667
2668                 memcpy(to, from, len);
2669                 poffset += len;
2670                 sge->length = 0;
2671                 put_page(sg_page(sge));
2672
2673                 sk_msg_iter_var_next(i);
2674         } while (i != last_sge);
2675
2676         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2677
2678         /* To repair sg ring we need to shift entries. If we only
2679          * had a single entry though we can just replace it and
2680          * be done. Otherwise walk the ring and shift the entries.
2681          */
2682         WARN_ON_ONCE(last_sge == first_sge);
2683         shift = last_sge > first_sge ?
2684                 last_sge - first_sge - 1 :
2685                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2686         if (!shift)
2687                 goto out;
2688
2689         i = first_sge;
2690         sk_msg_iter_var_next(i);
2691         do {
2692                 u32 move_from;
2693
2694                 if (i + shift >= NR_MSG_FRAG_IDS)
2695                         move_from = i + shift - NR_MSG_FRAG_IDS;
2696                 else
2697                         move_from = i + shift;
2698                 if (move_from == msg->sg.end)
2699                         break;
2700
2701                 msg->sg.data[i] = msg->sg.data[move_from];
2702                 msg->sg.data[move_from].length = 0;
2703                 msg->sg.data[move_from].page_link = 0;
2704                 msg->sg.data[move_from].offset = 0;
2705                 sk_msg_iter_var_next(i);
2706         } while (1);
2707
2708         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2709                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2710                       msg->sg.end - shift;
2711 out:
2712         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2713         msg->data_end = msg->data + bytes;
2714         return 0;
2715 }
2716
2717 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2718         .func           = bpf_msg_pull_data,
2719         .gpl_only       = false,
2720         .ret_type       = RET_INTEGER,
2721         .arg1_type      = ARG_PTR_TO_CTX,
2722         .arg2_type      = ARG_ANYTHING,
2723         .arg3_type      = ARG_ANYTHING,
2724         .arg4_type      = ARG_ANYTHING,
2725 };
2726
2727 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2728            u32, len, u64, flags)
2729 {
2730         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2731         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2732         u8 *raw, *to, *from;
2733         struct page *page;
2734
2735         if (unlikely(flags))
2736                 return -EINVAL;
2737
2738         if (unlikely(len == 0))
2739                 return 0;
2740
2741         /* First find the starting scatterlist element */
2742         i = msg->sg.start;
2743         do {
2744                 offset += l;
2745                 l = sk_msg_elem(msg, i)->length;
2746
2747                 if (start < offset + l)
2748                         break;
2749                 sk_msg_iter_var_next(i);
2750         } while (i != msg->sg.end);
2751
2752         if (start >= offset + l)
2753                 return -EINVAL;
2754
2755         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2756
2757         /* If no space available will fallback to copy, we need at
2758          * least one scatterlist elem available to push data into
2759          * when start aligns to the beginning of an element or two
2760          * when it falls inside an element. We handle the start equals
2761          * offset case because its the common case for inserting a
2762          * header.
2763          */
2764         if (!space || (space == 1 && start != offset))
2765                 copy = msg->sg.data[i].length;
2766
2767         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2768                            get_order(copy + len));
2769         if (unlikely(!page))
2770                 return -ENOMEM;
2771
2772         if (copy) {
2773                 int front, back;
2774
2775                 raw = page_address(page);
2776
2777                 psge = sk_msg_elem(msg, i);
2778                 front = start - offset;
2779                 back = psge->length - front;
2780                 from = sg_virt(psge);
2781
2782                 if (front)
2783                         memcpy(raw, from, front);
2784
2785                 if (back) {
2786                         from += front;
2787                         to = raw + front + len;
2788
2789                         memcpy(to, from, back);
2790                 }
2791
2792                 put_page(sg_page(psge));
2793         } else if (start - offset) {
2794                 psge = sk_msg_elem(msg, i);
2795                 rsge = sk_msg_elem_cpy(msg, i);
2796
2797                 psge->length = start - offset;
2798                 rsge.length -= psge->length;
2799                 rsge.offset += start;
2800
2801                 sk_msg_iter_var_next(i);
2802                 sg_unmark_end(psge);
2803                 sg_unmark_end(&rsge);
2804                 sk_msg_iter_next(msg, end);
2805         }
2806
2807         /* Slot(s) to place newly allocated data */
2808         new = i;
2809
2810         /* Shift one or two slots as needed */
2811         if (!copy) {
2812                 sge = sk_msg_elem_cpy(msg, i);
2813
2814                 sk_msg_iter_var_next(i);
2815                 sg_unmark_end(&sge);
2816                 sk_msg_iter_next(msg, end);
2817
2818                 nsge = sk_msg_elem_cpy(msg, i);
2819                 if (rsge.length) {
2820                         sk_msg_iter_var_next(i);
2821                         nnsge = sk_msg_elem_cpy(msg, i);
2822                 }
2823
2824                 while (i != msg->sg.end) {
2825                         msg->sg.data[i] = sge;
2826                         sge = nsge;
2827                         sk_msg_iter_var_next(i);
2828                         if (rsge.length) {
2829                                 nsge = nnsge;
2830                                 nnsge = sk_msg_elem_cpy(msg, i);
2831                         } else {
2832                                 nsge = sk_msg_elem_cpy(msg, i);
2833                         }
2834                 }
2835         }
2836
2837         /* Place newly allocated data buffer */
2838         sk_mem_charge(msg->sk, len);
2839         msg->sg.size += len;
2840         __clear_bit(new, msg->sg.copy);
2841         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2842         if (rsge.length) {
2843                 get_page(sg_page(&rsge));
2844                 sk_msg_iter_var_next(new);
2845                 msg->sg.data[new] = rsge;
2846         }
2847
2848         sk_msg_compute_data_pointers(msg);
2849         return 0;
2850 }
2851
2852 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2853         .func           = bpf_msg_push_data,
2854         .gpl_only       = false,
2855         .ret_type       = RET_INTEGER,
2856         .arg1_type      = ARG_PTR_TO_CTX,
2857         .arg2_type      = ARG_ANYTHING,
2858         .arg3_type      = ARG_ANYTHING,
2859         .arg4_type      = ARG_ANYTHING,
2860 };
2861
2862 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2863 {
2864         int prev;
2865
2866         do {
2867                 prev = i;
2868                 sk_msg_iter_var_next(i);
2869                 msg->sg.data[prev] = msg->sg.data[i];
2870         } while (i != msg->sg.end);
2871
2872         sk_msg_iter_prev(msg, end);
2873 }
2874
2875 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2876 {
2877         struct scatterlist tmp, sge;
2878
2879         sk_msg_iter_next(msg, end);
2880         sge = sk_msg_elem_cpy(msg, i);
2881         sk_msg_iter_var_next(i);
2882         tmp = sk_msg_elem_cpy(msg, i);
2883
2884         while (i != msg->sg.end) {
2885                 msg->sg.data[i] = sge;
2886                 sk_msg_iter_var_next(i);
2887                 sge = tmp;
2888                 tmp = sk_msg_elem_cpy(msg, i);
2889         }
2890 }
2891
2892 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2893            u32, len, u64, flags)
2894 {
2895         u32 i = 0, l = 0, space, offset = 0;
2896         u64 last = start + len;
2897         int pop;
2898
2899         if (unlikely(flags))
2900                 return -EINVAL;
2901
2902         /* First find the starting scatterlist element */
2903         i = msg->sg.start;
2904         do {
2905                 offset += l;
2906                 l = sk_msg_elem(msg, i)->length;
2907
2908                 if (start < offset + l)
2909                         break;
2910                 sk_msg_iter_var_next(i);
2911         } while (i != msg->sg.end);
2912
2913         /* Bounds checks: start and pop must be inside message */
2914         if (start >= offset + l || last >= msg->sg.size)
2915                 return -EINVAL;
2916
2917         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2918
2919         pop = len;
2920         /* --------------| offset
2921          * -| start      |-------- len -------|
2922          *
2923          *  |----- a ----|-------- pop -------|----- b ----|
2924          *  |______________________________________________| length
2925          *
2926          *
2927          * a:   region at front of scatter element to save
2928          * b:   region at back of scatter element to save when length > A + pop
2929          * pop: region to pop from element, same as input 'pop' here will be
2930          *      decremented below per iteration.
2931          *
2932          * Two top-level cases to handle when start != offset, first B is non
2933          * zero and second B is zero corresponding to when a pop includes more
2934          * than one element.
2935          *
2936          * Then if B is non-zero AND there is no space allocate space and
2937          * compact A, B regions into page. If there is space shift ring to
2938          * the rigth free'ing the next element in ring to place B, leaving
2939          * A untouched except to reduce length.
2940          */
2941         if (start != offset) {
2942                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2943                 int a = start;
2944                 int b = sge->length - pop - a;
2945
2946                 sk_msg_iter_var_next(i);
2947
2948                 if (pop < sge->length - a) {
2949                         if (space) {
2950                                 sge->length = a;
2951                                 sk_msg_shift_right(msg, i);
2952                                 nsge = sk_msg_elem(msg, i);
2953                                 get_page(sg_page(sge));
2954                                 sg_set_page(nsge,
2955                                             sg_page(sge),
2956                                             b, sge->offset + pop + a);
2957                         } else {
2958                                 struct page *page, *orig;
2959                                 u8 *to, *from;
2960
2961                                 page = alloc_pages(__GFP_NOWARN |
2962                                                    __GFP_COMP   | GFP_ATOMIC,
2963                                                    get_order(a + b));
2964                                 if (unlikely(!page))
2965                                         return -ENOMEM;
2966
2967                                 sge->length = a;
2968                                 orig = sg_page(sge);
2969                                 from = sg_virt(sge);
2970                                 to = page_address(page);
2971                                 memcpy(to, from, a);
2972                                 memcpy(to + a, from + a + pop, b);
2973                                 sg_set_page(sge, page, a + b, 0);
2974                                 put_page(orig);
2975                         }
2976                         pop = 0;
2977                 } else if (pop >= sge->length - a) {
2978                         pop -= (sge->length - a);
2979                         sge->length = a;
2980                 }
2981         }
2982
2983         /* From above the current layout _must_ be as follows,
2984          *
2985          * -| offset
2986          * -| start
2987          *
2988          *  |---- pop ---|---------------- b ------------|
2989          *  |____________________________________________| length
2990          *
2991          * Offset and start of the current msg elem are equal because in the
2992          * previous case we handled offset != start and either consumed the
2993          * entire element and advanced to the next element OR pop == 0.
2994          *
2995          * Two cases to handle here are first pop is less than the length
2996          * leaving some remainder b above. Simply adjust the element's layout
2997          * in this case. Or pop >= length of the element so that b = 0. In this
2998          * case advance to next element decrementing pop.
2999          */
3000         while (pop) {
3001                 struct scatterlist *sge = sk_msg_elem(msg, i);
3002
3003                 if (pop < sge->length) {
3004                         sge->length -= pop;
3005                         sge->offset += pop;
3006                         pop = 0;
3007                 } else {
3008                         pop -= sge->length;
3009                         sk_msg_shift_left(msg, i);
3010                 }
3011                 sk_msg_iter_var_next(i);
3012         }
3013
3014         sk_mem_uncharge(msg->sk, len - pop);
3015         msg->sg.size -= (len - pop);
3016         sk_msg_compute_data_pointers(msg);
3017         return 0;
3018 }
3019
3020 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3021         .func           = bpf_msg_pop_data,
3022         .gpl_only       = false,
3023         .ret_type       = RET_INTEGER,
3024         .arg1_type      = ARG_PTR_TO_CTX,
3025         .arg2_type      = ARG_ANYTHING,
3026         .arg3_type      = ARG_ANYTHING,
3027         .arg4_type      = ARG_ANYTHING,
3028 };
3029
3030 #ifdef CONFIG_CGROUP_NET_CLASSID
3031 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3032 {
3033         return __task_get_classid(current);
3034 }
3035
3036 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3037         .func           = bpf_get_cgroup_classid_curr,
3038         .gpl_only       = false,
3039         .ret_type       = RET_INTEGER,
3040 };
3041
3042 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3043 {
3044         struct sock *sk = skb_to_full_sk(skb);
3045
3046         if (!sk || !sk_fullsock(sk))
3047                 return 0;
3048
3049         return sock_cgroup_classid(&sk->sk_cgrp_data);
3050 }
3051
3052 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3053         .func           = bpf_skb_cgroup_classid,
3054         .gpl_only       = false,
3055         .ret_type       = RET_INTEGER,
3056         .arg1_type      = ARG_PTR_TO_CTX,
3057 };
3058 #endif
3059
3060 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3061 {
3062         return task_get_classid(skb);
3063 }
3064
3065 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3066         .func           = bpf_get_cgroup_classid,
3067         .gpl_only       = false,
3068         .ret_type       = RET_INTEGER,
3069         .arg1_type      = ARG_PTR_TO_CTX,
3070 };
3071
3072 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3073 {
3074         return dst_tclassid(skb);
3075 }
3076
3077 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3078         .func           = bpf_get_route_realm,
3079         .gpl_only       = false,
3080         .ret_type       = RET_INTEGER,
3081         .arg1_type      = ARG_PTR_TO_CTX,
3082 };
3083
3084 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3085 {
3086         /* If skb_clear_hash() was called due to mangling, we can
3087          * trigger SW recalculation here. Later access to hash
3088          * can then use the inline skb->hash via context directly
3089          * instead of calling this helper again.
3090          */
3091         return skb_get_hash(skb);
3092 }
3093
3094 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3095         .func           = bpf_get_hash_recalc,
3096         .gpl_only       = false,
3097         .ret_type       = RET_INTEGER,
3098         .arg1_type      = ARG_PTR_TO_CTX,
3099 };
3100
3101 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3102 {
3103         /* After all direct packet write, this can be used once for
3104          * triggering a lazy recalc on next skb_get_hash() invocation.
3105          */
3106         skb_clear_hash(skb);
3107         return 0;
3108 }
3109
3110 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3111         .func           = bpf_set_hash_invalid,
3112         .gpl_only       = false,
3113         .ret_type       = RET_INTEGER,
3114         .arg1_type      = ARG_PTR_TO_CTX,
3115 };
3116
3117 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3118 {
3119         /* Set user specified hash as L4(+), so that it gets returned
3120          * on skb_get_hash() call unless BPF prog later on triggers a
3121          * skb_clear_hash().
3122          */
3123         __skb_set_sw_hash(skb, hash, true);
3124         return 0;
3125 }
3126
3127 static const struct bpf_func_proto bpf_set_hash_proto = {
3128         .func           = bpf_set_hash,
3129         .gpl_only       = false,
3130         .ret_type       = RET_INTEGER,
3131         .arg1_type      = ARG_PTR_TO_CTX,
3132         .arg2_type      = ARG_ANYTHING,
3133 };
3134
3135 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3136            u16, vlan_tci)
3137 {
3138         int ret;
3139
3140         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3141                      vlan_proto != htons(ETH_P_8021AD)))
3142                 vlan_proto = htons(ETH_P_8021Q);
3143
3144         bpf_push_mac_rcsum(skb);
3145         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3146         bpf_pull_mac_rcsum(skb);
3147
3148         bpf_compute_data_pointers(skb);
3149         return ret;
3150 }
3151
3152 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3153         .func           = bpf_skb_vlan_push,
3154         .gpl_only       = false,
3155         .ret_type       = RET_INTEGER,
3156         .arg1_type      = ARG_PTR_TO_CTX,
3157         .arg2_type      = ARG_ANYTHING,
3158         .arg3_type      = ARG_ANYTHING,
3159 };
3160
3161 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3162 {
3163         int ret;
3164
3165         bpf_push_mac_rcsum(skb);
3166         ret = skb_vlan_pop(skb);
3167         bpf_pull_mac_rcsum(skb);
3168
3169         bpf_compute_data_pointers(skb);
3170         return ret;
3171 }
3172
3173 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3174         .func           = bpf_skb_vlan_pop,
3175         .gpl_only       = false,
3176         .ret_type       = RET_INTEGER,
3177         .arg1_type      = ARG_PTR_TO_CTX,
3178 };
3179
3180 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3181 {
3182         /* Caller already did skb_cow() with len as headroom,
3183          * so no need to do it here.
3184          */
3185         skb_push(skb, len);
3186         memmove(skb->data, skb->data + len, off);
3187         memset(skb->data + off, 0, len);
3188
3189         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3190          * needed here as it does not change the skb->csum
3191          * result for checksum complete when summing over
3192          * zeroed blocks.
3193          */
3194         return 0;
3195 }
3196
3197 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3198 {
3199         void *old_data;
3200
3201         /* skb_ensure_writable() is not needed here, as we're
3202          * already working on an uncloned skb.
3203          */
3204         if (unlikely(!pskb_may_pull(skb, off + len)))
3205                 return -ENOMEM;
3206
3207         old_data = skb->data;
3208         __skb_pull(skb, len);
3209         skb_postpull_rcsum(skb, old_data + off, len);
3210         memmove(skb->data, old_data, off);
3211
3212         return 0;
3213 }
3214
3215 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3216 {
3217         bool trans_same = skb->transport_header == skb->network_header;
3218         int ret;
3219
3220         /* There's no need for __skb_push()/__skb_pull() pair to
3221          * get to the start of the mac header as we're guaranteed
3222          * to always start from here under eBPF.
3223          */
3224         ret = bpf_skb_generic_push(skb, off, len);
3225         if (likely(!ret)) {
3226                 skb->mac_header -= len;
3227                 skb->network_header -= len;
3228                 if (trans_same)
3229                         skb->transport_header = skb->network_header;
3230         }
3231
3232         return ret;
3233 }
3234
3235 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3236 {
3237         bool trans_same = skb->transport_header == skb->network_header;
3238         int ret;
3239
3240         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3241         ret = bpf_skb_generic_pop(skb, off, len);
3242         if (likely(!ret)) {
3243                 skb->mac_header += len;
3244                 skb->network_header += len;
3245                 if (trans_same)
3246                         skb->transport_header = skb->network_header;
3247         }
3248
3249         return ret;
3250 }
3251
3252 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3253 {
3254         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3255         u32 off = skb_mac_header_len(skb);
3256         int ret;
3257
3258         ret = skb_cow(skb, len_diff);
3259         if (unlikely(ret < 0))
3260                 return ret;
3261
3262         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3263         if (unlikely(ret < 0))
3264                 return ret;
3265
3266         if (skb_is_gso(skb)) {
3267                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3268
3269                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3270                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3271                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3272                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3273                 }
3274         }
3275
3276         skb->protocol = htons(ETH_P_IPV6);
3277         skb_clear_hash(skb);
3278
3279         return 0;
3280 }
3281
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3283 {
3284         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285         u32 off = skb_mac_header_len(skb);
3286         int ret;
3287
3288         ret = skb_unclone(skb, GFP_ATOMIC);
3289         if (unlikely(ret < 0))
3290                 return ret;
3291
3292         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3293         if (unlikely(ret < 0))
3294                 return ret;
3295
3296         if (skb_is_gso(skb)) {
3297                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3298
3299                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3300                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3301                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3302                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3303                 }
3304         }
3305
3306         skb->protocol = htons(ETH_P_IP);
3307         skb_clear_hash(skb);
3308
3309         return 0;
3310 }
3311
3312 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3313 {
3314         __be16 from_proto = skb->protocol;
3315
3316         if (from_proto == htons(ETH_P_IP) &&
3317               to_proto == htons(ETH_P_IPV6))
3318                 return bpf_skb_proto_4_to_6(skb);
3319
3320         if (from_proto == htons(ETH_P_IPV6) &&
3321               to_proto == htons(ETH_P_IP))
3322                 return bpf_skb_proto_6_to_4(skb);
3323
3324         return -ENOTSUPP;
3325 }
3326
3327 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3328            u64, flags)
3329 {
3330         int ret;
3331
3332         if (unlikely(flags))
3333                 return -EINVAL;
3334
3335         /* General idea is that this helper does the basic groundwork
3336          * needed for changing the protocol, and eBPF program fills the
3337          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3338          * and other helpers, rather than passing a raw buffer here.
3339          *
3340          * The rationale is to keep this minimal and without a need to
3341          * deal with raw packet data. F.e. even if we would pass buffers
3342          * here, the program still needs to call the bpf_lX_csum_replace()
3343          * helpers anyway. Plus, this way we keep also separation of
3344          * concerns, since f.e. bpf_skb_store_bytes() should only take
3345          * care of stores.
3346          *
3347          * Currently, additional options and extension header space are
3348          * not supported, but flags register is reserved so we can adapt
3349          * that. For offloads, we mark packet as dodgy, so that headers
3350          * need to be verified first.
3351          */
3352         ret = bpf_skb_proto_xlat(skb, proto);
3353         bpf_compute_data_pointers(skb);
3354         return ret;
3355 }
3356
3357 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3358         .func           = bpf_skb_change_proto,
3359         .gpl_only       = false,
3360         .ret_type       = RET_INTEGER,
3361         .arg1_type      = ARG_PTR_TO_CTX,
3362         .arg2_type      = ARG_ANYTHING,
3363         .arg3_type      = ARG_ANYTHING,
3364 };
3365
3366 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3367 {
3368         /* We only allow a restricted subset to be changed for now. */
3369         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3370                      !skb_pkt_type_ok(pkt_type)))
3371                 return -EINVAL;
3372
3373         skb->pkt_type = pkt_type;
3374         return 0;
3375 }
3376
3377 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3378         .func           = bpf_skb_change_type,
3379         .gpl_only       = false,
3380         .ret_type       = RET_INTEGER,
3381         .arg1_type      = ARG_PTR_TO_CTX,
3382         .arg2_type      = ARG_ANYTHING,
3383 };
3384
3385 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3386 {
3387         switch (skb->protocol) {
3388         case htons(ETH_P_IP):
3389                 return sizeof(struct iphdr);
3390         case htons(ETH_P_IPV6):
3391                 return sizeof(struct ipv6hdr);
3392         default:
3393                 return ~0U;
3394         }
3395 }
3396
3397 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3398                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3399
3400 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK    (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3401                                          BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3402
3403 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3404                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3405                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3406                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3407                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3408                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3409                                           BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3410                                          BPF_F_ADJ_ROOM_DECAP_L3_MASK)
3411
3412 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3413                             u64 flags)
3414 {
3415         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3416         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3417         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3418         unsigned int gso_type = SKB_GSO_DODGY;
3419         int ret;
3420
3421         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3422                 /* udp gso_size delineates datagrams, only allow if fixed */
3423                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3424                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3425                         return -ENOTSUPP;
3426         }
3427
3428         ret = skb_cow_head(skb, len_diff);
3429         if (unlikely(ret < 0))
3430                 return ret;
3431
3432         if (encap) {
3433                 if (skb->protocol != htons(ETH_P_IP) &&
3434                     skb->protocol != htons(ETH_P_IPV6))
3435                         return -ENOTSUPP;
3436
3437                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3438                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3439                         return -EINVAL;
3440
3441                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3442                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3443                         return -EINVAL;
3444
3445                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3446                     inner_mac_len < ETH_HLEN)
3447                         return -EINVAL;
3448
3449                 if (skb->encapsulation)
3450                         return -EALREADY;
3451
3452                 mac_len = skb->network_header - skb->mac_header;
3453                 inner_net = skb->network_header;
3454                 if (inner_mac_len > len_diff)
3455                         return -EINVAL;
3456                 inner_trans = skb->transport_header;
3457         }
3458
3459         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3460         if (unlikely(ret < 0))
3461                 return ret;
3462
3463         if (encap) {
3464                 skb->inner_mac_header = inner_net - inner_mac_len;
3465                 skb->inner_network_header = inner_net;
3466                 skb->inner_transport_header = inner_trans;
3467
3468                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3469                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3470                 else
3471                         skb_set_inner_protocol(skb, skb->protocol);
3472
3473                 skb->encapsulation = 1;
3474                 skb_set_network_header(skb, mac_len);
3475
3476                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3477                         gso_type |= SKB_GSO_UDP_TUNNEL;
3478                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3479                         gso_type |= SKB_GSO_GRE;
3480                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3481                         gso_type |= SKB_GSO_IPXIP6;
3482                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3483                         gso_type |= SKB_GSO_IPXIP4;
3484
3485                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3486                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3487                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3488                                         sizeof(struct ipv6hdr) :
3489                                         sizeof(struct iphdr);
3490
3491                         skb_set_transport_header(skb, mac_len + nh_len);
3492                 }
3493
3494                 /* Match skb->protocol to new outer l3 protocol */
3495                 if (skb->protocol == htons(ETH_P_IP) &&
3496                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3497                         skb->protocol = htons(ETH_P_IPV6);
3498                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3499                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3500                         skb->protocol = htons(ETH_P_IP);
3501         }
3502
3503         if (skb_is_gso(skb)) {
3504                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3505
3506                 /* Due to header grow, MSS needs to be downgraded. */
3507                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3508                         skb_decrease_gso_size(shinfo, len_diff);
3509
3510                 /* Header must be checked, and gso_segs recomputed. */
3511                 shinfo->gso_type |= gso_type;
3512                 shinfo->gso_segs = 0;
3513         }
3514
3515         return 0;
3516 }
3517
3518 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3519                               u64 flags)
3520 {
3521         int ret;
3522
3523         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3524                                BPF_F_ADJ_ROOM_DECAP_L3_MASK |
3525                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3526                 return -EINVAL;
3527
3528         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3529                 /* udp gso_size delineates datagrams, only allow if fixed */
3530                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3531                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3532                         return -ENOTSUPP;
3533         }
3534
3535         ret = skb_unclone(skb, GFP_ATOMIC);
3536         if (unlikely(ret < 0))
3537                 return ret;
3538
3539         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3540         if (unlikely(ret < 0))
3541                 return ret;
3542
3543         /* Match skb->protocol to new outer l3 protocol */
3544         if (skb->protocol == htons(ETH_P_IP) &&
3545             flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3546                 skb->protocol = htons(ETH_P_IPV6);
3547         else if (skb->protocol == htons(ETH_P_IPV6) &&
3548                  flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3549                 skb->protocol = htons(ETH_P_IP);
3550
3551         if (skb_is_gso(skb)) {
3552                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3553
3554                 /* Due to header shrink, MSS can be upgraded. */
3555                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3556                         skb_increase_gso_size(shinfo, len_diff);
3557
3558                 /* Header must be checked, and gso_segs recomputed. */
3559                 shinfo->gso_type |= SKB_GSO_DODGY;
3560                 shinfo->gso_segs = 0;
3561         }
3562
3563         return 0;
3564 }
3565
3566 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3567
3568 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3569            u32, mode, u64, flags)
3570 {
3571         u32 len_diff_abs = abs(len_diff);
3572         bool shrink = len_diff < 0;
3573         int ret = 0;
3574
3575         if (unlikely(flags || mode))
3576                 return -EINVAL;
3577         if (unlikely(len_diff_abs > 0xfffU))
3578                 return -EFAULT;
3579
3580         if (!shrink) {
3581                 ret = skb_cow(skb, len_diff);
3582                 if (unlikely(ret < 0))
3583                         return ret;
3584                 __skb_push(skb, len_diff_abs);
3585                 memset(skb->data, 0, len_diff_abs);
3586         } else {
3587                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3588                         return -ENOMEM;
3589                 __skb_pull(skb, len_diff_abs);
3590         }
3591         if (tls_sw_has_ctx_rx(skb->sk)) {
3592                 struct strp_msg *rxm = strp_msg(skb);
3593
3594                 rxm->full_len += len_diff;
3595         }
3596         return ret;
3597 }
3598
3599 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3600         .func           = sk_skb_adjust_room,
3601         .gpl_only       = false,
3602         .ret_type       = RET_INTEGER,
3603         .arg1_type      = ARG_PTR_TO_CTX,
3604         .arg2_type      = ARG_ANYTHING,
3605         .arg3_type      = ARG_ANYTHING,
3606         .arg4_type      = ARG_ANYTHING,
3607 };
3608
3609 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3610            u32, mode, u64, flags)
3611 {
3612         u32 len_cur, len_diff_abs = abs(len_diff);
3613         u32 len_min = bpf_skb_net_base_len(skb);
3614         u32 len_max = BPF_SKB_MAX_LEN;
3615         __be16 proto = skb->protocol;
3616         bool shrink = len_diff < 0;
3617         u32 off;
3618         int ret;
3619
3620         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3621                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3622                 return -EINVAL;
3623         if (unlikely(len_diff_abs > 0xfffU))
3624                 return -EFAULT;
3625         if (unlikely(proto != htons(ETH_P_IP) &&
3626                      proto != htons(ETH_P_IPV6)))
3627                 return -ENOTSUPP;
3628
3629         off = skb_mac_header_len(skb);
3630         switch (mode) {
3631         case BPF_ADJ_ROOM_NET:
3632                 off += bpf_skb_net_base_len(skb);
3633                 break;
3634         case BPF_ADJ_ROOM_MAC:
3635                 break;
3636         default:
3637                 return -ENOTSUPP;
3638         }
3639
3640         if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3641                 if (!shrink)
3642                         return -EINVAL;
3643
3644                 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3645                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3646                         len_min = sizeof(struct iphdr);
3647                         break;
3648                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3649                         len_min = sizeof(struct ipv6hdr);
3650                         break;
3651                 default:
3652                         return -EINVAL;
3653                 }
3654         }
3655
3656         len_cur = skb->len - skb_network_offset(skb);
3657         if ((shrink && (len_diff_abs >= len_cur ||
3658                         len_cur - len_diff_abs < len_min)) ||
3659             (!shrink && (skb->len + len_diff_abs > len_max &&
3660                          !skb_is_gso(skb))))
3661                 return -ENOTSUPP;
3662
3663         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3664                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3665         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3666                 __skb_reset_checksum_unnecessary(skb);
3667
3668         bpf_compute_data_pointers(skb);
3669         return ret;
3670 }
3671
3672 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3673         .func           = bpf_skb_adjust_room,
3674         .gpl_only       = false,
3675         .ret_type       = RET_INTEGER,
3676         .arg1_type      = ARG_PTR_TO_CTX,
3677         .arg2_type      = ARG_ANYTHING,
3678         .arg3_type      = ARG_ANYTHING,
3679         .arg4_type      = ARG_ANYTHING,
3680 };
3681
3682 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3683 {
3684         u32 min_len = skb_network_offset(skb);
3685
3686         if (skb_transport_header_was_set(skb))
3687                 min_len = skb_transport_offset(skb);
3688         if (skb->ip_summed == CHECKSUM_PARTIAL)
3689                 min_len = skb_checksum_start_offset(skb) +
3690                           skb->csum_offset + sizeof(__sum16);
3691         return min_len;
3692 }
3693
3694 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3695 {
3696         unsigned int old_len = skb->len;
3697         int ret;
3698
3699         ret = __skb_grow_rcsum(skb, new_len);
3700         if (!ret)
3701                 memset(skb->data + old_len, 0, new_len - old_len);
3702         return ret;
3703 }
3704
3705 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3706 {
3707         return __skb_trim_rcsum(skb, new_len);
3708 }
3709
3710 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3711                                         u64 flags)
3712 {
3713         u32 max_len = BPF_SKB_MAX_LEN;
3714         u32 min_len = __bpf_skb_min_len(skb);
3715         int ret;
3716
3717         if (unlikely(flags || new_len > max_len || new_len < min_len))
3718                 return -EINVAL;
3719         if (skb->encapsulation)
3720                 return -ENOTSUPP;
3721
3722         /* The basic idea of this helper is that it's performing the
3723          * needed work to either grow or trim an skb, and eBPF program
3724          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3725          * bpf_lX_csum_replace() and others rather than passing a raw
3726          * buffer here. This one is a slow path helper and intended
3727          * for replies with control messages.
3728          *
3729          * Like in bpf_skb_change_proto(), we want to keep this rather
3730          * minimal and without protocol specifics so that we are able
3731          * to separate concerns as in bpf_skb_store_bytes() should only
3732          * be the one responsible for writing buffers.
3733          *
3734          * It's really expected to be a slow path operation here for
3735          * control message replies, so we're implicitly linearizing,
3736          * uncloning and drop offloads from the skb by this.
3737          */
3738         ret = __bpf_try_make_writable(skb, skb->len);
3739         if (!ret) {
3740                 if (new_len > skb->len)
3741                         ret = bpf_skb_grow_rcsum(skb, new_len);
3742                 else if (new_len < skb->len)
3743                         ret = bpf_skb_trim_rcsum(skb, new_len);
3744                 if (!ret && skb_is_gso(skb))
3745                         skb_gso_reset(skb);
3746         }
3747         return ret;
3748 }
3749
3750 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3751            u64, flags)
3752 {
3753         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3754
3755         bpf_compute_data_pointers(skb);
3756         return ret;
3757 }
3758
3759 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3760         .func           = bpf_skb_change_tail,
3761         .gpl_only       = false,
3762         .ret_type       = RET_INTEGER,
3763         .arg1_type      = ARG_PTR_TO_CTX,
3764         .arg2_type      = ARG_ANYTHING,
3765         .arg3_type      = ARG_ANYTHING,
3766 };
3767
3768 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3769            u64, flags)
3770 {
3771         return __bpf_skb_change_tail(skb, new_len, flags);
3772 }
3773
3774 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3775         .func           = sk_skb_change_tail,
3776         .gpl_only       = false,
3777         .ret_type       = RET_INTEGER,
3778         .arg1_type      = ARG_PTR_TO_CTX,
3779         .arg2_type      = ARG_ANYTHING,
3780         .arg3_type      = ARG_ANYTHING,
3781 };
3782
3783 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3784                                         u64 flags)
3785 {
3786         u32 max_len = BPF_SKB_MAX_LEN;
3787         u32 new_len = skb->len + head_room;
3788         int ret;
3789
3790         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3791                      new_len < skb->len))
3792                 return -EINVAL;
3793
3794         ret = skb_cow(skb, head_room);
3795         if (likely(!ret)) {
3796                 /* Idea for this helper is that we currently only
3797                  * allow to expand on mac header. This means that
3798                  * skb->protocol network header, etc, stay as is.
3799                  * Compared to bpf_skb_change_tail(), we're more
3800                  * flexible due to not needing to linearize or
3801                  * reset GSO. Intention for this helper is to be
3802                  * used by an L3 skb that needs to push mac header
3803                  * for redirection into L2 device.
3804                  */
3805                 __skb_push(skb, head_room);
3806                 memset(skb->data, 0, head_room);
3807                 skb_reset_mac_header(skb);
3808                 skb_reset_mac_len(skb);
3809         }
3810
3811         return ret;
3812 }
3813
3814 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3815            u64, flags)
3816 {
3817         int ret = __bpf_skb_change_head(skb, head_room, flags);
3818
3819         bpf_compute_data_pointers(skb);
3820         return ret;
3821 }
3822
3823 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3824         .func           = bpf_skb_change_head,
3825         .gpl_only       = false,
3826         .ret_type       = RET_INTEGER,
3827         .arg1_type      = ARG_PTR_TO_CTX,
3828         .arg2_type      = ARG_ANYTHING,
3829         .arg3_type      = ARG_ANYTHING,
3830 };
3831
3832 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3833            u64, flags)
3834 {
3835         return __bpf_skb_change_head(skb, head_room, flags);
3836 }
3837
3838 static const struct bpf_func_proto sk_skb_change_head_proto = {
3839         .func           = sk_skb_change_head,
3840         .gpl_only       = false,
3841         .ret_type       = RET_INTEGER,
3842         .arg1_type      = ARG_PTR_TO_CTX,
3843         .arg2_type      = ARG_ANYTHING,
3844         .arg3_type      = ARG_ANYTHING,
3845 };
3846
3847 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp)
3848 {
3849         return xdp_get_buff_len(xdp);
3850 }
3851
3852 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3853         .func           = bpf_xdp_get_buff_len,
3854         .gpl_only       = false,
3855         .ret_type       = RET_INTEGER,
3856         .arg1_type      = ARG_PTR_TO_CTX,
3857 };
3858
3859 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3860
3861 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3862         .func           = bpf_xdp_get_buff_len,
3863         .gpl_only       = false,
3864         .arg1_type      = ARG_PTR_TO_BTF_ID,
3865         .arg1_btf_id    = &bpf_xdp_get_buff_len_bpf_ids[0],
3866 };
3867
3868 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3869 {
3870         return xdp_data_meta_unsupported(xdp) ? 0 :
3871                xdp->data - xdp->data_meta;
3872 }
3873
3874 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3875 {
3876         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3877         unsigned long metalen = xdp_get_metalen(xdp);
3878         void *data_start = xdp_frame_end + metalen;
3879         void *data = xdp->data + offset;
3880
3881         if (unlikely(data < data_start ||
3882                      data > xdp->data_end - ETH_HLEN))
3883                 return -EINVAL;
3884
3885         if (metalen)
3886                 memmove(xdp->data_meta + offset,
3887                         xdp->data_meta, metalen);
3888         xdp->data_meta += offset;
3889         xdp->data = data;
3890
3891         return 0;
3892 }
3893
3894 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3895         .func           = bpf_xdp_adjust_head,
3896         .gpl_only       = false,
3897         .ret_type       = RET_INTEGER,
3898         .arg1_type      = ARG_PTR_TO_CTX,
3899         .arg2_type      = ARG_ANYTHING,
3900 };
3901
3902 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3903                       void *buf, unsigned long len, bool flush)
3904 {
3905         unsigned long ptr_len, ptr_off = 0;
3906         skb_frag_t *next_frag, *end_frag;
3907         struct skb_shared_info *sinfo;
3908         void *src, *dst;
3909         u8 *ptr_buf;
3910
3911         if (likely(xdp->data_end - xdp->data >= off + len)) {
3912                 src = flush ? buf : xdp->data + off;
3913                 dst = flush ? xdp->data + off : buf;
3914                 memcpy(dst, src, len);
3915                 return;
3916         }
3917
3918         sinfo = xdp_get_shared_info_from_buff(xdp);
3919         end_frag = &sinfo->frags[sinfo->nr_frags];
3920         next_frag = &sinfo->frags[0];
3921
3922         ptr_len = xdp->data_end - xdp->data;
3923         ptr_buf = xdp->data;
3924
3925         while (true) {
3926                 if (off < ptr_off + ptr_len) {
3927                         unsigned long copy_off = off - ptr_off;
3928                         unsigned long copy_len = min(len, ptr_len - copy_off);
3929
3930                         src = flush ? buf : ptr_buf + copy_off;
3931                         dst = flush ? ptr_buf + copy_off : buf;
3932                         memcpy(dst, src, copy_len);
3933
3934                         off += copy_len;
3935                         len -= copy_len;
3936                         buf += copy_len;
3937                 }
3938
3939                 if (!len || next_frag == end_frag)
3940                         break;
3941
3942                 ptr_off += ptr_len;
3943                 ptr_buf = skb_frag_address(next_frag);
3944                 ptr_len = skb_frag_size(next_frag);
3945                 next_frag++;
3946         }
3947 }
3948
3949 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3950 {
3951         u32 size = xdp->data_end - xdp->data;
3952         struct skb_shared_info *sinfo;
3953         void *addr = xdp->data;
3954         int i;
3955
3956         if (unlikely(offset > 0xffff || len > 0xffff))
3957                 return ERR_PTR(-EFAULT);
3958
3959         if (unlikely(offset + len > xdp_get_buff_len(xdp)))
3960                 return ERR_PTR(-EINVAL);
3961
3962         if (likely(offset < size)) /* linear area */
3963                 goto out;
3964
3965         sinfo = xdp_get_shared_info_from_buff(xdp);
3966         offset -= size;
3967         for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3968                 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3969
3970                 if  (offset < frag_size) {
3971                         addr = skb_frag_address(&sinfo->frags[i]);
3972                         size = frag_size;
3973                         break;
3974                 }
3975                 offset -= frag_size;
3976         }
3977 out:
3978         return offset + len <= size ? addr + offset : NULL;
3979 }
3980
3981 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3982            void *, buf, u32, len)
3983 {
3984         void *ptr;
3985
3986         ptr = bpf_xdp_pointer(xdp, offset, len);
3987         if (IS_ERR(ptr))
3988                 return PTR_ERR(ptr);
3989
3990         if (!ptr)
3991                 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3992         else
3993                 memcpy(buf, ptr, len);
3994
3995         return 0;
3996 }
3997
3998 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3999         .func           = bpf_xdp_load_bytes,
4000         .gpl_only       = false,
4001         .ret_type       = RET_INTEGER,
4002         .arg1_type      = ARG_PTR_TO_CTX,
4003         .arg2_type      = ARG_ANYTHING,
4004         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4005         .arg4_type      = ARG_CONST_SIZE,
4006 };
4007
4008 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4009 {
4010         return ____bpf_xdp_load_bytes(xdp, offset, buf, len);
4011 }
4012
4013 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
4014            void *, buf, u32, len)
4015 {
4016         void *ptr;
4017
4018         ptr = bpf_xdp_pointer(xdp, offset, len);
4019         if (IS_ERR(ptr))
4020                 return PTR_ERR(ptr);
4021
4022         if (!ptr)
4023                 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4024         else
4025                 memcpy(ptr, buf, len);
4026
4027         return 0;
4028 }
4029
4030 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4031         .func           = bpf_xdp_store_bytes,
4032         .gpl_only       = false,
4033         .ret_type       = RET_INTEGER,
4034         .arg1_type      = ARG_PTR_TO_CTX,
4035         .arg2_type      = ARG_ANYTHING,
4036         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4037         .arg4_type      = ARG_CONST_SIZE,
4038 };
4039
4040 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len)
4041 {
4042         return ____bpf_xdp_store_bytes(xdp, offset, buf, len);
4043 }
4044
4045 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4046 {
4047         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4048         skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4049         struct xdp_rxq_info *rxq = xdp->rxq;
4050         unsigned int tailroom;
4051
4052         if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4053                 return -EOPNOTSUPP;
4054
4055         tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4056         if (unlikely(offset > tailroom))
4057                 return -EINVAL;
4058
4059         memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4060         skb_frag_size_add(frag, offset);
4061         sinfo->xdp_frags_size += offset;
4062
4063         return 0;
4064 }
4065
4066 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4067 {
4068         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4069         int i, n_frags_free = 0, len_free = 0;
4070
4071         if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4072                 return -EINVAL;
4073
4074         for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4075                 skb_frag_t *frag = &sinfo->frags[i];
4076                 int shrink = min_t(int, offset, skb_frag_size(frag));
4077
4078                 len_free += shrink;
4079                 offset -= shrink;
4080
4081                 if (skb_frag_size(frag) == shrink) {
4082                         struct page *page = skb_frag_page(frag);
4083
4084                         __xdp_return(page_address(page), &xdp->rxq->mem,
4085                                      false, NULL);
4086                         n_frags_free++;
4087                 } else {
4088                         skb_frag_size_sub(frag, shrink);
4089                         break;
4090                 }
4091         }
4092         sinfo->nr_frags -= n_frags_free;
4093         sinfo->xdp_frags_size -= len_free;
4094
4095         if (unlikely(!sinfo->nr_frags)) {
4096                 xdp_buff_clear_frags_flag(xdp);
4097                 xdp->data_end -= offset;
4098         }
4099
4100         return 0;
4101 }
4102
4103 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4104 {
4105         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4106         void *data_end = xdp->data_end + offset;
4107
4108         if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4109                 if (offset < 0)
4110                         return bpf_xdp_frags_shrink_tail(xdp, -offset);
4111
4112                 return bpf_xdp_frags_increase_tail(xdp, offset);
4113         }
4114
4115         /* Notice that xdp_data_hard_end have reserved some tailroom */
4116         if (unlikely(data_end > data_hard_end))
4117                 return -EINVAL;
4118
4119         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4120         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4121                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4122                 return -EINVAL;
4123         }
4124
4125         if (unlikely(data_end < xdp->data + ETH_HLEN))
4126                 return -EINVAL;
4127
4128         /* Clear memory area on grow, can contain uninit kernel memory */
4129         if (offset > 0)
4130                 memset(xdp->data_end, 0, offset);
4131
4132         xdp->data_end = data_end;
4133
4134         return 0;
4135 }
4136
4137 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4138         .func           = bpf_xdp_adjust_tail,
4139         .gpl_only       = false,
4140         .ret_type       = RET_INTEGER,
4141         .arg1_type      = ARG_PTR_TO_CTX,
4142         .arg2_type      = ARG_ANYTHING,
4143 };
4144
4145 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4146 {
4147         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4148         void *meta = xdp->data_meta + offset;
4149         unsigned long metalen = xdp->data - meta;
4150
4151         if (xdp_data_meta_unsupported(xdp))
4152                 return -ENOTSUPP;
4153         if (unlikely(meta < xdp_frame_end ||
4154                      meta > xdp->data))
4155                 return -EINVAL;
4156         if (unlikely(xdp_metalen_invalid(metalen)))
4157                 return -EACCES;
4158
4159         xdp->data_meta = meta;
4160
4161         return 0;
4162 }
4163
4164 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4165         .func           = bpf_xdp_adjust_meta,
4166         .gpl_only       = false,
4167         .ret_type       = RET_INTEGER,
4168         .arg1_type      = ARG_PTR_TO_CTX,
4169         .arg2_type      = ARG_ANYTHING,
4170 };
4171
4172 /**
4173  * DOC: xdp redirect
4174  *
4175  * XDP_REDIRECT works by a three-step process, implemented in the functions
4176  * below:
4177  *
4178  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4179  *    of the redirect and store it (along with some other metadata) in a per-CPU
4180  *    struct bpf_redirect_info.
4181  *
4182  * 2. When the program returns the XDP_REDIRECT return code, the driver will
4183  *    call xdp_do_redirect() which will use the information in struct
4184  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
4185  *    bulk queue structure.
4186  *
4187  * 3. Before exiting its NAPI poll loop, the driver will call
4188  *    xdp_do_flush(), which will flush all the different bulk queues,
4189  *    thus completing the redirect. Note that xdp_do_flush() must be
4190  *    called before napi_complete_done() in the driver, as the
4191  *    XDP_REDIRECT logic relies on being inside a single NAPI instance
4192  *    through to the xdp_do_flush() call for RCU protection of all
4193  *    in-kernel data structures.
4194  */
4195 /*
4196  * Pointers to the map entries will be kept around for this whole sequence of
4197  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4198  * the core code; instead, the RCU protection relies on everything happening
4199  * inside a single NAPI poll sequence, which means it's between a pair of calls
4200  * to local_bh_disable()/local_bh_enable().
4201  *
4202  * The map entries are marked as __rcu and the map code makes sure to
4203  * dereference those pointers with rcu_dereference_check() in a way that works
4204  * for both sections that to hold an rcu_read_lock() and sections that are
4205  * called from NAPI without a separate rcu_read_lock(). The code below does not
4206  * use RCU annotations, but relies on those in the map code.
4207  */
4208 void xdp_do_flush(void)
4209 {
4210         __dev_flush();
4211         __cpu_map_flush();
4212         __xsk_map_flush();
4213 }
4214 EXPORT_SYMBOL_GPL(xdp_do_flush);
4215
4216 void bpf_clear_redirect_map(struct bpf_map *map)
4217 {
4218         struct bpf_redirect_info *ri;
4219         int cpu;
4220
4221         for_each_possible_cpu(cpu) {
4222                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4223                 /* Avoid polluting remote cacheline due to writes if
4224                  * not needed. Once we pass this test, we need the
4225                  * cmpxchg() to make sure it hasn't been changed in
4226                  * the meantime by remote CPU.
4227                  */
4228                 if (unlikely(READ_ONCE(ri->map) == map))
4229                         cmpxchg(&ri->map, map, NULL);
4230         }
4231 }
4232
4233 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4234 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4235
4236 u32 xdp_master_redirect(struct xdp_buff *xdp)
4237 {
4238         struct net_device *master, *slave;
4239         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4240
4241         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4242         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4243         if (slave && slave != xdp->rxq->dev) {
4244                 /* The target device is different from the receiving device, so
4245                  * redirect it to the new device.
4246                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4247                  * drivers to unmap the packet from their rx ring.
4248                  */
4249                 ri->tgt_index = slave->ifindex;
4250                 ri->map_id = INT_MAX;
4251                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4252                 return XDP_REDIRECT;
4253         }
4254         return XDP_TX;
4255 }
4256 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4257
4258 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4259                                         struct net_device *dev,
4260                                         struct xdp_buff *xdp,
4261                                         struct bpf_prog *xdp_prog)
4262 {
4263         enum bpf_map_type map_type = ri->map_type;
4264         void *fwd = ri->tgt_value;
4265         u32 map_id = ri->map_id;
4266         int err;
4267
4268         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4269         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4270
4271         err = __xsk_map_redirect(fwd, xdp);
4272         if (unlikely(err))
4273                 goto err;
4274
4275         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4276         return 0;
4277 err:
4278         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4279         return err;
4280 }
4281
4282 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4283                                                    struct net_device *dev,
4284                                                    struct xdp_frame *xdpf,
4285                                                    struct bpf_prog *xdp_prog)
4286 {
4287         enum bpf_map_type map_type = ri->map_type;
4288         void *fwd = ri->tgt_value;
4289         u32 map_id = ri->map_id;
4290         struct bpf_map *map;
4291         int err;
4292
4293         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4294         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4295
4296         if (unlikely(!xdpf)) {
4297                 err = -EOVERFLOW;
4298                 goto err;
4299         }
4300
4301         switch (map_type) {
4302         case BPF_MAP_TYPE_DEVMAP:
4303                 fallthrough;
4304         case BPF_MAP_TYPE_DEVMAP_HASH:
4305                 map = READ_ONCE(ri->map);
4306                 if (unlikely(map)) {
4307                         WRITE_ONCE(ri->map, NULL);
4308                         err = dev_map_enqueue_multi(xdpf, dev, map,
4309                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
4310                 } else {
4311                         err = dev_map_enqueue(fwd, xdpf, dev);
4312                 }
4313                 break;
4314         case BPF_MAP_TYPE_CPUMAP:
4315                 err = cpu_map_enqueue(fwd, xdpf, dev);
4316                 break;
4317         case BPF_MAP_TYPE_UNSPEC:
4318                 if (map_id == INT_MAX) {
4319                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4320                         if (unlikely(!fwd)) {
4321                                 err = -EINVAL;
4322                                 break;
4323                         }
4324                         err = dev_xdp_enqueue(fwd, xdpf, dev);
4325                         break;
4326                 }
4327                 fallthrough;
4328         default:
4329                 err = -EBADRQC;
4330         }
4331
4332         if (unlikely(err))
4333                 goto err;
4334
4335         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4336         return 0;
4337 err:
4338         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4339         return err;
4340 }
4341
4342 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4343                     struct bpf_prog *xdp_prog)
4344 {
4345         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4346         enum bpf_map_type map_type = ri->map_type;
4347
4348         if (map_type == BPF_MAP_TYPE_XSKMAP)
4349                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4350
4351         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4352                                        xdp_prog);
4353 }
4354 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4355
4356 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4357                           struct xdp_frame *xdpf, 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, xdpf, xdp_prog);
4366 }
4367 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4368
4369 static int xdp_do_generic_redirect_map(struct net_device *dev,
4370                                        struct sk_buff *skb,
4371                                        struct xdp_buff *xdp,
4372                                        struct bpf_prog *xdp_prog,
4373                                        void *fwd,
4374                                        enum bpf_map_type map_type, u32 map_id)
4375 {
4376         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4377         struct bpf_map *map;
4378         int err;
4379
4380         switch (map_type) {
4381         case BPF_MAP_TYPE_DEVMAP:
4382                 fallthrough;
4383         case BPF_MAP_TYPE_DEVMAP_HASH:
4384                 map = READ_ONCE(ri->map);
4385                 if (unlikely(map)) {
4386                         WRITE_ONCE(ri->map, NULL);
4387                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4388                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4389                 } else {
4390                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4391                 }
4392                 if (unlikely(err))
4393                         goto err;
4394                 break;
4395         case BPF_MAP_TYPE_XSKMAP:
4396                 err = xsk_generic_rcv(fwd, xdp);
4397                 if (err)
4398                         goto err;
4399                 consume_skb(skb);
4400                 break;
4401         case BPF_MAP_TYPE_CPUMAP:
4402                 err = cpu_map_generic_redirect(fwd, skb);
4403                 if (unlikely(err))
4404                         goto err;
4405                 break;
4406         default:
4407                 err = -EBADRQC;
4408                 goto err;
4409         }
4410
4411         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4412         return 0;
4413 err:
4414         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4415         return err;
4416 }
4417
4418 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4419                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4420 {
4421         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4422         enum bpf_map_type map_type = ri->map_type;
4423         void *fwd = ri->tgt_value;
4424         u32 map_id = ri->map_id;
4425         int err;
4426
4427         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4428         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4429
4430         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4431                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4432                 if (unlikely(!fwd)) {
4433                         err = -EINVAL;
4434                         goto err;
4435                 }
4436
4437                 err = xdp_ok_fwd_dev(fwd, skb->len);
4438                 if (unlikely(err))
4439                         goto err;
4440
4441                 skb->dev = fwd;
4442                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4443                 generic_xdp_tx(skb, xdp_prog);
4444                 return 0;
4445         }
4446
4447         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4448 err:
4449         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4450         return err;
4451 }
4452
4453 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4454 {
4455         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4456
4457         if (unlikely(flags))
4458                 return XDP_ABORTED;
4459
4460         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4461          * by map_idr) is used for ifindex based XDP redirect.
4462          */
4463         ri->tgt_index = ifindex;
4464         ri->map_id = INT_MAX;
4465         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4466
4467         return XDP_REDIRECT;
4468 }
4469
4470 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4471         .func           = bpf_xdp_redirect,
4472         .gpl_only       = false,
4473         .ret_type       = RET_INTEGER,
4474         .arg1_type      = ARG_ANYTHING,
4475         .arg2_type      = ARG_ANYTHING,
4476 };
4477
4478 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4479            u64, flags)
4480 {
4481         return map->ops->map_redirect(map, key, flags);
4482 }
4483
4484 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4485         .func           = bpf_xdp_redirect_map,
4486         .gpl_only       = false,
4487         .ret_type       = RET_INTEGER,
4488         .arg1_type      = ARG_CONST_MAP_PTR,
4489         .arg2_type      = ARG_ANYTHING,
4490         .arg3_type      = ARG_ANYTHING,
4491 };
4492
4493 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4494                                   unsigned long off, unsigned long len)
4495 {
4496         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4497
4498         if (unlikely(!ptr))
4499                 return len;
4500         if (ptr != dst_buff)
4501                 memcpy(dst_buff, ptr, len);
4502
4503         return 0;
4504 }
4505
4506 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4507            u64, flags, void *, meta, u64, meta_size)
4508 {
4509         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4510
4511         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4512                 return -EINVAL;
4513         if (unlikely(!skb || skb_size > skb->len))
4514                 return -EFAULT;
4515
4516         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4517                                 bpf_skb_copy);
4518 }
4519
4520 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4521         .func           = bpf_skb_event_output,
4522         .gpl_only       = true,
4523         .ret_type       = RET_INTEGER,
4524         .arg1_type      = ARG_PTR_TO_CTX,
4525         .arg2_type      = ARG_CONST_MAP_PTR,
4526         .arg3_type      = ARG_ANYTHING,
4527         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4528         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4529 };
4530
4531 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4532
4533 const struct bpf_func_proto bpf_skb_output_proto = {
4534         .func           = bpf_skb_event_output,
4535         .gpl_only       = true,
4536         .ret_type       = RET_INTEGER,
4537         .arg1_type      = ARG_PTR_TO_BTF_ID,
4538         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
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 static unsigned short bpf_tunnel_key_af(u64 flags)
4546 {
4547         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4548 }
4549
4550 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4551            u32, size, u64, flags)
4552 {
4553         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4554         u8 compat[sizeof(struct bpf_tunnel_key)];
4555         void *to_orig = to;
4556         int err;
4557
4558         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4559                                          BPF_F_TUNINFO_FLAGS)))) {
4560                 err = -EINVAL;
4561                 goto err_clear;
4562         }
4563         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4564                 err = -EPROTO;
4565                 goto err_clear;
4566         }
4567         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4568                 err = -EINVAL;
4569                 switch (size) {
4570                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4571                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4572                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4573                         goto set_compat;
4574                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4575                         /* Fixup deprecated structure layouts here, so we have
4576                          * a common path later on.
4577                          */
4578                         if (ip_tunnel_info_af(info) != AF_INET)
4579                                 goto err_clear;
4580 set_compat:
4581                         to = (struct bpf_tunnel_key *)compat;
4582                         break;
4583                 default:
4584                         goto err_clear;
4585                 }
4586         }
4587
4588         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4589         to->tunnel_tos = info->key.tos;
4590         to->tunnel_ttl = info->key.ttl;
4591         if (flags & BPF_F_TUNINFO_FLAGS)
4592                 to->tunnel_flags = info->key.tun_flags;
4593         else
4594                 to->tunnel_ext = 0;
4595
4596         if (flags & BPF_F_TUNINFO_IPV6) {
4597                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4598                        sizeof(to->remote_ipv6));
4599                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4600                        sizeof(to->local_ipv6));
4601                 to->tunnel_label = be32_to_cpu(info->key.label);
4602         } else {
4603                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4604                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4605                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4606                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4607                 to->tunnel_label = 0;
4608         }
4609
4610         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4611                 memcpy(to_orig, to, size);
4612
4613         return 0;
4614 err_clear:
4615         memset(to_orig, 0, size);
4616         return err;
4617 }
4618
4619 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4620         .func           = bpf_skb_get_tunnel_key,
4621         .gpl_only       = false,
4622         .ret_type       = RET_INTEGER,
4623         .arg1_type      = ARG_PTR_TO_CTX,
4624         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4625         .arg3_type      = ARG_CONST_SIZE,
4626         .arg4_type      = ARG_ANYTHING,
4627 };
4628
4629 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4630 {
4631         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4632         int err;
4633
4634         if (unlikely(!info ||
4635                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4636                 err = -ENOENT;
4637                 goto err_clear;
4638         }
4639         if (unlikely(size < info->options_len)) {
4640                 err = -ENOMEM;
4641                 goto err_clear;
4642         }
4643
4644         ip_tunnel_info_opts_get(to, info);
4645         if (size > info->options_len)
4646                 memset(to + info->options_len, 0, size - info->options_len);
4647
4648         return info->options_len;
4649 err_clear:
4650         memset(to, 0, size);
4651         return err;
4652 }
4653
4654 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4655         .func           = bpf_skb_get_tunnel_opt,
4656         .gpl_only       = false,
4657         .ret_type       = RET_INTEGER,
4658         .arg1_type      = ARG_PTR_TO_CTX,
4659         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4660         .arg3_type      = ARG_CONST_SIZE,
4661 };
4662
4663 static struct metadata_dst __percpu *md_dst;
4664
4665 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4666            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4667 {
4668         struct metadata_dst *md = this_cpu_ptr(md_dst);
4669         u8 compat[sizeof(struct bpf_tunnel_key)];
4670         struct ip_tunnel_info *info;
4671
4672         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4673                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4674                                BPF_F_NO_TUNNEL_KEY)))
4675                 return -EINVAL;
4676         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4677                 switch (size) {
4678                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4679                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4680                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4681                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4682                         /* Fixup deprecated structure layouts here, so we have
4683                          * a common path later on.
4684                          */
4685                         memcpy(compat, from, size);
4686                         memset(compat + size, 0, sizeof(compat) - size);
4687                         from = (const struct bpf_tunnel_key *) compat;
4688                         break;
4689                 default:
4690                         return -EINVAL;
4691                 }
4692         }
4693         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4694                      from->tunnel_ext))
4695                 return -EINVAL;
4696
4697         skb_dst_drop(skb);
4698         dst_hold((struct dst_entry *) md);
4699         skb_dst_set(skb, (struct dst_entry *) md);
4700
4701         info = &md->u.tun_info;
4702         memset(info, 0, sizeof(*info));
4703         info->mode = IP_TUNNEL_INFO_TX;
4704
4705         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4706         if (flags & BPF_F_DONT_FRAGMENT)
4707                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4708         if (flags & BPF_F_ZERO_CSUM_TX)
4709                 info->key.tun_flags &= ~TUNNEL_CSUM;
4710         if (flags & BPF_F_SEQ_NUMBER)
4711                 info->key.tun_flags |= TUNNEL_SEQ;
4712         if (flags & BPF_F_NO_TUNNEL_KEY)
4713                 info->key.tun_flags &= ~TUNNEL_KEY;
4714
4715         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4716         info->key.tos = from->tunnel_tos;
4717         info->key.ttl = from->tunnel_ttl;
4718
4719         if (flags & BPF_F_TUNINFO_IPV6) {
4720                 info->mode |= IP_TUNNEL_INFO_IPV6;
4721                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4722                        sizeof(from->remote_ipv6));
4723                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4724                        sizeof(from->local_ipv6));
4725                 info->key.label = cpu_to_be32(from->tunnel_label) &
4726                                   IPV6_FLOWLABEL_MASK;
4727         } else {
4728                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4729                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4730                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4731         }
4732
4733         return 0;
4734 }
4735
4736 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4737         .func           = bpf_skb_set_tunnel_key,
4738         .gpl_only       = false,
4739         .ret_type       = RET_INTEGER,
4740         .arg1_type      = ARG_PTR_TO_CTX,
4741         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4742         .arg3_type      = ARG_CONST_SIZE,
4743         .arg4_type      = ARG_ANYTHING,
4744 };
4745
4746 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4747            const u8 *, from, u32, size)
4748 {
4749         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4750         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4751
4752         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4753                 return -EINVAL;
4754         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4755                 return -ENOMEM;
4756
4757         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4758
4759         return 0;
4760 }
4761
4762 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4763         .func           = bpf_skb_set_tunnel_opt,
4764         .gpl_only       = false,
4765         .ret_type       = RET_INTEGER,
4766         .arg1_type      = ARG_PTR_TO_CTX,
4767         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4768         .arg3_type      = ARG_CONST_SIZE,
4769 };
4770
4771 static const struct bpf_func_proto *
4772 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4773 {
4774         if (!md_dst) {
4775                 struct metadata_dst __percpu *tmp;
4776
4777                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4778                                                 METADATA_IP_TUNNEL,
4779                                                 GFP_KERNEL);
4780                 if (!tmp)
4781                         return NULL;
4782                 if (cmpxchg(&md_dst, NULL, tmp))
4783                         metadata_dst_free_percpu(tmp);
4784         }
4785
4786         switch (which) {
4787         case BPF_FUNC_skb_set_tunnel_key:
4788                 return &bpf_skb_set_tunnel_key_proto;
4789         case BPF_FUNC_skb_set_tunnel_opt:
4790                 return &bpf_skb_set_tunnel_opt_proto;
4791         default:
4792                 return NULL;
4793         }
4794 }
4795
4796 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4797            u32, idx)
4798 {
4799         struct bpf_array *array = container_of(map, struct bpf_array, map);
4800         struct cgroup *cgrp;
4801         struct sock *sk;
4802
4803         sk = skb_to_full_sk(skb);
4804         if (!sk || !sk_fullsock(sk))
4805                 return -ENOENT;
4806         if (unlikely(idx >= array->map.max_entries))
4807                 return -E2BIG;
4808
4809         cgrp = READ_ONCE(array->ptrs[idx]);
4810         if (unlikely(!cgrp))
4811                 return -EAGAIN;
4812
4813         return sk_under_cgroup_hierarchy(sk, cgrp);
4814 }
4815
4816 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4817         .func           = bpf_skb_under_cgroup,
4818         .gpl_only       = false,
4819         .ret_type       = RET_INTEGER,
4820         .arg1_type      = ARG_PTR_TO_CTX,
4821         .arg2_type      = ARG_CONST_MAP_PTR,
4822         .arg3_type      = ARG_ANYTHING,
4823 };
4824
4825 #ifdef CONFIG_SOCK_CGROUP_DATA
4826 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4827 {
4828         struct cgroup *cgrp;
4829
4830         sk = sk_to_full_sk(sk);
4831         if (!sk || !sk_fullsock(sk))
4832                 return 0;
4833
4834         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4835         return cgroup_id(cgrp);
4836 }
4837
4838 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4839 {
4840         return __bpf_sk_cgroup_id(skb->sk);
4841 }
4842
4843 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4844         .func           = bpf_skb_cgroup_id,
4845         .gpl_only       = false,
4846         .ret_type       = RET_INTEGER,
4847         .arg1_type      = ARG_PTR_TO_CTX,
4848 };
4849
4850 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4851                                               int ancestor_level)
4852 {
4853         struct cgroup *ancestor;
4854         struct cgroup *cgrp;
4855
4856         sk = sk_to_full_sk(sk);
4857         if (!sk || !sk_fullsock(sk))
4858                 return 0;
4859
4860         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4861         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4862         if (!ancestor)
4863                 return 0;
4864
4865         return cgroup_id(ancestor);
4866 }
4867
4868 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4869            ancestor_level)
4870 {
4871         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4872 }
4873
4874 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4875         .func           = bpf_skb_ancestor_cgroup_id,
4876         .gpl_only       = false,
4877         .ret_type       = RET_INTEGER,
4878         .arg1_type      = ARG_PTR_TO_CTX,
4879         .arg2_type      = ARG_ANYTHING,
4880 };
4881
4882 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4883 {
4884         return __bpf_sk_cgroup_id(sk);
4885 }
4886
4887 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4888         .func           = bpf_sk_cgroup_id,
4889         .gpl_only       = false,
4890         .ret_type       = RET_INTEGER,
4891         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4892 };
4893
4894 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4895 {
4896         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4897 }
4898
4899 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4900         .func           = bpf_sk_ancestor_cgroup_id,
4901         .gpl_only       = false,
4902         .ret_type       = RET_INTEGER,
4903         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4904         .arg2_type      = ARG_ANYTHING,
4905 };
4906 #endif
4907
4908 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4909                                   unsigned long off, unsigned long len)
4910 {
4911         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4912
4913         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4914         return 0;
4915 }
4916
4917 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4918            u64, flags, void *, meta, u64, meta_size)
4919 {
4920         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4921
4922         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4923                 return -EINVAL;
4924
4925         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4926                 return -EFAULT;
4927
4928         return bpf_event_output(map, flags, meta, meta_size, xdp,
4929                                 xdp_size, bpf_xdp_copy);
4930 }
4931
4932 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4933         .func           = bpf_xdp_event_output,
4934         .gpl_only       = true,
4935         .ret_type       = RET_INTEGER,
4936         .arg1_type      = ARG_PTR_TO_CTX,
4937         .arg2_type      = ARG_CONST_MAP_PTR,
4938         .arg3_type      = ARG_ANYTHING,
4939         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4940         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4941 };
4942
4943 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4944
4945 const struct bpf_func_proto bpf_xdp_output_proto = {
4946         .func           = bpf_xdp_event_output,
4947         .gpl_only       = true,
4948         .ret_type       = RET_INTEGER,
4949         .arg1_type      = ARG_PTR_TO_BTF_ID,
4950         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
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 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4958 {
4959         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4960 }
4961
4962 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4963         .func           = bpf_get_socket_cookie,
4964         .gpl_only       = false,
4965         .ret_type       = RET_INTEGER,
4966         .arg1_type      = ARG_PTR_TO_CTX,
4967 };
4968
4969 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4970 {
4971         return __sock_gen_cookie(ctx->sk);
4972 }
4973
4974 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4975         .func           = bpf_get_socket_cookie_sock_addr,
4976         .gpl_only       = false,
4977         .ret_type       = RET_INTEGER,
4978         .arg1_type      = ARG_PTR_TO_CTX,
4979 };
4980
4981 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4982 {
4983         return __sock_gen_cookie(ctx);
4984 }
4985
4986 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4987         .func           = bpf_get_socket_cookie_sock,
4988         .gpl_only       = false,
4989         .ret_type       = RET_INTEGER,
4990         .arg1_type      = ARG_PTR_TO_CTX,
4991 };
4992
4993 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4994 {
4995         return sk ? sock_gen_cookie(sk) : 0;
4996 }
4997
4998 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4999         .func           = bpf_get_socket_ptr_cookie,
5000         .gpl_only       = false,
5001         .ret_type       = RET_INTEGER,
5002         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5003 };
5004
5005 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5006 {
5007         return __sock_gen_cookie(ctx->sk);
5008 }
5009
5010 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5011         .func           = bpf_get_socket_cookie_sock_ops,
5012         .gpl_only       = false,
5013         .ret_type       = RET_INTEGER,
5014         .arg1_type      = ARG_PTR_TO_CTX,
5015 };
5016
5017 static u64 __bpf_get_netns_cookie(struct sock *sk)
5018 {
5019         const struct net *net = sk ? sock_net(sk) : &init_net;
5020
5021         return net->net_cookie;
5022 }
5023
5024 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5025 {
5026         return __bpf_get_netns_cookie(ctx);
5027 }
5028
5029 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5030         .func           = bpf_get_netns_cookie_sock,
5031         .gpl_only       = false,
5032         .ret_type       = RET_INTEGER,
5033         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5034 };
5035
5036 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5037 {
5038         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5039 }
5040
5041 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5042         .func           = bpf_get_netns_cookie_sock_addr,
5043         .gpl_only       = false,
5044         .ret_type       = RET_INTEGER,
5045         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5046 };
5047
5048 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5049 {
5050         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5051 }
5052
5053 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5054         .func           = bpf_get_netns_cookie_sock_ops,
5055         .gpl_only       = false,
5056         .ret_type       = RET_INTEGER,
5057         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5058 };
5059
5060 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5061 {
5062         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5063 }
5064
5065 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5066         .func           = bpf_get_netns_cookie_sk_msg,
5067         .gpl_only       = false,
5068         .ret_type       = RET_INTEGER,
5069         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5070 };
5071
5072 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5073 {
5074         struct sock *sk = sk_to_full_sk(skb->sk);
5075         kuid_t kuid;
5076
5077         if (!sk || !sk_fullsock(sk))
5078                 return overflowuid;
5079         kuid = sock_net_uid(sock_net(sk), sk);
5080         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5081 }
5082
5083 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5084         .func           = bpf_get_socket_uid,
5085         .gpl_only       = false,
5086         .ret_type       = RET_INTEGER,
5087         .arg1_type      = ARG_PTR_TO_CTX,
5088 };
5089
5090 static int sol_socket_sockopt(struct sock *sk, int optname,
5091                               char *optval, int *optlen,
5092                               bool getopt)
5093 {
5094         switch (optname) {
5095         case SO_REUSEADDR:
5096         case SO_SNDBUF:
5097         case SO_RCVBUF:
5098         case SO_KEEPALIVE:
5099         case SO_PRIORITY:
5100         case SO_REUSEPORT:
5101         case SO_RCVLOWAT:
5102         case SO_MARK:
5103         case SO_MAX_PACING_RATE:
5104         case SO_BINDTOIFINDEX:
5105         case SO_TXREHASH:
5106                 if (*optlen != sizeof(int))
5107                         return -EINVAL;
5108                 break;
5109         case SO_BINDTODEVICE:
5110                 break;
5111         default:
5112                 return -EINVAL;
5113         }
5114
5115         if (getopt) {
5116                 if (optname == SO_BINDTODEVICE)
5117                         return -EINVAL;
5118                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5119                                      KERNEL_SOCKPTR(optval),
5120                                      KERNEL_SOCKPTR(optlen));
5121         }
5122
5123         return sk_setsockopt(sk, SOL_SOCKET, optname,
5124                              KERNEL_SOCKPTR(optval), *optlen);
5125 }
5126
5127 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5128                                   char *optval, int optlen)
5129 {
5130         struct tcp_sock *tp = tcp_sk(sk);
5131         unsigned long timeout;
5132         int val;
5133
5134         if (optlen != sizeof(int))
5135                 return -EINVAL;
5136
5137         val = *(int *)optval;
5138
5139         /* Only some options are supported */
5140         switch (optname) {
5141         case TCP_BPF_IW:
5142                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5143                         return -EINVAL;
5144                 tcp_snd_cwnd_set(tp, val);
5145                 break;
5146         case TCP_BPF_SNDCWND_CLAMP:
5147                 if (val <= 0)
5148                         return -EINVAL;
5149                 tp->snd_cwnd_clamp = val;
5150                 tp->snd_ssthresh = val;
5151                 break;
5152         case TCP_BPF_DELACK_MAX:
5153                 timeout = usecs_to_jiffies(val);
5154                 if (timeout > TCP_DELACK_MAX ||
5155                     timeout < TCP_TIMEOUT_MIN)
5156                         return -EINVAL;
5157                 inet_csk(sk)->icsk_delack_max = timeout;
5158                 break;
5159         case TCP_BPF_RTO_MIN:
5160                 timeout = usecs_to_jiffies(val);
5161                 if (timeout > TCP_RTO_MIN ||
5162                     timeout < TCP_TIMEOUT_MIN)
5163                         return -EINVAL;
5164                 inet_csk(sk)->icsk_rto_min = timeout;
5165                 break;
5166         default:
5167                 return -EINVAL;
5168         }
5169
5170         return 0;
5171 }
5172
5173 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5174                                       int *optlen, bool getopt)
5175 {
5176         struct tcp_sock *tp;
5177         int ret;
5178
5179         if (*optlen < 2)
5180                 return -EINVAL;
5181
5182         if (getopt) {
5183                 if (!inet_csk(sk)->icsk_ca_ops)
5184                         return -EINVAL;
5185                 /* BPF expects NULL-terminated tcp-cc string */
5186                 optval[--(*optlen)] = '\0';
5187                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5188                                          KERNEL_SOCKPTR(optval),
5189                                          KERNEL_SOCKPTR(optlen));
5190         }
5191
5192         /* "cdg" is the only cc that alloc a ptr
5193          * in inet_csk_ca area.  The bpf-tcp-cc may
5194          * overwrite this ptr after switching to cdg.
5195          */
5196         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5197                 return -ENOTSUPP;
5198
5199         /* It stops this looping
5200          *
5201          * .init => bpf_setsockopt(tcp_cc) => .init =>
5202          * bpf_setsockopt(tcp_cc)" => .init => ....
5203          *
5204          * The second bpf_setsockopt(tcp_cc) is not allowed
5205          * in order to break the loop when both .init
5206          * are the same bpf prog.
5207          *
5208          * This applies even the second bpf_setsockopt(tcp_cc)
5209          * does not cause a loop.  This limits only the first
5210          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5211          * pick a fallback cc (eg. peer does not support ECN)
5212          * and the second '.init' cannot fallback to
5213          * another.
5214          */
5215         tp = tcp_sk(sk);
5216         if (tp->bpf_chg_cc_inprogress)
5217                 return -EBUSY;
5218
5219         tp->bpf_chg_cc_inprogress = 1;
5220         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5221                                 KERNEL_SOCKPTR(optval), *optlen);
5222         tp->bpf_chg_cc_inprogress = 0;
5223         return ret;
5224 }
5225
5226 static int sol_tcp_sockopt(struct sock *sk, int optname,
5227                            char *optval, int *optlen,
5228                            bool getopt)
5229 {
5230         if (sk->sk_protocol != IPPROTO_TCP)
5231                 return -EINVAL;
5232
5233         switch (optname) {
5234         case TCP_NODELAY:
5235         case TCP_MAXSEG:
5236         case TCP_KEEPIDLE:
5237         case TCP_KEEPINTVL:
5238         case TCP_KEEPCNT:
5239         case TCP_SYNCNT:
5240         case TCP_WINDOW_CLAMP:
5241         case TCP_THIN_LINEAR_TIMEOUTS:
5242         case TCP_USER_TIMEOUT:
5243         case TCP_NOTSENT_LOWAT:
5244         case TCP_SAVE_SYN:
5245                 if (*optlen != sizeof(int))
5246                         return -EINVAL;
5247                 break;
5248         case TCP_CONGESTION:
5249                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5250         case TCP_SAVED_SYN:
5251                 if (*optlen < 1)
5252                         return -EINVAL;
5253                 break;
5254         default:
5255                 if (getopt)
5256                         return -EINVAL;
5257                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5258         }
5259
5260         if (getopt) {
5261                 if (optname == TCP_SAVED_SYN) {
5262                         struct tcp_sock *tp = tcp_sk(sk);
5263
5264                         if (!tp->saved_syn ||
5265                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5266                                 return -EINVAL;
5267                         memcpy(optval, tp->saved_syn->data, *optlen);
5268                         /* It cannot free tp->saved_syn here because it
5269                          * does not know if the user space still needs it.
5270                          */
5271                         return 0;
5272                 }
5273
5274                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5275                                          KERNEL_SOCKPTR(optval),
5276                                          KERNEL_SOCKPTR(optlen));
5277         }
5278
5279         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5280                                  KERNEL_SOCKPTR(optval), *optlen);
5281 }
5282
5283 static int sol_ip_sockopt(struct sock *sk, int optname,
5284                           char *optval, int *optlen,
5285                           bool getopt)
5286 {
5287         if (sk->sk_family != AF_INET)
5288                 return -EINVAL;
5289
5290         switch (optname) {
5291         case IP_TOS:
5292                 if (*optlen != sizeof(int))
5293                         return -EINVAL;
5294                 break;
5295         default:
5296                 return -EINVAL;
5297         }
5298
5299         if (getopt)
5300                 return do_ip_getsockopt(sk, SOL_IP, optname,
5301                                         KERNEL_SOCKPTR(optval),
5302                                         KERNEL_SOCKPTR(optlen));
5303
5304         return do_ip_setsockopt(sk, SOL_IP, optname,
5305                                 KERNEL_SOCKPTR(optval), *optlen);
5306 }
5307
5308 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5309                             char *optval, int *optlen,
5310                             bool getopt)
5311 {
5312         if (sk->sk_family != AF_INET6)
5313                 return -EINVAL;
5314
5315         switch (optname) {
5316         case IPV6_TCLASS:
5317         case IPV6_AUTOFLOWLABEL:
5318                 if (*optlen != sizeof(int))
5319                         return -EINVAL;
5320                 break;
5321         default:
5322                 return -EINVAL;
5323         }
5324
5325         if (getopt)
5326                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5327                                                       KERNEL_SOCKPTR(optval),
5328                                                       KERNEL_SOCKPTR(optlen));
5329
5330         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5331                                               KERNEL_SOCKPTR(optval), *optlen);
5332 }
5333
5334 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5335                             char *optval, int optlen)
5336 {
5337         if (!sk_fullsock(sk))
5338                 return -EINVAL;
5339
5340         if (level == SOL_SOCKET)
5341                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5342         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5343                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5344         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5345                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5346         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5347                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5348
5349         return -EINVAL;
5350 }
5351
5352 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5353                            char *optval, int optlen)
5354 {
5355         if (sk_fullsock(sk))
5356                 sock_owned_by_me(sk);
5357         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5358 }
5359
5360 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5361                             char *optval, int optlen)
5362 {
5363         int err, saved_optlen = optlen;
5364
5365         if (!sk_fullsock(sk)) {
5366                 err = -EINVAL;
5367                 goto done;
5368         }
5369
5370         if (level == SOL_SOCKET)
5371                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5372         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5373                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5374         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5375                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5376         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5377                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5378         else
5379                 err = -EINVAL;
5380
5381 done:
5382         if (err)
5383                 optlen = 0;
5384         if (optlen < saved_optlen)
5385                 memset(optval + optlen, 0, saved_optlen - optlen);
5386         return err;
5387 }
5388
5389 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5390                            char *optval, int optlen)
5391 {
5392         if (sk_fullsock(sk))
5393                 sock_owned_by_me(sk);
5394         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5395 }
5396
5397 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5398            int, optname, char *, optval, int, optlen)
5399 {
5400         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5401 }
5402
5403 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5404         .func           = bpf_sk_setsockopt,
5405         .gpl_only       = false,
5406         .ret_type       = RET_INTEGER,
5407         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5408         .arg2_type      = ARG_ANYTHING,
5409         .arg3_type      = ARG_ANYTHING,
5410         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5411         .arg5_type      = ARG_CONST_SIZE,
5412 };
5413
5414 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5415            int, optname, char *, optval, int, optlen)
5416 {
5417         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5418 }
5419
5420 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5421         .func           = bpf_sk_getsockopt,
5422         .gpl_only       = false,
5423         .ret_type       = RET_INTEGER,
5424         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5425         .arg2_type      = ARG_ANYTHING,
5426         .arg3_type      = ARG_ANYTHING,
5427         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5428         .arg5_type      = ARG_CONST_SIZE,
5429 };
5430
5431 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5432            int, optname, char *, optval, int, optlen)
5433 {
5434         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5435 }
5436
5437 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5438         .func           = bpf_unlocked_sk_setsockopt,
5439         .gpl_only       = false,
5440         .ret_type       = RET_INTEGER,
5441         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5442         .arg2_type      = ARG_ANYTHING,
5443         .arg3_type      = ARG_ANYTHING,
5444         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5445         .arg5_type      = ARG_CONST_SIZE,
5446 };
5447
5448 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5449            int, optname, char *, optval, int, optlen)
5450 {
5451         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5452 }
5453
5454 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5455         .func           = bpf_unlocked_sk_getsockopt,
5456         .gpl_only       = false,
5457         .ret_type       = RET_INTEGER,
5458         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5459         .arg2_type      = ARG_ANYTHING,
5460         .arg3_type      = ARG_ANYTHING,
5461         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5462         .arg5_type      = ARG_CONST_SIZE,
5463 };
5464
5465 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5466            int, level, int, optname, char *, optval, int, optlen)
5467 {
5468         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5469 }
5470
5471 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5472         .func           = bpf_sock_addr_setsockopt,
5473         .gpl_only       = false,
5474         .ret_type       = RET_INTEGER,
5475         .arg1_type      = ARG_PTR_TO_CTX,
5476         .arg2_type      = ARG_ANYTHING,
5477         .arg3_type      = ARG_ANYTHING,
5478         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5479         .arg5_type      = ARG_CONST_SIZE,
5480 };
5481
5482 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5483            int, level, int, optname, char *, optval, int, optlen)
5484 {
5485         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5486 }
5487
5488 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5489         .func           = bpf_sock_addr_getsockopt,
5490         .gpl_only       = false,
5491         .ret_type       = RET_INTEGER,
5492         .arg1_type      = ARG_PTR_TO_CTX,
5493         .arg2_type      = ARG_ANYTHING,
5494         .arg3_type      = ARG_ANYTHING,
5495         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5496         .arg5_type      = ARG_CONST_SIZE,
5497 };
5498
5499 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5500            int, level, int, optname, char *, optval, int, optlen)
5501 {
5502         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5503 }
5504
5505 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5506         .func           = bpf_sock_ops_setsockopt,
5507         .gpl_only       = false,
5508         .ret_type       = RET_INTEGER,
5509         .arg1_type      = ARG_PTR_TO_CTX,
5510         .arg2_type      = ARG_ANYTHING,
5511         .arg3_type      = ARG_ANYTHING,
5512         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5513         .arg5_type      = ARG_CONST_SIZE,
5514 };
5515
5516 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5517                                 int optname, const u8 **start)
5518 {
5519         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5520         const u8 *hdr_start;
5521         int ret;
5522
5523         if (syn_skb) {
5524                 /* sk is a request_sock here */
5525
5526                 if (optname == TCP_BPF_SYN) {
5527                         hdr_start = syn_skb->data;
5528                         ret = tcp_hdrlen(syn_skb);
5529                 } else if (optname == TCP_BPF_SYN_IP) {
5530                         hdr_start = skb_network_header(syn_skb);
5531                         ret = skb_network_header_len(syn_skb) +
5532                                 tcp_hdrlen(syn_skb);
5533                 } else {
5534                         /* optname == TCP_BPF_SYN_MAC */
5535                         hdr_start = skb_mac_header(syn_skb);
5536                         ret = skb_mac_header_len(syn_skb) +
5537                                 skb_network_header_len(syn_skb) +
5538                                 tcp_hdrlen(syn_skb);
5539                 }
5540         } else {
5541                 struct sock *sk = bpf_sock->sk;
5542                 struct saved_syn *saved_syn;
5543
5544                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5545                         /* synack retransmit. bpf_sock->syn_skb will
5546                          * not be available.  It has to resort to
5547                          * saved_syn (if it is saved).
5548                          */
5549                         saved_syn = inet_reqsk(sk)->saved_syn;
5550                 else
5551                         saved_syn = tcp_sk(sk)->saved_syn;
5552
5553                 if (!saved_syn)
5554                         return -ENOENT;
5555
5556                 if (optname == TCP_BPF_SYN) {
5557                         hdr_start = saved_syn->data +
5558                                 saved_syn->mac_hdrlen +
5559                                 saved_syn->network_hdrlen;
5560                         ret = saved_syn->tcp_hdrlen;
5561                 } else if (optname == TCP_BPF_SYN_IP) {
5562                         hdr_start = saved_syn->data +
5563                                 saved_syn->mac_hdrlen;
5564                         ret = saved_syn->network_hdrlen +
5565                                 saved_syn->tcp_hdrlen;
5566                 } else {
5567                         /* optname == TCP_BPF_SYN_MAC */
5568
5569                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5570                         if (!saved_syn->mac_hdrlen)
5571                                 return -ENOENT;
5572
5573                         hdr_start = saved_syn->data;
5574                         ret = saved_syn->mac_hdrlen +
5575                                 saved_syn->network_hdrlen +
5576                                 saved_syn->tcp_hdrlen;
5577                 }
5578         }
5579
5580         *start = hdr_start;
5581         return ret;
5582 }
5583
5584 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5585            int, level, int, optname, char *, optval, int, optlen)
5586 {
5587         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5588             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5589                 int ret, copy_len = 0;
5590                 const u8 *start;
5591
5592                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5593                 if (ret > 0) {
5594                         copy_len = ret;
5595                         if (optlen < copy_len) {
5596                                 copy_len = optlen;
5597                                 ret = -ENOSPC;
5598                         }
5599
5600                         memcpy(optval, start, copy_len);
5601                 }
5602
5603                 /* Zero out unused buffer at the end */
5604                 memset(optval + copy_len, 0, optlen - copy_len);
5605
5606                 return ret;
5607         }
5608
5609         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5610 }
5611
5612 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5613         .func           = bpf_sock_ops_getsockopt,
5614         .gpl_only       = false,
5615         .ret_type       = RET_INTEGER,
5616         .arg1_type      = ARG_PTR_TO_CTX,
5617         .arg2_type      = ARG_ANYTHING,
5618         .arg3_type      = ARG_ANYTHING,
5619         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5620         .arg5_type      = ARG_CONST_SIZE,
5621 };
5622
5623 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5624            int, argval)
5625 {
5626         struct sock *sk = bpf_sock->sk;
5627         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5628
5629         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5630                 return -EINVAL;
5631
5632         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5633
5634         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5635 }
5636
5637 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5638         .func           = bpf_sock_ops_cb_flags_set,
5639         .gpl_only       = false,
5640         .ret_type       = RET_INTEGER,
5641         .arg1_type      = ARG_PTR_TO_CTX,
5642         .arg2_type      = ARG_ANYTHING,
5643 };
5644
5645 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5646 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5647
5648 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5649            int, addr_len)
5650 {
5651 #ifdef CONFIG_INET
5652         struct sock *sk = ctx->sk;
5653         u32 flags = BIND_FROM_BPF;
5654         int err;
5655
5656         err = -EINVAL;
5657         if (addr_len < offsetofend(struct sockaddr, sa_family))
5658                 return err;
5659         if (addr->sa_family == AF_INET) {
5660                 if (addr_len < sizeof(struct sockaddr_in))
5661                         return err;
5662                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5663                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5664                 return __inet_bind(sk, addr, addr_len, flags);
5665 #if IS_ENABLED(CONFIG_IPV6)
5666         } else if (addr->sa_family == AF_INET6) {
5667                 if (addr_len < SIN6_LEN_RFC2133)
5668                         return err;
5669                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5670                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5671                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5672                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5673                  */
5674                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5675 #endif /* CONFIG_IPV6 */
5676         }
5677 #endif /* CONFIG_INET */
5678
5679         return -EAFNOSUPPORT;
5680 }
5681
5682 static const struct bpf_func_proto bpf_bind_proto = {
5683         .func           = bpf_bind,
5684         .gpl_only       = false,
5685         .ret_type       = RET_INTEGER,
5686         .arg1_type      = ARG_PTR_TO_CTX,
5687         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5688         .arg3_type      = ARG_CONST_SIZE,
5689 };
5690
5691 #ifdef CONFIG_XFRM
5692
5693 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5694     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5695
5696 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5697 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5698
5699 #endif
5700
5701 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5702            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5703 {
5704         const struct sec_path *sp = skb_sec_path(skb);
5705         const struct xfrm_state *x;
5706
5707         if (!sp || unlikely(index >= sp->len || flags))
5708                 goto err_clear;
5709
5710         x = sp->xvec[index];
5711
5712         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5713                 goto err_clear;
5714
5715         to->reqid = x->props.reqid;
5716         to->spi = x->id.spi;
5717         to->family = x->props.family;
5718         to->ext = 0;
5719
5720         if (to->family == AF_INET6) {
5721                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5722                        sizeof(to->remote_ipv6));
5723         } else {
5724                 to->remote_ipv4 = x->props.saddr.a4;
5725                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5726         }
5727
5728         return 0;
5729 err_clear:
5730         memset(to, 0, size);
5731         return -EINVAL;
5732 }
5733
5734 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5735         .func           = bpf_skb_get_xfrm_state,
5736         .gpl_only       = false,
5737         .ret_type       = RET_INTEGER,
5738         .arg1_type      = ARG_PTR_TO_CTX,
5739         .arg2_type      = ARG_ANYTHING,
5740         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5741         .arg4_type      = ARG_CONST_SIZE,
5742         .arg5_type      = ARG_ANYTHING,
5743 };
5744 #endif
5745
5746 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5747 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5748 {
5749         params->h_vlan_TCI = 0;
5750         params->h_vlan_proto = 0;
5751         if (mtu)
5752                 params->mtu_result = mtu; /* union with tot_len */
5753
5754         return 0;
5755 }
5756 #endif
5757
5758 #if IS_ENABLED(CONFIG_INET)
5759 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5760                                u32 flags, bool check_mtu)
5761 {
5762         struct fib_nh_common *nhc;
5763         struct in_device *in_dev;
5764         struct neighbour *neigh;
5765         struct net_device *dev;
5766         struct fib_result res;
5767         struct flowi4 fl4;
5768         u32 mtu = 0;
5769         int err;
5770
5771         dev = dev_get_by_index_rcu(net, params->ifindex);
5772         if (unlikely(!dev))
5773                 return -ENODEV;
5774
5775         /* verify forwarding is enabled on this interface */
5776         in_dev = __in_dev_get_rcu(dev);
5777         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5778                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5779
5780         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5781                 fl4.flowi4_iif = 1;
5782                 fl4.flowi4_oif = params->ifindex;
5783         } else {
5784                 fl4.flowi4_iif = params->ifindex;
5785                 fl4.flowi4_oif = 0;
5786         }
5787         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5788         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5789         fl4.flowi4_flags = 0;
5790
5791         fl4.flowi4_proto = params->l4_protocol;
5792         fl4.daddr = params->ipv4_dst;
5793         fl4.saddr = params->ipv4_src;
5794         fl4.fl4_sport = params->sport;
5795         fl4.fl4_dport = params->dport;
5796         fl4.flowi4_multipath_hash = 0;
5797
5798         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5799                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5800                 struct fib_table *tb;
5801
5802                 if (flags & BPF_FIB_LOOKUP_TBID) {
5803                         tbid = params->tbid;
5804                         /* zero out for vlan output */
5805                         params->tbid = 0;
5806                 }
5807
5808                 tb = fib_get_table(net, tbid);
5809                 if (unlikely(!tb))
5810                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5811
5812                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5813         } else {
5814                 fl4.flowi4_mark = 0;
5815                 fl4.flowi4_secid = 0;
5816                 fl4.flowi4_tun_key.tun_id = 0;
5817                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5818
5819                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5820         }
5821
5822         if (err) {
5823                 /* map fib lookup errors to RTN_ type */
5824                 if (err == -EINVAL)
5825                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5826                 if (err == -EHOSTUNREACH)
5827                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5828                 if (err == -EACCES)
5829                         return BPF_FIB_LKUP_RET_PROHIBIT;
5830
5831                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5832         }
5833
5834         if (res.type != RTN_UNICAST)
5835                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5836
5837         if (fib_info_num_path(res.fi) > 1)
5838                 fib_select_path(net, &res, &fl4, NULL);
5839
5840         if (check_mtu) {
5841                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5842                 if (params->tot_len > mtu) {
5843                         params->mtu_result = mtu; /* union with tot_len */
5844                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5845                 }
5846         }
5847
5848         nhc = res.nhc;
5849
5850         /* do not handle lwt encaps right now */
5851         if (nhc->nhc_lwtstate)
5852                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5853
5854         dev = nhc->nhc_dev;
5855
5856         params->rt_metric = res.fi->fib_priority;
5857         params->ifindex = dev->ifindex;
5858
5859         /* xdp and cls_bpf programs are run in RCU-bh so
5860          * rcu_read_lock_bh is not needed here
5861          */
5862         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5863                 if (nhc->nhc_gw_family)
5864                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5865         } else {
5866                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5867
5868                 params->family = AF_INET6;
5869                 *dst = nhc->nhc_gw.ipv6;
5870         }
5871
5872         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5873                 goto set_fwd_params;
5874
5875         if (likely(nhc->nhc_gw_family != AF_INET6))
5876                 neigh = __ipv4_neigh_lookup_noref(dev,
5877                                                   (__force u32)params->ipv4_dst);
5878         else
5879                 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5880
5881         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5882                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5883         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5884         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5885
5886 set_fwd_params:
5887         return bpf_fib_set_fwd_params(params, mtu);
5888 }
5889 #endif
5890
5891 #if IS_ENABLED(CONFIG_IPV6)
5892 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5893                                u32 flags, bool check_mtu)
5894 {
5895         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5896         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5897         struct fib6_result res = {};
5898         struct neighbour *neigh;
5899         struct net_device *dev;
5900         struct inet6_dev *idev;
5901         struct flowi6 fl6;
5902         int strict = 0;
5903         int oif, err;
5904         u32 mtu = 0;
5905
5906         /* link local addresses are never forwarded */
5907         if (rt6_need_strict(dst) || rt6_need_strict(src))
5908                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5909
5910         dev = dev_get_by_index_rcu(net, params->ifindex);
5911         if (unlikely(!dev))
5912                 return -ENODEV;
5913
5914         idev = __in6_dev_get_safely(dev);
5915         if (unlikely(!idev || !idev->cnf.forwarding))
5916                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5917
5918         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5919                 fl6.flowi6_iif = 1;
5920                 oif = fl6.flowi6_oif = params->ifindex;
5921         } else {
5922                 oif = fl6.flowi6_iif = params->ifindex;
5923                 fl6.flowi6_oif = 0;
5924                 strict = RT6_LOOKUP_F_HAS_SADDR;
5925         }
5926         fl6.flowlabel = params->flowinfo;
5927         fl6.flowi6_scope = 0;
5928         fl6.flowi6_flags = 0;
5929         fl6.mp_hash = 0;
5930
5931         fl6.flowi6_proto = params->l4_protocol;
5932         fl6.daddr = *dst;
5933         fl6.saddr = *src;
5934         fl6.fl6_sport = params->sport;
5935         fl6.fl6_dport = params->dport;
5936
5937         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5938                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5939                 struct fib6_table *tb;
5940
5941                 if (flags & BPF_FIB_LOOKUP_TBID) {
5942                         tbid = params->tbid;
5943                         /* zero out for vlan output */
5944                         params->tbid = 0;
5945                 }
5946
5947                 tb = ipv6_stub->fib6_get_table(net, tbid);
5948                 if (unlikely(!tb))
5949                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5950
5951                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5952                                                    strict);
5953         } else {
5954                 fl6.flowi6_mark = 0;
5955                 fl6.flowi6_secid = 0;
5956                 fl6.flowi6_tun_key.tun_id = 0;
5957                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5958
5959                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5960         }
5961
5962         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5963                      res.f6i == net->ipv6.fib6_null_entry))
5964                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5965
5966         switch (res.fib6_type) {
5967         /* only unicast is forwarded */
5968         case RTN_UNICAST:
5969                 break;
5970         case RTN_BLACKHOLE:
5971                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5972         case RTN_UNREACHABLE:
5973                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5974         case RTN_PROHIBIT:
5975                 return BPF_FIB_LKUP_RET_PROHIBIT;
5976         default:
5977                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5978         }
5979
5980         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5981                                     fl6.flowi6_oif != 0, NULL, strict);
5982
5983         if (check_mtu) {
5984                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5985                 if (params->tot_len > mtu) {
5986                         params->mtu_result = mtu; /* union with tot_len */
5987                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5988                 }
5989         }
5990
5991         if (res.nh->fib_nh_lws)
5992                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5993
5994         if (res.nh->fib_nh_gw_family)
5995                 *dst = res.nh->fib_nh_gw6;
5996
5997         dev = res.nh->fib_nh_dev;
5998         params->rt_metric = res.f6i->fib6_metric;
5999         params->ifindex = dev->ifindex;
6000
6001         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6002                 goto set_fwd_params;
6003
6004         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6005          * not needed here.
6006          */
6007         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6008         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6009                 return BPF_FIB_LKUP_RET_NO_NEIGH;
6010         memcpy(params->dmac, neigh->ha, ETH_ALEN);
6011         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6012
6013 set_fwd_params:
6014         return bpf_fib_set_fwd_params(params, mtu);
6015 }
6016 #endif
6017
6018 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6019                              BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6020
6021 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6022            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6023 {
6024         if (plen < sizeof(*params))
6025                 return -EINVAL;
6026
6027         if (flags & ~BPF_FIB_LOOKUP_MASK)
6028                 return -EINVAL;
6029
6030         switch (params->family) {
6031 #if IS_ENABLED(CONFIG_INET)
6032         case AF_INET:
6033                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6034                                            flags, true);
6035 #endif
6036 #if IS_ENABLED(CONFIG_IPV6)
6037         case AF_INET6:
6038                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6039                                            flags, true);
6040 #endif
6041         }
6042         return -EAFNOSUPPORT;
6043 }
6044
6045 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6046         .func           = bpf_xdp_fib_lookup,
6047         .gpl_only       = true,
6048         .ret_type       = RET_INTEGER,
6049         .arg1_type      = ARG_PTR_TO_CTX,
6050         .arg2_type      = ARG_PTR_TO_MEM,
6051         .arg3_type      = ARG_CONST_SIZE,
6052         .arg4_type      = ARG_ANYTHING,
6053 };
6054
6055 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6056            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6057 {
6058         struct net *net = dev_net(skb->dev);
6059         int rc = -EAFNOSUPPORT;
6060         bool check_mtu = false;
6061
6062         if (plen < sizeof(*params))
6063                 return -EINVAL;
6064
6065         if (flags & ~BPF_FIB_LOOKUP_MASK)
6066                 return -EINVAL;
6067
6068         if (params->tot_len)
6069                 check_mtu = true;
6070
6071         switch (params->family) {
6072 #if IS_ENABLED(CONFIG_INET)
6073         case AF_INET:
6074                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6075                 break;
6076 #endif
6077 #if IS_ENABLED(CONFIG_IPV6)
6078         case AF_INET6:
6079                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6080                 break;
6081 #endif
6082         }
6083
6084         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6085                 struct net_device *dev;
6086
6087                 /* When tot_len isn't provided by user, check skb
6088                  * against MTU of FIB lookup resulting net_device
6089                  */
6090                 dev = dev_get_by_index_rcu(net, params->ifindex);
6091                 if (!is_skb_forwardable(dev, skb))
6092                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6093
6094                 params->mtu_result = dev->mtu; /* union with tot_len */
6095         }
6096
6097         return rc;
6098 }
6099
6100 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6101         .func           = bpf_skb_fib_lookup,
6102         .gpl_only       = true,
6103         .ret_type       = RET_INTEGER,
6104         .arg1_type      = ARG_PTR_TO_CTX,
6105         .arg2_type      = ARG_PTR_TO_MEM,
6106         .arg3_type      = ARG_CONST_SIZE,
6107         .arg4_type      = ARG_ANYTHING,
6108 };
6109
6110 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6111                                             u32 ifindex)
6112 {
6113         struct net *netns = dev_net(dev_curr);
6114
6115         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6116         if (ifindex == 0)
6117                 return dev_curr;
6118
6119         return dev_get_by_index_rcu(netns, ifindex);
6120 }
6121
6122 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6123            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6124 {
6125         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6126         struct net_device *dev = skb->dev;
6127         int skb_len, dev_len;
6128         int mtu;
6129
6130         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6131                 return -EINVAL;
6132
6133         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6134                 return -EINVAL;
6135
6136         dev = __dev_via_ifindex(dev, ifindex);
6137         if (unlikely(!dev))
6138                 return -ENODEV;
6139
6140         mtu = READ_ONCE(dev->mtu);
6141
6142         dev_len = mtu + dev->hard_header_len;
6143
6144         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6145         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6146
6147         skb_len += len_diff; /* minus result pass check */
6148         if (skb_len <= dev_len) {
6149                 ret = BPF_MTU_CHK_RET_SUCCESS;
6150                 goto out;
6151         }
6152         /* At this point, skb->len exceed MTU, but as it include length of all
6153          * segments, it can still be below MTU.  The SKB can possibly get
6154          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6155          * must choose if segs are to be MTU checked.
6156          */
6157         if (skb_is_gso(skb)) {
6158                 ret = BPF_MTU_CHK_RET_SUCCESS;
6159
6160                 if (flags & BPF_MTU_CHK_SEGS &&
6161                     !skb_gso_validate_network_len(skb, mtu))
6162                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6163         }
6164 out:
6165         /* BPF verifier guarantees valid pointer */
6166         *mtu_len = mtu;
6167
6168         return ret;
6169 }
6170
6171 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6172            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6173 {
6174         struct net_device *dev = xdp->rxq->dev;
6175         int xdp_len = xdp->data_end - xdp->data;
6176         int ret = BPF_MTU_CHK_RET_SUCCESS;
6177         int mtu, dev_len;
6178
6179         /* XDP variant doesn't support multi-buffer segment check (yet) */
6180         if (unlikely(flags))
6181                 return -EINVAL;
6182
6183         dev = __dev_via_ifindex(dev, ifindex);
6184         if (unlikely(!dev))
6185                 return -ENODEV;
6186
6187         mtu = READ_ONCE(dev->mtu);
6188
6189         /* Add L2-header as dev MTU is L3 size */
6190         dev_len = mtu + dev->hard_header_len;
6191
6192         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6193         if (*mtu_len)
6194                 xdp_len = *mtu_len + dev->hard_header_len;
6195
6196         xdp_len += len_diff; /* minus result pass check */
6197         if (xdp_len > dev_len)
6198                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6199
6200         /* BPF verifier guarantees valid pointer */
6201         *mtu_len = mtu;
6202
6203         return ret;
6204 }
6205
6206 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6207         .func           = bpf_skb_check_mtu,
6208         .gpl_only       = true,
6209         .ret_type       = RET_INTEGER,
6210         .arg1_type      = ARG_PTR_TO_CTX,
6211         .arg2_type      = ARG_ANYTHING,
6212         .arg3_type      = ARG_PTR_TO_INT,
6213         .arg4_type      = ARG_ANYTHING,
6214         .arg5_type      = ARG_ANYTHING,
6215 };
6216
6217 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6218         .func           = bpf_xdp_check_mtu,
6219         .gpl_only       = true,
6220         .ret_type       = RET_INTEGER,
6221         .arg1_type      = ARG_PTR_TO_CTX,
6222         .arg2_type      = ARG_ANYTHING,
6223         .arg3_type      = ARG_PTR_TO_INT,
6224         .arg4_type      = ARG_ANYTHING,
6225         .arg5_type      = ARG_ANYTHING,
6226 };
6227
6228 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6229 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6230 {
6231         int err;
6232         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6233
6234         if (!seg6_validate_srh(srh, len, false))
6235                 return -EINVAL;
6236
6237         switch (type) {
6238         case BPF_LWT_ENCAP_SEG6_INLINE:
6239                 if (skb->protocol != htons(ETH_P_IPV6))
6240                         return -EBADMSG;
6241
6242                 err = seg6_do_srh_inline(skb, srh);
6243                 break;
6244         case BPF_LWT_ENCAP_SEG6:
6245                 skb_reset_inner_headers(skb);
6246                 skb->encapsulation = 1;
6247                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6248                 break;
6249         default:
6250                 return -EINVAL;
6251         }
6252
6253         bpf_compute_data_pointers(skb);
6254         if (err)
6255                 return err;
6256
6257         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6258
6259         return seg6_lookup_nexthop(skb, NULL, 0);
6260 }
6261 #endif /* CONFIG_IPV6_SEG6_BPF */
6262
6263 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6264 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6265                              bool ingress)
6266 {
6267         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6268 }
6269 #endif
6270
6271 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6272            u32, len)
6273 {
6274         switch (type) {
6275 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6276         case BPF_LWT_ENCAP_SEG6:
6277         case BPF_LWT_ENCAP_SEG6_INLINE:
6278                 return bpf_push_seg6_encap(skb, type, hdr, len);
6279 #endif
6280 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6281         case BPF_LWT_ENCAP_IP:
6282                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6283 #endif
6284         default:
6285                 return -EINVAL;
6286         }
6287 }
6288
6289 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6290            void *, hdr, u32, len)
6291 {
6292         switch (type) {
6293 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6294         case BPF_LWT_ENCAP_IP:
6295                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6296 #endif
6297         default:
6298                 return -EINVAL;
6299         }
6300 }
6301
6302 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6303         .func           = bpf_lwt_in_push_encap,
6304         .gpl_only       = false,
6305         .ret_type       = RET_INTEGER,
6306         .arg1_type      = ARG_PTR_TO_CTX,
6307         .arg2_type      = ARG_ANYTHING,
6308         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6309         .arg4_type      = ARG_CONST_SIZE
6310 };
6311
6312 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6313         .func           = bpf_lwt_xmit_push_encap,
6314         .gpl_only       = false,
6315         .ret_type       = RET_INTEGER,
6316         .arg1_type      = ARG_PTR_TO_CTX,
6317         .arg2_type      = ARG_ANYTHING,
6318         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6319         .arg4_type      = ARG_CONST_SIZE
6320 };
6321
6322 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6323 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6324            const void *, from, u32, len)
6325 {
6326         struct seg6_bpf_srh_state *srh_state =
6327                 this_cpu_ptr(&seg6_bpf_srh_states);
6328         struct ipv6_sr_hdr *srh = srh_state->srh;
6329         void *srh_tlvs, *srh_end, *ptr;
6330         int srhoff = 0;
6331
6332         if (srh == NULL)
6333                 return -EINVAL;
6334
6335         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6336         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6337
6338         ptr = skb->data + offset;
6339         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6340                 srh_state->valid = false;
6341         else if (ptr < (void *)&srh->flags ||
6342                  ptr + len > (void *)&srh->segments)
6343                 return -EFAULT;
6344
6345         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6346                 return -EFAULT;
6347         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6348                 return -EINVAL;
6349         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6350
6351         memcpy(skb->data + offset, from, len);
6352         return 0;
6353 }
6354
6355 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6356         .func           = bpf_lwt_seg6_store_bytes,
6357         .gpl_only       = false,
6358         .ret_type       = RET_INTEGER,
6359         .arg1_type      = ARG_PTR_TO_CTX,
6360         .arg2_type      = ARG_ANYTHING,
6361         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6362         .arg4_type      = ARG_CONST_SIZE
6363 };
6364
6365 static void bpf_update_srh_state(struct sk_buff *skb)
6366 {
6367         struct seg6_bpf_srh_state *srh_state =
6368                 this_cpu_ptr(&seg6_bpf_srh_states);
6369         int srhoff = 0;
6370
6371         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6372                 srh_state->srh = NULL;
6373         } else {
6374                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6375                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6376                 srh_state->valid = true;
6377         }
6378 }
6379
6380 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6381            u32, action, void *, param, u32, param_len)
6382 {
6383         struct seg6_bpf_srh_state *srh_state =
6384                 this_cpu_ptr(&seg6_bpf_srh_states);
6385         int hdroff = 0;
6386         int err;
6387
6388         switch (action) {
6389         case SEG6_LOCAL_ACTION_END_X:
6390                 if (!seg6_bpf_has_valid_srh(skb))
6391                         return -EBADMSG;
6392                 if (param_len != sizeof(struct in6_addr))
6393                         return -EINVAL;
6394                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6395         case SEG6_LOCAL_ACTION_END_T:
6396                 if (!seg6_bpf_has_valid_srh(skb))
6397                         return -EBADMSG;
6398                 if (param_len != sizeof(int))
6399                         return -EINVAL;
6400                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6401         case SEG6_LOCAL_ACTION_END_DT6:
6402                 if (!seg6_bpf_has_valid_srh(skb))
6403                         return -EBADMSG;
6404                 if (param_len != sizeof(int))
6405                         return -EINVAL;
6406
6407                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6408                         return -EBADMSG;
6409                 if (!pskb_pull(skb, hdroff))
6410                         return -EBADMSG;
6411
6412                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6413                 skb_reset_network_header(skb);
6414                 skb_reset_transport_header(skb);
6415                 skb->encapsulation = 0;
6416
6417                 bpf_compute_data_pointers(skb);
6418                 bpf_update_srh_state(skb);
6419                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6420         case SEG6_LOCAL_ACTION_END_B6:
6421                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6422                         return -EBADMSG;
6423                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6424                                           param, param_len);
6425                 if (!err)
6426                         bpf_update_srh_state(skb);
6427
6428                 return err;
6429         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6430                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6431                         return -EBADMSG;
6432                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6433                                           param, param_len);
6434                 if (!err)
6435                         bpf_update_srh_state(skb);
6436
6437                 return err;
6438         default:
6439                 return -EINVAL;
6440         }
6441 }
6442
6443 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6444         .func           = bpf_lwt_seg6_action,
6445         .gpl_only       = false,
6446         .ret_type       = RET_INTEGER,
6447         .arg1_type      = ARG_PTR_TO_CTX,
6448         .arg2_type      = ARG_ANYTHING,
6449         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6450         .arg4_type      = ARG_CONST_SIZE
6451 };
6452
6453 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6454            s32, len)
6455 {
6456         struct seg6_bpf_srh_state *srh_state =
6457                 this_cpu_ptr(&seg6_bpf_srh_states);
6458         struct ipv6_sr_hdr *srh = srh_state->srh;
6459         void *srh_end, *srh_tlvs, *ptr;
6460         struct ipv6hdr *hdr;
6461         int srhoff = 0;
6462         int ret;
6463
6464         if (unlikely(srh == NULL))
6465                 return -EINVAL;
6466
6467         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6468                         ((srh->first_segment + 1) << 4));
6469         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6470                         srh_state->hdrlen);
6471         ptr = skb->data + offset;
6472
6473         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6474                 return -EFAULT;
6475         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6476                 return -EFAULT;
6477
6478         if (len > 0) {
6479                 ret = skb_cow_head(skb, len);
6480                 if (unlikely(ret < 0))
6481                         return ret;
6482
6483                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6484         } else {
6485                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6486         }
6487
6488         bpf_compute_data_pointers(skb);
6489         if (unlikely(ret < 0))
6490                 return ret;
6491
6492         hdr = (struct ipv6hdr *)skb->data;
6493         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6494
6495         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6496                 return -EINVAL;
6497         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6498         srh_state->hdrlen += len;
6499         srh_state->valid = false;
6500         return 0;
6501 }
6502
6503 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6504         .func           = bpf_lwt_seg6_adjust_srh,
6505         .gpl_only       = false,
6506         .ret_type       = RET_INTEGER,
6507         .arg1_type      = ARG_PTR_TO_CTX,
6508         .arg2_type      = ARG_ANYTHING,
6509         .arg3_type      = ARG_ANYTHING,
6510 };
6511 #endif /* CONFIG_IPV6_SEG6_BPF */
6512
6513 #ifdef CONFIG_INET
6514 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6515                               int dif, int sdif, u8 family, u8 proto)
6516 {
6517         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6518         bool refcounted = false;
6519         struct sock *sk = NULL;
6520
6521         if (family == AF_INET) {
6522                 __be32 src4 = tuple->ipv4.saddr;
6523                 __be32 dst4 = tuple->ipv4.daddr;
6524
6525                 if (proto == IPPROTO_TCP)
6526                         sk = __inet_lookup(net, hinfo, NULL, 0,
6527                                            src4, tuple->ipv4.sport,
6528                                            dst4, tuple->ipv4.dport,
6529                                            dif, sdif, &refcounted);
6530                 else
6531                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6532                                                dst4, tuple->ipv4.dport,
6533                                                dif, sdif, net->ipv4.udp_table, NULL);
6534 #if IS_ENABLED(CONFIG_IPV6)
6535         } else {
6536                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6537                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6538
6539                 if (proto == IPPROTO_TCP)
6540                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6541                                             src6, tuple->ipv6.sport,
6542                                             dst6, ntohs(tuple->ipv6.dport),
6543                                             dif, sdif, &refcounted);
6544                 else if (likely(ipv6_bpf_stub))
6545                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6546                                                             src6, tuple->ipv6.sport,
6547                                                             dst6, tuple->ipv6.dport,
6548                                                             dif, sdif,
6549                                                             net->ipv4.udp_table, NULL);
6550 #endif
6551         }
6552
6553         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6554                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6555                 sk = NULL;
6556         }
6557         return sk;
6558 }
6559
6560 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6561  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6562  */
6563 static struct sock *
6564 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6565                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6566                  u64 flags, int sdif)
6567 {
6568         struct sock *sk = NULL;
6569         struct net *net;
6570         u8 family;
6571
6572         if (len == sizeof(tuple->ipv4))
6573                 family = AF_INET;
6574         else if (len == sizeof(tuple->ipv6))
6575                 family = AF_INET6;
6576         else
6577                 return NULL;
6578
6579         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6580                 goto out;
6581
6582         if (sdif < 0) {
6583                 if (family == AF_INET)
6584                         sdif = inet_sdif(skb);
6585                 else
6586                         sdif = inet6_sdif(skb);
6587         }
6588
6589         if ((s32)netns_id < 0) {
6590                 net = caller_net;
6591                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6592         } else {
6593                 net = get_net_ns_by_id(caller_net, netns_id);
6594                 if (unlikely(!net))
6595                         goto out;
6596                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6597                 put_net(net);
6598         }
6599
6600 out:
6601         return sk;
6602 }
6603
6604 static struct sock *
6605 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6606                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6607                 u64 flags, int sdif)
6608 {
6609         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6610                                            ifindex, proto, netns_id, flags,
6611                                            sdif);
6612
6613         if (sk) {
6614                 struct sock *sk2 = sk_to_full_sk(sk);
6615
6616                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6617                  * sock refcnt is decremented to prevent a request_sock leak.
6618                  */
6619                 if (!sk_fullsock(sk2))
6620                         sk2 = NULL;
6621                 if (sk2 != sk) {
6622                         sock_gen_put(sk);
6623                         /* Ensure there is no need to bump sk2 refcnt */
6624                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6625                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6626                                 return NULL;
6627                         }
6628                         sk = sk2;
6629                 }
6630         }
6631
6632         return sk;
6633 }
6634
6635 static struct sock *
6636 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6637                u8 proto, u64 netns_id, u64 flags)
6638 {
6639         struct net *caller_net;
6640         int ifindex;
6641
6642         if (skb->dev) {
6643                 caller_net = dev_net(skb->dev);
6644                 ifindex = skb->dev->ifindex;
6645         } else {
6646                 caller_net = sock_net(skb->sk);
6647                 ifindex = 0;
6648         }
6649
6650         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6651                                 netns_id, flags, -1);
6652 }
6653
6654 static struct sock *
6655 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6656               u8 proto, u64 netns_id, u64 flags)
6657 {
6658         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6659                                          flags);
6660
6661         if (sk) {
6662                 struct sock *sk2 = sk_to_full_sk(sk);
6663
6664                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6665                  * sock refcnt is decremented to prevent a request_sock leak.
6666                  */
6667                 if (!sk_fullsock(sk2))
6668                         sk2 = NULL;
6669                 if (sk2 != sk) {
6670                         sock_gen_put(sk);
6671                         /* Ensure there is no need to bump sk2 refcnt */
6672                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6673                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6674                                 return NULL;
6675                         }
6676                         sk = sk2;
6677                 }
6678         }
6679
6680         return sk;
6681 }
6682
6683 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6684            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6685 {
6686         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6687                                              netns_id, flags);
6688 }
6689
6690 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6691         .func           = bpf_skc_lookup_tcp,
6692         .gpl_only       = false,
6693         .pkt_access     = true,
6694         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6695         .arg1_type      = ARG_PTR_TO_CTX,
6696         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6697         .arg3_type      = ARG_CONST_SIZE,
6698         .arg4_type      = ARG_ANYTHING,
6699         .arg5_type      = ARG_ANYTHING,
6700 };
6701
6702 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6703            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6704 {
6705         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6706                                             netns_id, flags);
6707 }
6708
6709 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6710         .func           = bpf_sk_lookup_tcp,
6711         .gpl_only       = false,
6712         .pkt_access     = true,
6713         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6714         .arg1_type      = ARG_PTR_TO_CTX,
6715         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6716         .arg3_type      = ARG_CONST_SIZE,
6717         .arg4_type      = ARG_ANYTHING,
6718         .arg5_type      = ARG_ANYTHING,
6719 };
6720
6721 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6722            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6723 {
6724         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6725                                             netns_id, flags);
6726 }
6727
6728 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6729         .func           = bpf_sk_lookup_udp,
6730         .gpl_only       = false,
6731         .pkt_access     = true,
6732         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6733         .arg1_type      = ARG_PTR_TO_CTX,
6734         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6735         .arg3_type      = ARG_CONST_SIZE,
6736         .arg4_type      = ARG_ANYTHING,
6737         .arg5_type      = ARG_ANYTHING,
6738 };
6739
6740 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6741            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6742 {
6743         struct net_device *dev = skb->dev;
6744         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6745         struct net *caller_net = dev_net(dev);
6746
6747         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6748                                                ifindex, IPPROTO_TCP, netns_id,
6749                                                flags, sdif);
6750 }
6751
6752 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6753         .func           = bpf_tc_skc_lookup_tcp,
6754         .gpl_only       = false,
6755         .pkt_access     = true,
6756         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6757         .arg1_type      = ARG_PTR_TO_CTX,
6758         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6759         .arg3_type      = ARG_CONST_SIZE,
6760         .arg4_type      = ARG_ANYTHING,
6761         .arg5_type      = ARG_ANYTHING,
6762 };
6763
6764 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6765            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6766 {
6767         struct net_device *dev = skb->dev;
6768         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6769         struct net *caller_net = dev_net(dev);
6770
6771         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6772                                               ifindex, IPPROTO_TCP, netns_id,
6773                                               flags, sdif);
6774 }
6775
6776 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6777         .func           = bpf_tc_sk_lookup_tcp,
6778         .gpl_only       = false,
6779         .pkt_access     = true,
6780         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6781         .arg1_type      = ARG_PTR_TO_CTX,
6782         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6783         .arg3_type      = ARG_CONST_SIZE,
6784         .arg4_type      = ARG_ANYTHING,
6785         .arg5_type      = ARG_ANYTHING,
6786 };
6787
6788 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6789            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6790 {
6791         struct net_device *dev = skb->dev;
6792         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6793         struct net *caller_net = dev_net(dev);
6794
6795         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6796                                               ifindex, IPPROTO_UDP, netns_id,
6797                                               flags, sdif);
6798 }
6799
6800 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6801         .func           = bpf_tc_sk_lookup_udp,
6802         .gpl_only       = false,
6803         .pkt_access     = true,
6804         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6805         .arg1_type      = ARG_PTR_TO_CTX,
6806         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6807         .arg3_type      = ARG_CONST_SIZE,
6808         .arg4_type      = ARG_ANYTHING,
6809         .arg5_type      = ARG_ANYTHING,
6810 };
6811
6812 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6813 {
6814         if (sk && sk_is_refcounted(sk))
6815                 sock_gen_put(sk);
6816         return 0;
6817 }
6818
6819 static const struct bpf_func_proto bpf_sk_release_proto = {
6820         .func           = bpf_sk_release,
6821         .gpl_only       = false,
6822         .ret_type       = RET_INTEGER,
6823         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6824 };
6825
6826 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6827            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6828 {
6829         struct net_device *dev = ctx->rxq->dev;
6830         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6831         struct net *caller_net = dev_net(dev);
6832
6833         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6834                                               ifindex, IPPROTO_UDP, netns_id,
6835                                               flags, sdif);
6836 }
6837
6838 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6839         .func           = bpf_xdp_sk_lookup_udp,
6840         .gpl_only       = false,
6841         .pkt_access     = true,
6842         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6843         .arg1_type      = ARG_PTR_TO_CTX,
6844         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6845         .arg3_type      = ARG_CONST_SIZE,
6846         .arg4_type      = ARG_ANYTHING,
6847         .arg5_type      = ARG_ANYTHING,
6848 };
6849
6850 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6851            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6852 {
6853         struct net_device *dev = ctx->rxq->dev;
6854         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6855         struct net *caller_net = dev_net(dev);
6856
6857         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6858                                                ifindex, IPPROTO_TCP, netns_id,
6859                                                flags, sdif);
6860 }
6861
6862 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6863         .func           = bpf_xdp_skc_lookup_tcp,
6864         .gpl_only       = false,
6865         .pkt_access     = true,
6866         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6867         .arg1_type      = ARG_PTR_TO_CTX,
6868         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6869         .arg3_type      = ARG_CONST_SIZE,
6870         .arg4_type      = ARG_ANYTHING,
6871         .arg5_type      = ARG_ANYTHING,
6872 };
6873
6874 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6875            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6876 {
6877         struct net_device *dev = ctx->rxq->dev;
6878         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6879         struct net *caller_net = dev_net(dev);
6880
6881         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6882                                               ifindex, IPPROTO_TCP, netns_id,
6883                                               flags, sdif);
6884 }
6885
6886 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6887         .func           = bpf_xdp_sk_lookup_tcp,
6888         .gpl_only       = false,
6889         .pkt_access     = true,
6890         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6891         .arg1_type      = ARG_PTR_TO_CTX,
6892         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6893         .arg3_type      = ARG_CONST_SIZE,
6894         .arg4_type      = ARG_ANYTHING,
6895         .arg5_type      = ARG_ANYTHING,
6896 };
6897
6898 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6899            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6900 {
6901         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6902                                                sock_net(ctx->sk), 0,
6903                                                IPPROTO_TCP, netns_id, flags,
6904                                                -1);
6905 }
6906
6907 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6908         .func           = bpf_sock_addr_skc_lookup_tcp,
6909         .gpl_only       = false,
6910         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6911         .arg1_type      = ARG_PTR_TO_CTX,
6912         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6913         .arg3_type      = ARG_CONST_SIZE,
6914         .arg4_type      = ARG_ANYTHING,
6915         .arg5_type      = ARG_ANYTHING,
6916 };
6917
6918 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6919            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6920 {
6921         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6922                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6923                                               netns_id, flags, -1);
6924 }
6925
6926 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6927         .func           = bpf_sock_addr_sk_lookup_tcp,
6928         .gpl_only       = false,
6929         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6930         .arg1_type      = ARG_PTR_TO_CTX,
6931         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6932         .arg3_type      = ARG_CONST_SIZE,
6933         .arg4_type      = ARG_ANYTHING,
6934         .arg5_type      = ARG_ANYTHING,
6935 };
6936
6937 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6938            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6939 {
6940         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6941                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6942                                               netns_id, flags, -1);
6943 }
6944
6945 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6946         .func           = bpf_sock_addr_sk_lookup_udp,
6947         .gpl_only       = false,
6948         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6949         .arg1_type      = ARG_PTR_TO_CTX,
6950         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6951         .arg3_type      = ARG_CONST_SIZE,
6952         .arg4_type      = ARG_ANYTHING,
6953         .arg5_type      = ARG_ANYTHING,
6954 };
6955
6956 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6957                                   struct bpf_insn_access_aux *info)
6958 {
6959         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6960                                           icsk_retransmits))
6961                 return false;
6962
6963         if (off % size != 0)
6964                 return false;
6965
6966         switch (off) {
6967         case offsetof(struct bpf_tcp_sock, bytes_received):
6968         case offsetof(struct bpf_tcp_sock, bytes_acked):
6969                 return size == sizeof(__u64);
6970         default:
6971                 return size == sizeof(__u32);
6972         }
6973 }
6974
6975 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6976                                     const struct bpf_insn *si,
6977                                     struct bpf_insn *insn_buf,
6978                                     struct bpf_prog *prog, u32 *target_size)
6979 {
6980         struct bpf_insn *insn = insn_buf;
6981
6982 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6983         do {                                                            \
6984                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6985                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6986                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6987                                       si->dst_reg, si->src_reg,         \
6988                                       offsetof(struct tcp_sock, FIELD)); \
6989         } while (0)
6990
6991 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6992         do {                                                            \
6993                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6994                                           FIELD) >                      \
6995                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6997                                         struct inet_connection_sock,    \
6998                                         FIELD),                         \
6999                                       si->dst_reg, si->src_reg,         \
7000                                       offsetof(                         \
7001                                         struct inet_connection_sock,    \
7002                                         FIELD));                        \
7003         } while (0)
7004
7005         BTF_TYPE_EMIT(struct bpf_tcp_sock);
7006
7007         switch (si->off) {
7008         case offsetof(struct bpf_tcp_sock, rtt_min):
7009                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7010                              sizeof(struct minmax));
7011                 BUILD_BUG_ON(sizeof(struct minmax) <
7012                              sizeof(struct minmax_sample));
7013
7014                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7015                                       offsetof(struct tcp_sock, rtt_min) +
7016                                       offsetof(struct minmax_sample, v));
7017                 break;
7018         case offsetof(struct bpf_tcp_sock, snd_cwnd):
7019                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7020                 break;
7021         case offsetof(struct bpf_tcp_sock, srtt_us):
7022                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7023                 break;
7024         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7025                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7026                 break;
7027         case offsetof(struct bpf_tcp_sock, rcv_nxt):
7028                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7029                 break;
7030         case offsetof(struct bpf_tcp_sock, snd_nxt):
7031                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7032                 break;
7033         case offsetof(struct bpf_tcp_sock, snd_una):
7034                 BPF_TCP_SOCK_GET_COMMON(snd_una);
7035                 break;
7036         case offsetof(struct bpf_tcp_sock, mss_cache):
7037                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7038                 break;
7039         case offsetof(struct bpf_tcp_sock, ecn_flags):
7040                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7041                 break;
7042         case offsetof(struct bpf_tcp_sock, rate_delivered):
7043                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7044                 break;
7045         case offsetof(struct bpf_tcp_sock, rate_interval_us):
7046                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7047                 break;
7048         case offsetof(struct bpf_tcp_sock, packets_out):
7049                 BPF_TCP_SOCK_GET_COMMON(packets_out);
7050                 break;
7051         case offsetof(struct bpf_tcp_sock, retrans_out):
7052                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7053                 break;
7054         case offsetof(struct bpf_tcp_sock, total_retrans):
7055                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7056                 break;
7057         case offsetof(struct bpf_tcp_sock, segs_in):
7058                 BPF_TCP_SOCK_GET_COMMON(segs_in);
7059                 break;
7060         case offsetof(struct bpf_tcp_sock, data_segs_in):
7061                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7062                 break;
7063         case offsetof(struct bpf_tcp_sock, segs_out):
7064                 BPF_TCP_SOCK_GET_COMMON(segs_out);
7065                 break;
7066         case offsetof(struct bpf_tcp_sock, data_segs_out):
7067                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7068                 break;
7069         case offsetof(struct bpf_tcp_sock, lost_out):
7070                 BPF_TCP_SOCK_GET_COMMON(lost_out);
7071                 break;
7072         case offsetof(struct bpf_tcp_sock, sacked_out):
7073                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7074                 break;
7075         case offsetof(struct bpf_tcp_sock, bytes_received):
7076                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7077                 break;
7078         case offsetof(struct bpf_tcp_sock, bytes_acked):
7079                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7080                 break;
7081         case offsetof(struct bpf_tcp_sock, dsack_dups):
7082                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7083                 break;
7084         case offsetof(struct bpf_tcp_sock, delivered):
7085                 BPF_TCP_SOCK_GET_COMMON(delivered);
7086                 break;
7087         case offsetof(struct bpf_tcp_sock, delivered_ce):
7088                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7089                 break;
7090         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7091                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7092                 break;
7093         }
7094
7095         return insn - insn_buf;
7096 }
7097
7098 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7099 {
7100         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7101                 return (unsigned long)sk;
7102
7103         return (unsigned long)NULL;
7104 }
7105
7106 const struct bpf_func_proto bpf_tcp_sock_proto = {
7107         .func           = bpf_tcp_sock,
7108         .gpl_only       = false,
7109         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
7110         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7111 };
7112
7113 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7114 {
7115         sk = sk_to_full_sk(sk);
7116
7117         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7118                 return (unsigned long)sk;
7119
7120         return (unsigned long)NULL;
7121 }
7122
7123 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7124         .func           = bpf_get_listener_sock,
7125         .gpl_only       = false,
7126         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7127         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7128 };
7129
7130 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7131 {
7132         unsigned int iphdr_len;
7133
7134         switch (skb_protocol(skb, true)) {
7135         case cpu_to_be16(ETH_P_IP):
7136                 iphdr_len = sizeof(struct iphdr);
7137                 break;
7138         case cpu_to_be16(ETH_P_IPV6):
7139                 iphdr_len = sizeof(struct ipv6hdr);
7140                 break;
7141         default:
7142                 return 0;
7143         }
7144
7145         if (skb_headlen(skb) < iphdr_len)
7146                 return 0;
7147
7148         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7149                 return 0;
7150
7151         return INET_ECN_set_ce(skb);
7152 }
7153
7154 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7155                                   struct bpf_insn_access_aux *info)
7156 {
7157         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7158                 return false;
7159
7160         if (off % size != 0)
7161                 return false;
7162
7163         switch (off) {
7164         default:
7165                 return size == sizeof(__u32);
7166         }
7167 }
7168
7169 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7170                                     const struct bpf_insn *si,
7171                                     struct bpf_insn *insn_buf,
7172                                     struct bpf_prog *prog, u32 *target_size)
7173 {
7174         struct bpf_insn *insn = insn_buf;
7175
7176 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7177         do {                                                            \
7178                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7179                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7180                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7181                                       si->dst_reg, si->src_reg,         \
7182                                       offsetof(struct xdp_sock, FIELD)); \
7183         } while (0)
7184
7185         switch (si->off) {
7186         case offsetof(struct bpf_xdp_sock, queue_id):
7187                 BPF_XDP_SOCK_GET(queue_id);
7188                 break;
7189         }
7190
7191         return insn - insn_buf;
7192 }
7193
7194 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7195         .func           = bpf_skb_ecn_set_ce,
7196         .gpl_only       = false,
7197         .ret_type       = RET_INTEGER,
7198         .arg1_type      = ARG_PTR_TO_CTX,
7199 };
7200
7201 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7202            struct tcphdr *, th, u32, th_len)
7203 {
7204 #ifdef CONFIG_SYN_COOKIES
7205         u32 cookie;
7206         int ret;
7207
7208         if (unlikely(!sk || th_len < sizeof(*th)))
7209                 return -EINVAL;
7210
7211         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7212         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7213                 return -EINVAL;
7214
7215         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7216                 return -EINVAL;
7217
7218         if (!th->ack || th->rst || th->syn)
7219                 return -ENOENT;
7220
7221         if (unlikely(iph_len < sizeof(struct iphdr)))
7222                 return -EINVAL;
7223
7224         if (tcp_synq_no_recent_overflow(sk))
7225                 return -ENOENT;
7226
7227         cookie = ntohl(th->ack_seq) - 1;
7228
7229         /* Both struct iphdr and struct ipv6hdr have the version field at the
7230          * same offset so we can cast to the shorter header (struct iphdr).
7231          */
7232         switch (((struct iphdr *)iph)->version) {
7233         case 4:
7234                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7235                         return -EINVAL;
7236
7237                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7238                 break;
7239
7240 #if IS_BUILTIN(CONFIG_IPV6)
7241         case 6:
7242                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7243                         return -EINVAL;
7244
7245                 if (sk->sk_family != AF_INET6)
7246                         return -EINVAL;
7247
7248                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7249                 break;
7250 #endif /* CONFIG_IPV6 */
7251
7252         default:
7253                 return -EPROTONOSUPPORT;
7254         }
7255
7256         if (ret > 0)
7257                 return 0;
7258
7259         return -ENOENT;
7260 #else
7261         return -ENOTSUPP;
7262 #endif
7263 }
7264
7265 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7266         .func           = bpf_tcp_check_syncookie,
7267         .gpl_only       = true,
7268         .pkt_access     = true,
7269         .ret_type       = RET_INTEGER,
7270         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7271         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7272         .arg3_type      = ARG_CONST_SIZE,
7273         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7274         .arg5_type      = ARG_CONST_SIZE,
7275 };
7276
7277 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7278            struct tcphdr *, th, u32, th_len)
7279 {
7280 #ifdef CONFIG_SYN_COOKIES
7281         u32 cookie;
7282         u16 mss;
7283
7284         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7285                 return -EINVAL;
7286
7287         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7288                 return -EINVAL;
7289
7290         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7291                 return -ENOENT;
7292
7293         if (!th->syn || th->ack || th->fin || th->rst)
7294                 return -EINVAL;
7295
7296         if (unlikely(iph_len < sizeof(struct iphdr)))
7297                 return -EINVAL;
7298
7299         /* Both struct iphdr and struct ipv6hdr have the version field at the
7300          * same offset so we can cast to the shorter header (struct iphdr).
7301          */
7302         switch (((struct iphdr *)iph)->version) {
7303         case 4:
7304                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7305                         return -EINVAL;
7306
7307                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7308                 break;
7309
7310 #if IS_BUILTIN(CONFIG_IPV6)
7311         case 6:
7312                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7313                         return -EINVAL;
7314
7315                 if (sk->sk_family != AF_INET6)
7316                         return -EINVAL;
7317
7318                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7319                 break;
7320 #endif /* CONFIG_IPV6 */
7321
7322         default:
7323                 return -EPROTONOSUPPORT;
7324         }
7325         if (mss == 0)
7326                 return -ENOENT;
7327
7328         return cookie | ((u64)mss << 32);
7329 #else
7330         return -EOPNOTSUPP;
7331 #endif /* CONFIG_SYN_COOKIES */
7332 }
7333
7334 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7335         .func           = bpf_tcp_gen_syncookie,
7336         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7337         .pkt_access     = true,
7338         .ret_type       = RET_INTEGER,
7339         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7340         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7341         .arg3_type      = ARG_CONST_SIZE,
7342         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7343         .arg5_type      = ARG_CONST_SIZE,
7344 };
7345
7346 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7347 {
7348         if (!sk || flags != 0)
7349                 return -EINVAL;
7350         if (!skb_at_tc_ingress(skb))
7351                 return -EOPNOTSUPP;
7352         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7353                 return -ENETUNREACH;
7354         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7355                 return -ESOCKTNOSUPPORT;
7356         if (sk_is_refcounted(sk) &&
7357             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7358                 return -ENOENT;
7359
7360         skb_orphan(skb);
7361         skb->sk = sk;
7362         skb->destructor = sock_pfree;
7363
7364         return 0;
7365 }
7366
7367 static const struct bpf_func_proto bpf_sk_assign_proto = {
7368         .func           = bpf_sk_assign,
7369         .gpl_only       = false,
7370         .ret_type       = RET_INTEGER,
7371         .arg1_type      = ARG_PTR_TO_CTX,
7372         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7373         .arg3_type      = ARG_ANYTHING,
7374 };
7375
7376 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7377                                     u8 search_kind, const u8 *magic,
7378                                     u8 magic_len, bool *eol)
7379 {
7380         u8 kind, kind_len;
7381
7382         *eol = false;
7383
7384         while (op < opend) {
7385                 kind = op[0];
7386
7387                 if (kind == TCPOPT_EOL) {
7388                         *eol = true;
7389                         return ERR_PTR(-ENOMSG);
7390                 } else if (kind == TCPOPT_NOP) {
7391                         op++;
7392                         continue;
7393                 }
7394
7395                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7396                         /* Something is wrong in the received header.
7397                          * Follow the TCP stack's tcp_parse_options()
7398                          * and just bail here.
7399                          */
7400                         return ERR_PTR(-EFAULT);
7401
7402                 kind_len = op[1];
7403                 if (search_kind == kind) {
7404                         if (!magic_len)
7405                                 return op;
7406
7407                         if (magic_len > kind_len - 2)
7408                                 return ERR_PTR(-ENOMSG);
7409
7410                         if (!memcmp(&op[2], magic, magic_len))
7411                                 return op;
7412                 }
7413
7414                 op += kind_len;
7415         }
7416
7417         return ERR_PTR(-ENOMSG);
7418 }
7419
7420 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7421            void *, search_res, u32, len, u64, flags)
7422 {
7423         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7424         const u8 *op, *opend, *magic, *search = search_res;
7425         u8 search_kind, search_len, copy_len, magic_len;
7426         int ret;
7427
7428         /* 2 byte is the minimal option len except TCPOPT_NOP and
7429          * TCPOPT_EOL which are useless for the bpf prog to learn
7430          * and this helper disallow loading them also.
7431          */
7432         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7433                 return -EINVAL;
7434
7435         search_kind = search[0];
7436         search_len = search[1];
7437
7438         if (search_len > len || search_kind == TCPOPT_NOP ||
7439             search_kind == TCPOPT_EOL)
7440                 return -EINVAL;
7441
7442         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7443                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7444                 if (search_len != 4 && search_len != 6)
7445                         return -EINVAL;
7446                 magic = &search[2];
7447                 magic_len = search_len - 2;
7448         } else {
7449                 if (search_len)
7450                         return -EINVAL;
7451                 magic = NULL;
7452                 magic_len = 0;
7453         }
7454
7455         if (load_syn) {
7456                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7457                 if (ret < 0)
7458                         return ret;
7459
7460                 opend = op + ret;
7461                 op += sizeof(struct tcphdr);
7462         } else {
7463                 if (!bpf_sock->skb ||
7464                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7465                         /* This bpf_sock->op cannot call this helper */
7466                         return -EPERM;
7467
7468                 opend = bpf_sock->skb_data_end;
7469                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7470         }
7471
7472         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7473                                 &eol);
7474         if (IS_ERR(op))
7475                 return PTR_ERR(op);
7476
7477         copy_len = op[1];
7478         ret = copy_len;
7479         if (copy_len > len) {
7480                 ret = -ENOSPC;
7481                 copy_len = len;
7482         }
7483
7484         memcpy(search_res, op, copy_len);
7485         return ret;
7486 }
7487
7488 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7489         .func           = bpf_sock_ops_load_hdr_opt,
7490         .gpl_only       = false,
7491         .ret_type       = RET_INTEGER,
7492         .arg1_type      = ARG_PTR_TO_CTX,
7493         .arg2_type      = ARG_PTR_TO_MEM,
7494         .arg3_type      = ARG_CONST_SIZE,
7495         .arg4_type      = ARG_ANYTHING,
7496 };
7497
7498 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7499            const void *, from, u32, len, u64, flags)
7500 {
7501         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7502         const u8 *op, *new_op, *magic = NULL;
7503         struct sk_buff *skb;
7504         bool eol;
7505
7506         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7507                 return -EPERM;
7508
7509         if (len < 2 || flags)
7510                 return -EINVAL;
7511
7512         new_op = from;
7513         new_kind = new_op[0];
7514         new_kind_len = new_op[1];
7515
7516         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7517             new_kind == TCPOPT_EOL)
7518                 return -EINVAL;
7519
7520         if (new_kind_len > bpf_sock->remaining_opt_len)
7521                 return -ENOSPC;
7522
7523         /* 253 is another experimental kind */
7524         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7525                 if (new_kind_len < 4)
7526                         return -EINVAL;
7527                 /* Match for the 2 byte magic also.
7528                  * RFC 6994: the magic could be 2 or 4 bytes.
7529                  * Hence, matching by 2 byte only is on the
7530                  * conservative side but it is the right
7531                  * thing to do for the 'search-for-duplication'
7532                  * purpose.
7533                  */
7534                 magic = &new_op[2];
7535                 magic_len = 2;
7536         }
7537
7538         /* Check for duplication */
7539         skb = bpf_sock->skb;
7540         op = skb->data + sizeof(struct tcphdr);
7541         opend = bpf_sock->skb_data_end;
7542
7543         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7544                                 &eol);
7545         if (!IS_ERR(op))
7546                 return -EEXIST;
7547
7548         if (PTR_ERR(op) != -ENOMSG)
7549                 return PTR_ERR(op);
7550
7551         if (eol)
7552                 /* The option has been ended.  Treat it as no more
7553                  * header option can be written.
7554                  */
7555                 return -ENOSPC;
7556
7557         /* No duplication found.  Store the header option. */
7558         memcpy(opend, from, new_kind_len);
7559
7560         bpf_sock->remaining_opt_len -= new_kind_len;
7561         bpf_sock->skb_data_end += new_kind_len;
7562
7563         return 0;
7564 }
7565
7566 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7567         .func           = bpf_sock_ops_store_hdr_opt,
7568         .gpl_only       = false,
7569         .ret_type       = RET_INTEGER,
7570         .arg1_type      = ARG_PTR_TO_CTX,
7571         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7572         .arg3_type      = ARG_CONST_SIZE,
7573         .arg4_type      = ARG_ANYTHING,
7574 };
7575
7576 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7577            u32, len, u64, flags)
7578 {
7579         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7580                 return -EPERM;
7581
7582         if (flags || len < 2)
7583                 return -EINVAL;
7584
7585         if (len > bpf_sock->remaining_opt_len)
7586                 return -ENOSPC;
7587
7588         bpf_sock->remaining_opt_len -= len;
7589
7590         return 0;
7591 }
7592
7593 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7594         .func           = bpf_sock_ops_reserve_hdr_opt,
7595         .gpl_only       = false,
7596         .ret_type       = RET_INTEGER,
7597         .arg1_type      = ARG_PTR_TO_CTX,
7598         .arg2_type      = ARG_ANYTHING,
7599         .arg3_type      = ARG_ANYTHING,
7600 };
7601
7602 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7603            u64, tstamp, u32, tstamp_type)
7604 {
7605         /* skb_clear_delivery_time() is done for inet protocol */
7606         if (skb->protocol != htons(ETH_P_IP) &&
7607             skb->protocol != htons(ETH_P_IPV6))
7608                 return -EOPNOTSUPP;
7609
7610         switch (tstamp_type) {
7611         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7612                 if (!tstamp)
7613                         return -EINVAL;
7614                 skb->tstamp = tstamp;
7615                 skb->mono_delivery_time = 1;
7616                 break;
7617         case BPF_SKB_TSTAMP_UNSPEC:
7618                 if (tstamp)
7619                         return -EINVAL;
7620                 skb->tstamp = 0;
7621                 skb->mono_delivery_time = 0;
7622                 break;
7623         default:
7624                 return -EINVAL;
7625         }
7626
7627         return 0;
7628 }
7629
7630 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7631         .func           = bpf_skb_set_tstamp,
7632         .gpl_only       = false,
7633         .ret_type       = RET_INTEGER,
7634         .arg1_type      = ARG_PTR_TO_CTX,
7635         .arg2_type      = ARG_ANYTHING,
7636         .arg3_type      = ARG_ANYTHING,
7637 };
7638
7639 #ifdef CONFIG_SYN_COOKIES
7640 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7641            struct tcphdr *, th, u32, th_len)
7642 {
7643         u32 cookie;
7644         u16 mss;
7645
7646         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7647                 return -EINVAL;
7648
7649         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7650         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7651
7652         return cookie | ((u64)mss << 32);
7653 }
7654
7655 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7656         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7657         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7658         .pkt_access     = true,
7659         .ret_type       = RET_INTEGER,
7660         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7661         .arg1_size      = sizeof(struct iphdr),
7662         .arg2_type      = ARG_PTR_TO_MEM,
7663         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7664 };
7665
7666 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7667            struct tcphdr *, th, u32, th_len)
7668 {
7669 #if IS_BUILTIN(CONFIG_IPV6)
7670         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7671                 sizeof(struct ipv6hdr);
7672         u32 cookie;
7673         u16 mss;
7674
7675         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7676                 return -EINVAL;
7677
7678         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7679         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7680
7681         return cookie | ((u64)mss << 32);
7682 #else
7683         return -EPROTONOSUPPORT;
7684 #endif
7685 }
7686
7687 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7688         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7689         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7690         .pkt_access     = true,
7691         .ret_type       = RET_INTEGER,
7692         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7693         .arg1_size      = sizeof(struct ipv6hdr),
7694         .arg2_type      = ARG_PTR_TO_MEM,
7695         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7696 };
7697
7698 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7699            struct tcphdr *, th)
7700 {
7701         u32 cookie = ntohl(th->ack_seq) - 1;
7702
7703         if (__cookie_v4_check(iph, th, cookie) > 0)
7704                 return 0;
7705
7706         return -EACCES;
7707 }
7708
7709 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7710         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7711         .gpl_only       = true, /* __cookie_v4_check is GPL */
7712         .pkt_access     = true,
7713         .ret_type       = RET_INTEGER,
7714         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7715         .arg1_size      = sizeof(struct iphdr),
7716         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7717         .arg2_size      = sizeof(struct tcphdr),
7718 };
7719
7720 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7721            struct tcphdr *, th)
7722 {
7723 #if IS_BUILTIN(CONFIG_IPV6)
7724         u32 cookie = ntohl(th->ack_seq) - 1;
7725
7726         if (__cookie_v6_check(iph, th, cookie) > 0)
7727                 return 0;
7728
7729         return -EACCES;
7730 #else
7731         return -EPROTONOSUPPORT;
7732 #endif
7733 }
7734
7735 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7736         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7737         .gpl_only       = true, /* __cookie_v6_check is GPL */
7738         .pkt_access     = true,
7739         .ret_type       = RET_INTEGER,
7740         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7741         .arg1_size      = sizeof(struct ipv6hdr),
7742         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7743         .arg2_size      = sizeof(struct tcphdr),
7744 };
7745 #endif /* CONFIG_SYN_COOKIES */
7746
7747 #endif /* CONFIG_INET */
7748
7749 bool bpf_helper_changes_pkt_data(void *func)
7750 {
7751         if (func == bpf_skb_vlan_push ||
7752             func == bpf_skb_vlan_pop ||
7753             func == bpf_skb_store_bytes ||
7754             func == bpf_skb_change_proto ||
7755             func == bpf_skb_change_head ||
7756             func == sk_skb_change_head ||
7757             func == bpf_skb_change_tail ||
7758             func == sk_skb_change_tail ||
7759             func == bpf_skb_adjust_room ||
7760             func == sk_skb_adjust_room ||
7761             func == bpf_skb_pull_data ||
7762             func == sk_skb_pull_data ||
7763             func == bpf_clone_redirect ||
7764             func == bpf_l3_csum_replace ||
7765             func == bpf_l4_csum_replace ||
7766             func == bpf_xdp_adjust_head ||
7767             func == bpf_xdp_adjust_meta ||
7768             func == bpf_msg_pull_data ||
7769             func == bpf_msg_push_data ||
7770             func == bpf_msg_pop_data ||
7771             func == bpf_xdp_adjust_tail ||
7772 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7773             func == bpf_lwt_seg6_store_bytes ||
7774             func == bpf_lwt_seg6_adjust_srh ||
7775             func == bpf_lwt_seg6_action ||
7776 #endif
7777 #ifdef CONFIG_INET
7778             func == bpf_sock_ops_store_hdr_opt ||
7779 #endif
7780             func == bpf_lwt_in_push_encap ||
7781             func == bpf_lwt_xmit_push_encap)
7782                 return true;
7783
7784         return false;
7785 }
7786
7787 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7788 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7789
7790 static const struct bpf_func_proto *
7791 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7792 {
7793         const struct bpf_func_proto *func_proto;
7794
7795         func_proto = cgroup_common_func_proto(func_id, prog);
7796         if (func_proto)
7797                 return func_proto;
7798
7799         func_proto = cgroup_current_func_proto(func_id, prog);
7800         if (func_proto)
7801                 return func_proto;
7802
7803         switch (func_id) {
7804         case BPF_FUNC_get_socket_cookie:
7805                 return &bpf_get_socket_cookie_sock_proto;
7806         case BPF_FUNC_get_netns_cookie:
7807                 return &bpf_get_netns_cookie_sock_proto;
7808         case BPF_FUNC_perf_event_output:
7809                 return &bpf_event_output_data_proto;
7810         case BPF_FUNC_sk_storage_get:
7811                 return &bpf_sk_storage_get_cg_sock_proto;
7812         case BPF_FUNC_ktime_get_coarse_ns:
7813                 return &bpf_ktime_get_coarse_ns_proto;
7814         default:
7815                 return bpf_base_func_proto(func_id);
7816         }
7817 }
7818
7819 static const struct bpf_func_proto *
7820 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7821 {
7822         const struct bpf_func_proto *func_proto;
7823
7824         func_proto = cgroup_common_func_proto(func_id, prog);
7825         if (func_proto)
7826                 return func_proto;
7827
7828         func_proto = cgroup_current_func_proto(func_id, prog);
7829         if (func_proto)
7830                 return func_proto;
7831
7832         switch (func_id) {
7833         case BPF_FUNC_bind:
7834                 switch (prog->expected_attach_type) {
7835                 case BPF_CGROUP_INET4_CONNECT:
7836                 case BPF_CGROUP_INET6_CONNECT:
7837                         return &bpf_bind_proto;
7838                 default:
7839                         return NULL;
7840                 }
7841         case BPF_FUNC_get_socket_cookie:
7842                 return &bpf_get_socket_cookie_sock_addr_proto;
7843         case BPF_FUNC_get_netns_cookie:
7844                 return &bpf_get_netns_cookie_sock_addr_proto;
7845         case BPF_FUNC_perf_event_output:
7846                 return &bpf_event_output_data_proto;
7847 #ifdef CONFIG_INET
7848         case BPF_FUNC_sk_lookup_tcp:
7849                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7850         case BPF_FUNC_sk_lookup_udp:
7851                 return &bpf_sock_addr_sk_lookup_udp_proto;
7852         case BPF_FUNC_sk_release:
7853                 return &bpf_sk_release_proto;
7854         case BPF_FUNC_skc_lookup_tcp:
7855                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7856 #endif /* CONFIG_INET */
7857         case BPF_FUNC_sk_storage_get:
7858                 return &bpf_sk_storage_get_proto;
7859         case BPF_FUNC_sk_storage_delete:
7860                 return &bpf_sk_storage_delete_proto;
7861         case BPF_FUNC_setsockopt:
7862                 switch (prog->expected_attach_type) {
7863                 case BPF_CGROUP_INET4_BIND:
7864                 case BPF_CGROUP_INET6_BIND:
7865                 case BPF_CGROUP_INET4_CONNECT:
7866                 case BPF_CGROUP_INET6_CONNECT:
7867                 case BPF_CGROUP_UDP4_RECVMSG:
7868                 case BPF_CGROUP_UDP6_RECVMSG:
7869                 case BPF_CGROUP_UDP4_SENDMSG:
7870                 case BPF_CGROUP_UDP6_SENDMSG:
7871                 case BPF_CGROUP_INET4_GETPEERNAME:
7872                 case BPF_CGROUP_INET6_GETPEERNAME:
7873                 case BPF_CGROUP_INET4_GETSOCKNAME:
7874                 case BPF_CGROUP_INET6_GETSOCKNAME:
7875                         return &bpf_sock_addr_setsockopt_proto;
7876                 default:
7877                         return NULL;
7878                 }
7879         case BPF_FUNC_getsockopt:
7880                 switch (prog->expected_attach_type) {
7881                 case BPF_CGROUP_INET4_BIND:
7882                 case BPF_CGROUP_INET6_BIND:
7883                 case BPF_CGROUP_INET4_CONNECT:
7884                 case BPF_CGROUP_INET6_CONNECT:
7885                 case BPF_CGROUP_UDP4_RECVMSG:
7886                 case BPF_CGROUP_UDP6_RECVMSG:
7887                 case BPF_CGROUP_UDP4_SENDMSG:
7888                 case BPF_CGROUP_UDP6_SENDMSG:
7889                 case BPF_CGROUP_INET4_GETPEERNAME:
7890                 case BPF_CGROUP_INET6_GETPEERNAME:
7891                 case BPF_CGROUP_INET4_GETSOCKNAME:
7892                 case BPF_CGROUP_INET6_GETSOCKNAME:
7893                         return &bpf_sock_addr_getsockopt_proto;
7894                 default:
7895                         return NULL;
7896                 }
7897         default:
7898                 return bpf_sk_base_func_proto(func_id);
7899         }
7900 }
7901
7902 static const struct bpf_func_proto *
7903 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7904 {
7905         switch (func_id) {
7906         case BPF_FUNC_skb_load_bytes:
7907                 return &bpf_skb_load_bytes_proto;
7908         case BPF_FUNC_skb_load_bytes_relative:
7909                 return &bpf_skb_load_bytes_relative_proto;
7910         case BPF_FUNC_get_socket_cookie:
7911                 return &bpf_get_socket_cookie_proto;
7912         case BPF_FUNC_get_socket_uid:
7913                 return &bpf_get_socket_uid_proto;
7914         case BPF_FUNC_perf_event_output:
7915                 return &bpf_skb_event_output_proto;
7916         default:
7917                 return bpf_sk_base_func_proto(func_id);
7918         }
7919 }
7920
7921 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7922 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7923
7924 static const struct bpf_func_proto *
7925 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7926 {
7927         const struct bpf_func_proto *func_proto;
7928
7929         func_proto = cgroup_common_func_proto(func_id, prog);
7930         if (func_proto)
7931                 return func_proto;
7932
7933         switch (func_id) {
7934         case BPF_FUNC_sk_fullsock:
7935                 return &bpf_sk_fullsock_proto;
7936         case BPF_FUNC_sk_storage_get:
7937                 return &bpf_sk_storage_get_proto;
7938         case BPF_FUNC_sk_storage_delete:
7939                 return &bpf_sk_storage_delete_proto;
7940         case BPF_FUNC_perf_event_output:
7941                 return &bpf_skb_event_output_proto;
7942 #ifdef CONFIG_SOCK_CGROUP_DATA
7943         case BPF_FUNC_skb_cgroup_id:
7944                 return &bpf_skb_cgroup_id_proto;
7945         case BPF_FUNC_skb_ancestor_cgroup_id:
7946                 return &bpf_skb_ancestor_cgroup_id_proto;
7947         case BPF_FUNC_sk_cgroup_id:
7948                 return &bpf_sk_cgroup_id_proto;
7949         case BPF_FUNC_sk_ancestor_cgroup_id:
7950                 return &bpf_sk_ancestor_cgroup_id_proto;
7951 #endif
7952 #ifdef CONFIG_INET
7953         case BPF_FUNC_sk_lookup_tcp:
7954                 return &bpf_sk_lookup_tcp_proto;
7955         case BPF_FUNC_sk_lookup_udp:
7956                 return &bpf_sk_lookup_udp_proto;
7957         case BPF_FUNC_sk_release:
7958                 return &bpf_sk_release_proto;
7959         case BPF_FUNC_skc_lookup_tcp:
7960                 return &bpf_skc_lookup_tcp_proto;
7961         case BPF_FUNC_tcp_sock:
7962                 return &bpf_tcp_sock_proto;
7963         case BPF_FUNC_get_listener_sock:
7964                 return &bpf_get_listener_sock_proto;
7965         case BPF_FUNC_skb_ecn_set_ce:
7966                 return &bpf_skb_ecn_set_ce_proto;
7967 #endif
7968         default:
7969                 return sk_filter_func_proto(func_id, prog);
7970         }
7971 }
7972
7973 static const struct bpf_func_proto *
7974 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7975 {
7976         switch (func_id) {
7977         case BPF_FUNC_skb_store_bytes:
7978                 return &bpf_skb_store_bytes_proto;
7979         case BPF_FUNC_skb_load_bytes:
7980                 return &bpf_skb_load_bytes_proto;
7981         case BPF_FUNC_skb_load_bytes_relative:
7982                 return &bpf_skb_load_bytes_relative_proto;
7983         case BPF_FUNC_skb_pull_data:
7984                 return &bpf_skb_pull_data_proto;
7985         case BPF_FUNC_csum_diff:
7986                 return &bpf_csum_diff_proto;
7987         case BPF_FUNC_csum_update:
7988                 return &bpf_csum_update_proto;
7989         case BPF_FUNC_csum_level:
7990                 return &bpf_csum_level_proto;
7991         case BPF_FUNC_l3_csum_replace:
7992                 return &bpf_l3_csum_replace_proto;
7993         case BPF_FUNC_l4_csum_replace:
7994                 return &bpf_l4_csum_replace_proto;
7995         case BPF_FUNC_clone_redirect:
7996                 return &bpf_clone_redirect_proto;
7997         case BPF_FUNC_get_cgroup_classid:
7998                 return &bpf_get_cgroup_classid_proto;
7999         case BPF_FUNC_skb_vlan_push:
8000                 return &bpf_skb_vlan_push_proto;
8001         case BPF_FUNC_skb_vlan_pop:
8002                 return &bpf_skb_vlan_pop_proto;
8003         case BPF_FUNC_skb_change_proto:
8004                 return &bpf_skb_change_proto_proto;
8005         case BPF_FUNC_skb_change_type:
8006                 return &bpf_skb_change_type_proto;
8007         case BPF_FUNC_skb_adjust_room:
8008                 return &bpf_skb_adjust_room_proto;
8009         case BPF_FUNC_skb_change_tail:
8010                 return &bpf_skb_change_tail_proto;
8011         case BPF_FUNC_skb_change_head:
8012                 return &bpf_skb_change_head_proto;
8013         case BPF_FUNC_skb_get_tunnel_key:
8014                 return &bpf_skb_get_tunnel_key_proto;
8015         case BPF_FUNC_skb_set_tunnel_key:
8016                 return bpf_get_skb_set_tunnel_proto(func_id);
8017         case BPF_FUNC_skb_get_tunnel_opt:
8018                 return &bpf_skb_get_tunnel_opt_proto;
8019         case BPF_FUNC_skb_set_tunnel_opt:
8020                 return bpf_get_skb_set_tunnel_proto(func_id);
8021         case BPF_FUNC_redirect:
8022                 return &bpf_redirect_proto;
8023         case BPF_FUNC_redirect_neigh:
8024                 return &bpf_redirect_neigh_proto;
8025         case BPF_FUNC_redirect_peer:
8026                 return &bpf_redirect_peer_proto;
8027         case BPF_FUNC_get_route_realm:
8028                 return &bpf_get_route_realm_proto;
8029         case BPF_FUNC_get_hash_recalc:
8030                 return &bpf_get_hash_recalc_proto;
8031         case BPF_FUNC_set_hash_invalid:
8032                 return &bpf_set_hash_invalid_proto;
8033         case BPF_FUNC_set_hash:
8034                 return &bpf_set_hash_proto;
8035         case BPF_FUNC_perf_event_output:
8036                 return &bpf_skb_event_output_proto;
8037         case BPF_FUNC_get_smp_processor_id:
8038                 return &bpf_get_smp_processor_id_proto;
8039         case BPF_FUNC_skb_under_cgroup:
8040                 return &bpf_skb_under_cgroup_proto;
8041         case BPF_FUNC_get_socket_cookie:
8042                 return &bpf_get_socket_cookie_proto;
8043         case BPF_FUNC_get_socket_uid:
8044                 return &bpf_get_socket_uid_proto;
8045         case BPF_FUNC_fib_lookup:
8046                 return &bpf_skb_fib_lookup_proto;
8047         case BPF_FUNC_check_mtu:
8048                 return &bpf_skb_check_mtu_proto;
8049         case BPF_FUNC_sk_fullsock:
8050                 return &bpf_sk_fullsock_proto;
8051         case BPF_FUNC_sk_storage_get:
8052                 return &bpf_sk_storage_get_proto;
8053         case BPF_FUNC_sk_storage_delete:
8054                 return &bpf_sk_storage_delete_proto;
8055 #ifdef CONFIG_XFRM
8056         case BPF_FUNC_skb_get_xfrm_state:
8057                 return &bpf_skb_get_xfrm_state_proto;
8058 #endif
8059 #ifdef CONFIG_CGROUP_NET_CLASSID
8060         case BPF_FUNC_skb_cgroup_classid:
8061                 return &bpf_skb_cgroup_classid_proto;
8062 #endif
8063 #ifdef CONFIG_SOCK_CGROUP_DATA
8064         case BPF_FUNC_skb_cgroup_id:
8065                 return &bpf_skb_cgroup_id_proto;
8066         case BPF_FUNC_skb_ancestor_cgroup_id:
8067                 return &bpf_skb_ancestor_cgroup_id_proto;
8068 #endif
8069 #ifdef CONFIG_INET
8070         case BPF_FUNC_sk_lookup_tcp:
8071                 return &bpf_tc_sk_lookup_tcp_proto;
8072         case BPF_FUNC_sk_lookup_udp:
8073                 return &bpf_tc_sk_lookup_udp_proto;
8074         case BPF_FUNC_sk_release:
8075                 return &bpf_sk_release_proto;
8076         case BPF_FUNC_tcp_sock:
8077                 return &bpf_tcp_sock_proto;
8078         case BPF_FUNC_get_listener_sock:
8079                 return &bpf_get_listener_sock_proto;
8080         case BPF_FUNC_skc_lookup_tcp:
8081                 return &bpf_tc_skc_lookup_tcp_proto;
8082         case BPF_FUNC_tcp_check_syncookie:
8083                 return &bpf_tcp_check_syncookie_proto;
8084         case BPF_FUNC_skb_ecn_set_ce:
8085                 return &bpf_skb_ecn_set_ce_proto;
8086         case BPF_FUNC_tcp_gen_syncookie:
8087                 return &bpf_tcp_gen_syncookie_proto;
8088         case BPF_FUNC_sk_assign:
8089                 return &bpf_sk_assign_proto;
8090         case BPF_FUNC_skb_set_tstamp:
8091                 return &bpf_skb_set_tstamp_proto;
8092 #ifdef CONFIG_SYN_COOKIES
8093         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8094                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8095         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8096                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8097         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8098                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8099         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8100                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8101 #endif
8102 #endif
8103         default:
8104                 return bpf_sk_base_func_proto(func_id);
8105         }
8106 }
8107
8108 static const struct bpf_func_proto *
8109 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8110 {
8111         switch (func_id) {
8112         case BPF_FUNC_perf_event_output:
8113                 return &bpf_xdp_event_output_proto;
8114         case BPF_FUNC_get_smp_processor_id:
8115                 return &bpf_get_smp_processor_id_proto;
8116         case BPF_FUNC_csum_diff:
8117                 return &bpf_csum_diff_proto;
8118         case BPF_FUNC_xdp_adjust_head:
8119                 return &bpf_xdp_adjust_head_proto;
8120         case BPF_FUNC_xdp_adjust_meta:
8121                 return &bpf_xdp_adjust_meta_proto;
8122         case BPF_FUNC_redirect:
8123                 return &bpf_xdp_redirect_proto;
8124         case BPF_FUNC_redirect_map:
8125                 return &bpf_xdp_redirect_map_proto;
8126         case BPF_FUNC_xdp_adjust_tail:
8127                 return &bpf_xdp_adjust_tail_proto;
8128         case BPF_FUNC_xdp_get_buff_len:
8129                 return &bpf_xdp_get_buff_len_proto;
8130         case BPF_FUNC_xdp_load_bytes:
8131                 return &bpf_xdp_load_bytes_proto;
8132         case BPF_FUNC_xdp_store_bytes:
8133                 return &bpf_xdp_store_bytes_proto;
8134         case BPF_FUNC_fib_lookup:
8135                 return &bpf_xdp_fib_lookup_proto;
8136         case BPF_FUNC_check_mtu:
8137                 return &bpf_xdp_check_mtu_proto;
8138 #ifdef CONFIG_INET
8139         case BPF_FUNC_sk_lookup_udp:
8140                 return &bpf_xdp_sk_lookup_udp_proto;
8141         case BPF_FUNC_sk_lookup_tcp:
8142                 return &bpf_xdp_sk_lookup_tcp_proto;
8143         case BPF_FUNC_sk_release:
8144                 return &bpf_sk_release_proto;
8145         case BPF_FUNC_skc_lookup_tcp:
8146                 return &bpf_xdp_skc_lookup_tcp_proto;
8147         case BPF_FUNC_tcp_check_syncookie:
8148                 return &bpf_tcp_check_syncookie_proto;
8149         case BPF_FUNC_tcp_gen_syncookie:
8150                 return &bpf_tcp_gen_syncookie_proto;
8151 #ifdef CONFIG_SYN_COOKIES
8152         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8153                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8154         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8155                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8156         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8157                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8158         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8159                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8160 #endif
8161 #endif
8162         default:
8163                 return bpf_sk_base_func_proto(func_id);
8164         }
8165
8166 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8167         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8168          * kfuncs are defined in two different modules, and we want to be able
8169          * to use them interchangably with the same BTF type ID. Because modules
8170          * can't de-duplicate BTF IDs between each other, we need the type to be
8171          * referenced in the vmlinux BTF or the verifier will get confused about
8172          * the different types. So we add this dummy type reference which will
8173          * be included in vmlinux BTF, allowing both modules to refer to the
8174          * same type ID.
8175          */
8176         BTF_TYPE_EMIT(struct nf_conn___init);
8177 #endif
8178 }
8179
8180 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8181 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8182
8183 static const struct bpf_func_proto *
8184 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8185 {
8186         const struct bpf_func_proto *func_proto;
8187
8188         func_proto = cgroup_common_func_proto(func_id, prog);
8189         if (func_proto)
8190                 return func_proto;
8191
8192         switch (func_id) {
8193         case BPF_FUNC_setsockopt:
8194                 return &bpf_sock_ops_setsockopt_proto;
8195         case BPF_FUNC_getsockopt:
8196                 return &bpf_sock_ops_getsockopt_proto;
8197         case BPF_FUNC_sock_ops_cb_flags_set:
8198                 return &bpf_sock_ops_cb_flags_set_proto;
8199         case BPF_FUNC_sock_map_update:
8200                 return &bpf_sock_map_update_proto;
8201         case BPF_FUNC_sock_hash_update:
8202                 return &bpf_sock_hash_update_proto;
8203         case BPF_FUNC_get_socket_cookie:
8204                 return &bpf_get_socket_cookie_sock_ops_proto;
8205         case BPF_FUNC_perf_event_output:
8206                 return &bpf_event_output_data_proto;
8207         case BPF_FUNC_sk_storage_get:
8208                 return &bpf_sk_storage_get_proto;
8209         case BPF_FUNC_sk_storage_delete:
8210                 return &bpf_sk_storage_delete_proto;
8211         case BPF_FUNC_get_netns_cookie:
8212                 return &bpf_get_netns_cookie_sock_ops_proto;
8213 #ifdef CONFIG_INET
8214         case BPF_FUNC_load_hdr_opt:
8215                 return &bpf_sock_ops_load_hdr_opt_proto;
8216         case BPF_FUNC_store_hdr_opt:
8217                 return &bpf_sock_ops_store_hdr_opt_proto;
8218         case BPF_FUNC_reserve_hdr_opt:
8219                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8220         case BPF_FUNC_tcp_sock:
8221                 return &bpf_tcp_sock_proto;
8222 #endif /* CONFIG_INET */
8223         default:
8224                 return bpf_sk_base_func_proto(func_id);
8225         }
8226 }
8227
8228 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8229 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8230
8231 static const struct bpf_func_proto *
8232 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8233 {
8234         switch (func_id) {
8235         case BPF_FUNC_msg_redirect_map:
8236                 return &bpf_msg_redirect_map_proto;
8237         case BPF_FUNC_msg_redirect_hash:
8238                 return &bpf_msg_redirect_hash_proto;
8239         case BPF_FUNC_msg_apply_bytes:
8240                 return &bpf_msg_apply_bytes_proto;
8241         case BPF_FUNC_msg_cork_bytes:
8242                 return &bpf_msg_cork_bytes_proto;
8243         case BPF_FUNC_msg_pull_data:
8244                 return &bpf_msg_pull_data_proto;
8245         case BPF_FUNC_msg_push_data:
8246                 return &bpf_msg_push_data_proto;
8247         case BPF_FUNC_msg_pop_data:
8248                 return &bpf_msg_pop_data_proto;
8249         case BPF_FUNC_perf_event_output:
8250                 return &bpf_event_output_data_proto;
8251         case BPF_FUNC_get_current_uid_gid:
8252                 return &bpf_get_current_uid_gid_proto;
8253         case BPF_FUNC_get_current_pid_tgid:
8254                 return &bpf_get_current_pid_tgid_proto;
8255         case BPF_FUNC_sk_storage_get:
8256                 return &bpf_sk_storage_get_proto;
8257         case BPF_FUNC_sk_storage_delete:
8258                 return &bpf_sk_storage_delete_proto;
8259         case BPF_FUNC_get_netns_cookie:
8260                 return &bpf_get_netns_cookie_sk_msg_proto;
8261 #ifdef CONFIG_CGROUP_NET_CLASSID
8262         case BPF_FUNC_get_cgroup_classid:
8263                 return &bpf_get_cgroup_classid_curr_proto;
8264 #endif
8265         default:
8266                 return bpf_sk_base_func_proto(func_id);
8267         }
8268 }
8269
8270 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8271 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8272
8273 static const struct bpf_func_proto *
8274 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8275 {
8276         switch (func_id) {
8277         case BPF_FUNC_skb_store_bytes:
8278                 return &bpf_skb_store_bytes_proto;
8279         case BPF_FUNC_skb_load_bytes:
8280                 return &bpf_skb_load_bytes_proto;
8281         case BPF_FUNC_skb_pull_data:
8282                 return &sk_skb_pull_data_proto;
8283         case BPF_FUNC_skb_change_tail:
8284                 return &sk_skb_change_tail_proto;
8285         case BPF_FUNC_skb_change_head:
8286                 return &sk_skb_change_head_proto;
8287         case BPF_FUNC_skb_adjust_room:
8288                 return &sk_skb_adjust_room_proto;
8289         case BPF_FUNC_get_socket_cookie:
8290                 return &bpf_get_socket_cookie_proto;
8291         case BPF_FUNC_get_socket_uid:
8292                 return &bpf_get_socket_uid_proto;
8293         case BPF_FUNC_sk_redirect_map:
8294                 return &bpf_sk_redirect_map_proto;
8295         case BPF_FUNC_sk_redirect_hash:
8296                 return &bpf_sk_redirect_hash_proto;
8297         case BPF_FUNC_perf_event_output:
8298                 return &bpf_skb_event_output_proto;
8299 #ifdef CONFIG_INET
8300         case BPF_FUNC_sk_lookup_tcp:
8301                 return &bpf_sk_lookup_tcp_proto;
8302         case BPF_FUNC_sk_lookup_udp:
8303                 return &bpf_sk_lookup_udp_proto;
8304         case BPF_FUNC_sk_release:
8305                 return &bpf_sk_release_proto;
8306         case BPF_FUNC_skc_lookup_tcp:
8307                 return &bpf_skc_lookup_tcp_proto;
8308 #endif
8309         default:
8310                 return bpf_sk_base_func_proto(func_id);
8311         }
8312 }
8313
8314 static const struct bpf_func_proto *
8315 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8316 {
8317         switch (func_id) {
8318         case BPF_FUNC_skb_load_bytes:
8319                 return &bpf_flow_dissector_load_bytes_proto;
8320         default:
8321                 return bpf_sk_base_func_proto(func_id);
8322         }
8323 }
8324
8325 static const struct bpf_func_proto *
8326 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8327 {
8328         switch (func_id) {
8329         case BPF_FUNC_skb_load_bytes:
8330                 return &bpf_skb_load_bytes_proto;
8331         case BPF_FUNC_skb_pull_data:
8332                 return &bpf_skb_pull_data_proto;
8333         case BPF_FUNC_csum_diff:
8334                 return &bpf_csum_diff_proto;
8335         case BPF_FUNC_get_cgroup_classid:
8336                 return &bpf_get_cgroup_classid_proto;
8337         case BPF_FUNC_get_route_realm:
8338                 return &bpf_get_route_realm_proto;
8339         case BPF_FUNC_get_hash_recalc:
8340                 return &bpf_get_hash_recalc_proto;
8341         case BPF_FUNC_perf_event_output:
8342                 return &bpf_skb_event_output_proto;
8343         case BPF_FUNC_get_smp_processor_id:
8344                 return &bpf_get_smp_processor_id_proto;
8345         case BPF_FUNC_skb_under_cgroup:
8346                 return &bpf_skb_under_cgroup_proto;
8347         default:
8348                 return bpf_sk_base_func_proto(func_id);
8349         }
8350 }
8351
8352 static const struct bpf_func_proto *
8353 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8354 {
8355         switch (func_id) {
8356         case BPF_FUNC_lwt_push_encap:
8357                 return &bpf_lwt_in_push_encap_proto;
8358         default:
8359                 return lwt_out_func_proto(func_id, prog);
8360         }
8361 }
8362
8363 static const struct bpf_func_proto *
8364 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8365 {
8366         switch (func_id) {
8367         case BPF_FUNC_skb_get_tunnel_key:
8368                 return &bpf_skb_get_tunnel_key_proto;
8369         case BPF_FUNC_skb_set_tunnel_key:
8370                 return bpf_get_skb_set_tunnel_proto(func_id);
8371         case BPF_FUNC_skb_get_tunnel_opt:
8372                 return &bpf_skb_get_tunnel_opt_proto;
8373         case BPF_FUNC_skb_set_tunnel_opt:
8374                 return bpf_get_skb_set_tunnel_proto(func_id);
8375         case BPF_FUNC_redirect:
8376                 return &bpf_redirect_proto;
8377         case BPF_FUNC_clone_redirect:
8378                 return &bpf_clone_redirect_proto;
8379         case BPF_FUNC_skb_change_tail:
8380                 return &bpf_skb_change_tail_proto;
8381         case BPF_FUNC_skb_change_head:
8382                 return &bpf_skb_change_head_proto;
8383         case BPF_FUNC_skb_store_bytes:
8384                 return &bpf_skb_store_bytes_proto;
8385         case BPF_FUNC_csum_update:
8386                 return &bpf_csum_update_proto;
8387         case BPF_FUNC_csum_level:
8388                 return &bpf_csum_level_proto;
8389         case BPF_FUNC_l3_csum_replace:
8390                 return &bpf_l3_csum_replace_proto;
8391         case BPF_FUNC_l4_csum_replace:
8392                 return &bpf_l4_csum_replace_proto;
8393         case BPF_FUNC_set_hash_invalid:
8394                 return &bpf_set_hash_invalid_proto;
8395         case BPF_FUNC_lwt_push_encap:
8396                 return &bpf_lwt_xmit_push_encap_proto;
8397         default:
8398                 return lwt_out_func_proto(func_id, prog);
8399         }
8400 }
8401
8402 static const struct bpf_func_proto *
8403 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8404 {
8405         switch (func_id) {
8406 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8407         case BPF_FUNC_lwt_seg6_store_bytes:
8408                 return &bpf_lwt_seg6_store_bytes_proto;
8409         case BPF_FUNC_lwt_seg6_action:
8410                 return &bpf_lwt_seg6_action_proto;
8411         case BPF_FUNC_lwt_seg6_adjust_srh:
8412                 return &bpf_lwt_seg6_adjust_srh_proto;
8413 #endif
8414         default:
8415                 return lwt_out_func_proto(func_id, prog);
8416         }
8417 }
8418
8419 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8420                                     const struct bpf_prog *prog,
8421                                     struct bpf_insn_access_aux *info)
8422 {
8423         const int size_default = sizeof(__u32);
8424
8425         if (off < 0 || off >= sizeof(struct __sk_buff))
8426                 return false;
8427
8428         /* The verifier guarantees that size > 0. */
8429         if (off % size != 0)
8430                 return false;
8431
8432         switch (off) {
8433         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8434                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8435                         return false;
8436                 break;
8437         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8438         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8439         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8440         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8441         case bpf_ctx_range(struct __sk_buff, data):
8442         case bpf_ctx_range(struct __sk_buff, data_meta):
8443         case bpf_ctx_range(struct __sk_buff, data_end):
8444                 if (size != size_default)
8445                         return false;
8446                 break;
8447         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8448                 return false;
8449         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8450                 if (type == BPF_WRITE || size != sizeof(__u64))
8451                         return false;
8452                 break;
8453         case bpf_ctx_range(struct __sk_buff, tstamp):
8454                 if (size != sizeof(__u64))
8455                         return false;
8456                 break;
8457         case offsetof(struct __sk_buff, sk):
8458                 if (type == BPF_WRITE || size != sizeof(__u64))
8459                         return false;
8460                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8461                 break;
8462         case offsetof(struct __sk_buff, tstamp_type):
8463                 return false;
8464         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8465                 /* Explicitly prohibit access to padding in __sk_buff. */
8466                 return false;
8467         default:
8468                 /* Only narrow read access allowed for now. */
8469                 if (type == BPF_WRITE) {
8470                         if (size != size_default)
8471                                 return false;
8472                 } else {
8473                         bpf_ctx_record_field_size(info, size_default);
8474                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8475                                 return false;
8476                 }
8477         }
8478
8479         return true;
8480 }
8481
8482 static bool sk_filter_is_valid_access(int off, int size,
8483                                       enum bpf_access_type type,
8484                                       const struct bpf_prog *prog,
8485                                       struct bpf_insn_access_aux *info)
8486 {
8487         switch (off) {
8488         case bpf_ctx_range(struct __sk_buff, tc_classid):
8489         case bpf_ctx_range(struct __sk_buff, data):
8490         case bpf_ctx_range(struct __sk_buff, data_meta):
8491         case bpf_ctx_range(struct __sk_buff, data_end):
8492         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8493         case bpf_ctx_range(struct __sk_buff, tstamp):
8494         case bpf_ctx_range(struct __sk_buff, wire_len):
8495         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8496                 return false;
8497         }
8498
8499         if (type == BPF_WRITE) {
8500                 switch (off) {
8501                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8502                         break;
8503                 default:
8504                         return false;
8505                 }
8506         }
8507
8508         return bpf_skb_is_valid_access(off, size, type, prog, info);
8509 }
8510
8511 static bool cg_skb_is_valid_access(int off, int size,
8512                                    enum bpf_access_type type,
8513                                    const struct bpf_prog *prog,
8514                                    struct bpf_insn_access_aux *info)
8515 {
8516         switch (off) {
8517         case bpf_ctx_range(struct __sk_buff, tc_classid):
8518         case bpf_ctx_range(struct __sk_buff, data_meta):
8519         case bpf_ctx_range(struct __sk_buff, wire_len):
8520                 return false;
8521         case bpf_ctx_range(struct __sk_buff, data):
8522         case bpf_ctx_range(struct __sk_buff, data_end):
8523                 if (!bpf_capable())
8524                         return false;
8525                 break;
8526         }
8527
8528         if (type == BPF_WRITE) {
8529                 switch (off) {
8530                 case bpf_ctx_range(struct __sk_buff, mark):
8531                 case bpf_ctx_range(struct __sk_buff, priority):
8532                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8533                         break;
8534                 case bpf_ctx_range(struct __sk_buff, tstamp):
8535                         if (!bpf_capable())
8536                                 return false;
8537                         break;
8538                 default:
8539                         return false;
8540                 }
8541         }
8542
8543         switch (off) {
8544         case bpf_ctx_range(struct __sk_buff, data):
8545                 info->reg_type = PTR_TO_PACKET;
8546                 break;
8547         case bpf_ctx_range(struct __sk_buff, data_end):
8548                 info->reg_type = PTR_TO_PACKET_END;
8549                 break;
8550         }
8551
8552         return bpf_skb_is_valid_access(off, size, type, prog, info);
8553 }
8554
8555 static bool lwt_is_valid_access(int off, int size,
8556                                 enum bpf_access_type type,
8557                                 const struct bpf_prog *prog,
8558                                 struct bpf_insn_access_aux *info)
8559 {
8560         switch (off) {
8561         case bpf_ctx_range(struct __sk_buff, tc_classid):
8562         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8563         case bpf_ctx_range(struct __sk_buff, data_meta):
8564         case bpf_ctx_range(struct __sk_buff, tstamp):
8565         case bpf_ctx_range(struct __sk_buff, wire_len):
8566         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8567                 return false;
8568         }
8569
8570         if (type == BPF_WRITE) {
8571                 switch (off) {
8572                 case bpf_ctx_range(struct __sk_buff, mark):
8573                 case bpf_ctx_range(struct __sk_buff, priority):
8574                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8575                         break;
8576                 default:
8577                         return false;
8578                 }
8579         }
8580
8581         switch (off) {
8582         case bpf_ctx_range(struct __sk_buff, data):
8583                 info->reg_type = PTR_TO_PACKET;
8584                 break;
8585         case bpf_ctx_range(struct __sk_buff, data_end):
8586                 info->reg_type = PTR_TO_PACKET_END;
8587                 break;
8588         }
8589
8590         return bpf_skb_is_valid_access(off, size, type, prog, info);
8591 }
8592
8593 /* Attach type specific accesses */
8594 static bool __sock_filter_check_attach_type(int off,
8595                                             enum bpf_access_type access_type,
8596                                             enum bpf_attach_type attach_type)
8597 {
8598         switch (off) {
8599         case offsetof(struct bpf_sock, bound_dev_if):
8600         case offsetof(struct bpf_sock, mark):
8601         case offsetof(struct bpf_sock, priority):
8602                 switch (attach_type) {
8603                 case BPF_CGROUP_INET_SOCK_CREATE:
8604                 case BPF_CGROUP_INET_SOCK_RELEASE:
8605                         goto full_access;
8606                 default:
8607                         return false;
8608                 }
8609         case bpf_ctx_range(struct bpf_sock, src_ip4):
8610                 switch (attach_type) {
8611                 case BPF_CGROUP_INET4_POST_BIND:
8612                         goto read_only;
8613                 default:
8614                         return false;
8615                 }
8616         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8617                 switch (attach_type) {
8618                 case BPF_CGROUP_INET6_POST_BIND:
8619                         goto read_only;
8620                 default:
8621                         return false;
8622                 }
8623         case bpf_ctx_range(struct bpf_sock, src_port):
8624                 switch (attach_type) {
8625                 case BPF_CGROUP_INET4_POST_BIND:
8626                 case BPF_CGROUP_INET6_POST_BIND:
8627                         goto read_only;
8628                 default:
8629                         return false;
8630                 }
8631         }
8632 read_only:
8633         return access_type == BPF_READ;
8634 full_access:
8635         return true;
8636 }
8637
8638 bool bpf_sock_common_is_valid_access(int off, int size,
8639                                      enum bpf_access_type type,
8640                                      struct bpf_insn_access_aux *info)
8641 {
8642         switch (off) {
8643         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8644                 return false;
8645         default:
8646                 return bpf_sock_is_valid_access(off, size, type, info);
8647         }
8648 }
8649
8650 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8651                               struct bpf_insn_access_aux *info)
8652 {
8653         const int size_default = sizeof(__u32);
8654         int field_size;
8655
8656         if (off < 0 || off >= sizeof(struct bpf_sock))
8657                 return false;
8658         if (off % size != 0)
8659                 return false;
8660
8661         switch (off) {
8662         case offsetof(struct bpf_sock, state):
8663         case offsetof(struct bpf_sock, family):
8664         case offsetof(struct bpf_sock, type):
8665         case offsetof(struct bpf_sock, protocol):
8666         case offsetof(struct bpf_sock, src_port):
8667         case offsetof(struct bpf_sock, rx_queue_mapping):
8668         case bpf_ctx_range(struct bpf_sock, src_ip4):
8669         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8670         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8671         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8672                 bpf_ctx_record_field_size(info, size_default);
8673                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8674         case bpf_ctx_range(struct bpf_sock, dst_port):
8675                 field_size = size == size_default ?
8676                         size_default : sizeof_field(struct bpf_sock, dst_port);
8677                 bpf_ctx_record_field_size(info, field_size);
8678                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8679         case offsetofend(struct bpf_sock, dst_port) ...
8680              offsetof(struct bpf_sock, dst_ip4) - 1:
8681                 return false;
8682         }
8683
8684         return size == size_default;
8685 }
8686
8687 static bool sock_filter_is_valid_access(int off, int size,
8688                                         enum bpf_access_type type,
8689                                         const struct bpf_prog *prog,
8690                                         struct bpf_insn_access_aux *info)
8691 {
8692         if (!bpf_sock_is_valid_access(off, size, type, info))
8693                 return false;
8694         return __sock_filter_check_attach_type(off, type,
8695                                                prog->expected_attach_type);
8696 }
8697
8698 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8699                              const struct bpf_prog *prog)
8700 {
8701         /* Neither direct read nor direct write requires any preliminary
8702          * action.
8703          */
8704         return 0;
8705 }
8706
8707 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8708                                 const struct bpf_prog *prog, int drop_verdict)
8709 {
8710         struct bpf_insn *insn = insn_buf;
8711
8712         if (!direct_write)
8713                 return 0;
8714
8715         /* if (!skb->cloned)
8716          *       goto start;
8717          *
8718          * (Fast-path, otherwise approximation that we might be
8719          *  a clone, do the rest in helper.)
8720          */
8721         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8722         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8723         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8724
8725         /* ret = bpf_skb_pull_data(skb, 0); */
8726         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8727         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8728         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8729                                BPF_FUNC_skb_pull_data);
8730         /* if (!ret)
8731          *      goto restore;
8732          * return TC_ACT_SHOT;
8733          */
8734         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8735         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8736         *insn++ = BPF_EXIT_INSN();
8737
8738         /* restore: */
8739         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8740         /* start: */
8741         *insn++ = prog->insnsi[0];
8742
8743         return insn - insn_buf;
8744 }
8745
8746 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8747                           struct bpf_insn *insn_buf)
8748 {
8749         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8750         struct bpf_insn *insn = insn_buf;
8751
8752         if (!indirect) {
8753                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8754         } else {
8755                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8756                 if (orig->imm)
8757                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8758         }
8759         /* We're guaranteed here that CTX is in R6. */
8760         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8761
8762         switch (BPF_SIZE(orig->code)) {
8763         case BPF_B:
8764                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8765                 break;
8766         case BPF_H:
8767                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8768                 break;
8769         case BPF_W:
8770                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8771                 break;
8772         }
8773
8774         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8775         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8776         *insn++ = BPF_EXIT_INSN();
8777
8778         return insn - insn_buf;
8779 }
8780
8781 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8782                                const struct bpf_prog *prog)
8783 {
8784         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8785 }
8786
8787 static bool tc_cls_act_is_valid_access(int off, int size,
8788                                        enum bpf_access_type type,
8789                                        const struct bpf_prog *prog,
8790                                        struct bpf_insn_access_aux *info)
8791 {
8792         if (type == BPF_WRITE) {
8793                 switch (off) {
8794                 case bpf_ctx_range(struct __sk_buff, mark):
8795                 case bpf_ctx_range(struct __sk_buff, tc_index):
8796                 case bpf_ctx_range(struct __sk_buff, priority):
8797                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8798                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8799                 case bpf_ctx_range(struct __sk_buff, tstamp):
8800                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8801                         break;
8802                 default:
8803                         return false;
8804                 }
8805         }
8806
8807         switch (off) {
8808         case bpf_ctx_range(struct __sk_buff, data):
8809                 info->reg_type = PTR_TO_PACKET;
8810                 break;
8811         case bpf_ctx_range(struct __sk_buff, data_meta):
8812                 info->reg_type = PTR_TO_PACKET_META;
8813                 break;
8814         case bpf_ctx_range(struct __sk_buff, data_end):
8815                 info->reg_type = PTR_TO_PACKET_END;
8816                 break;
8817         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8818                 return false;
8819         case offsetof(struct __sk_buff, tstamp_type):
8820                 /* The convert_ctx_access() on reading and writing
8821                  * __sk_buff->tstamp depends on whether the bpf prog
8822                  * has used __sk_buff->tstamp_type or not.
8823                  * Thus, we need to set prog->tstamp_type_access
8824                  * earlier during is_valid_access() here.
8825                  */
8826                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8827                 return size == sizeof(__u8);
8828         }
8829
8830         return bpf_skb_is_valid_access(off, size, type, prog, info);
8831 }
8832
8833 DEFINE_MUTEX(nf_conn_btf_access_lock);
8834 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8835
8836 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8837                               const struct bpf_reg_state *reg,
8838                               int off, int size);
8839 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8840
8841 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8842                                         const struct bpf_reg_state *reg,
8843                                         int off, int size)
8844 {
8845         int ret = -EACCES;
8846
8847         mutex_lock(&nf_conn_btf_access_lock);
8848         if (nfct_btf_struct_access)
8849                 ret = nfct_btf_struct_access(log, reg, off, size);
8850         mutex_unlock(&nf_conn_btf_access_lock);
8851
8852         return ret;
8853 }
8854
8855 static bool __is_valid_xdp_access(int off, int size)
8856 {
8857         if (off < 0 || off >= sizeof(struct xdp_md))
8858                 return false;
8859         if (off % size != 0)
8860                 return false;
8861         if (size != sizeof(__u32))
8862                 return false;
8863
8864         return true;
8865 }
8866
8867 static bool xdp_is_valid_access(int off, int size,
8868                                 enum bpf_access_type type,
8869                                 const struct bpf_prog *prog,
8870                                 struct bpf_insn_access_aux *info)
8871 {
8872         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8873                 switch (off) {
8874                 case offsetof(struct xdp_md, egress_ifindex):
8875                         return false;
8876                 }
8877         }
8878
8879         if (type == BPF_WRITE) {
8880                 if (bpf_prog_is_offloaded(prog->aux)) {
8881                         switch (off) {
8882                         case offsetof(struct xdp_md, rx_queue_index):
8883                                 return __is_valid_xdp_access(off, size);
8884                         }
8885                 }
8886                 return false;
8887         }
8888
8889         switch (off) {
8890         case offsetof(struct xdp_md, data):
8891                 info->reg_type = PTR_TO_PACKET;
8892                 break;
8893         case offsetof(struct xdp_md, data_meta):
8894                 info->reg_type = PTR_TO_PACKET_META;
8895                 break;
8896         case offsetof(struct xdp_md, data_end):
8897                 info->reg_type = PTR_TO_PACKET_END;
8898                 break;
8899         }
8900
8901         return __is_valid_xdp_access(off, size);
8902 }
8903
8904 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8905 {
8906         const u32 act_max = XDP_REDIRECT;
8907
8908         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8909                      act > act_max ? "Illegal" : "Driver unsupported",
8910                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8911 }
8912 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8913
8914 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8915                                  const struct bpf_reg_state *reg,
8916                                  int off, int size)
8917 {
8918         int ret = -EACCES;
8919
8920         mutex_lock(&nf_conn_btf_access_lock);
8921         if (nfct_btf_struct_access)
8922                 ret = nfct_btf_struct_access(log, reg, off, size);
8923         mutex_unlock(&nf_conn_btf_access_lock);
8924
8925         return ret;
8926 }
8927
8928 static bool sock_addr_is_valid_access(int off, int size,
8929                                       enum bpf_access_type type,
8930                                       const struct bpf_prog *prog,
8931                                       struct bpf_insn_access_aux *info)
8932 {
8933         const int size_default = sizeof(__u32);
8934
8935         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8936                 return false;
8937         if (off % size != 0)
8938                 return false;
8939
8940         /* Disallow access to IPv6 fields from IPv4 contex and vise
8941          * versa.
8942          */
8943         switch (off) {
8944         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8945                 switch (prog->expected_attach_type) {
8946                 case BPF_CGROUP_INET4_BIND:
8947                 case BPF_CGROUP_INET4_CONNECT:
8948                 case BPF_CGROUP_INET4_GETPEERNAME:
8949                 case BPF_CGROUP_INET4_GETSOCKNAME:
8950                 case BPF_CGROUP_UDP4_SENDMSG:
8951                 case BPF_CGROUP_UDP4_RECVMSG:
8952                         break;
8953                 default:
8954                         return false;
8955                 }
8956                 break;
8957         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8958                 switch (prog->expected_attach_type) {
8959                 case BPF_CGROUP_INET6_BIND:
8960                 case BPF_CGROUP_INET6_CONNECT:
8961                 case BPF_CGROUP_INET6_GETPEERNAME:
8962                 case BPF_CGROUP_INET6_GETSOCKNAME:
8963                 case BPF_CGROUP_UDP6_SENDMSG:
8964                 case BPF_CGROUP_UDP6_RECVMSG:
8965                         break;
8966                 default:
8967                         return false;
8968                 }
8969                 break;
8970         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8971                 switch (prog->expected_attach_type) {
8972                 case BPF_CGROUP_UDP4_SENDMSG:
8973                         break;
8974                 default:
8975                         return false;
8976                 }
8977                 break;
8978         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8979                                 msg_src_ip6[3]):
8980                 switch (prog->expected_attach_type) {
8981                 case BPF_CGROUP_UDP6_SENDMSG:
8982                         break;
8983                 default:
8984                         return false;
8985                 }
8986                 break;
8987         }
8988
8989         switch (off) {
8990         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8991         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8992         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8993         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8994                                 msg_src_ip6[3]):
8995         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8996                 if (type == BPF_READ) {
8997                         bpf_ctx_record_field_size(info, size_default);
8998
8999                         if (bpf_ctx_wide_access_ok(off, size,
9000                                                    struct bpf_sock_addr,
9001                                                    user_ip6))
9002                                 return true;
9003
9004                         if (bpf_ctx_wide_access_ok(off, size,
9005                                                    struct bpf_sock_addr,
9006                                                    msg_src_ip6))
9007                                 return true;
9008
9009                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9010                                 return false;
9011                 } else {
9012                         if (bpf_ctx_wide_access_ok(off, size,
9013                                                    struct bpf_sock_addr,
9014                                                    user_ip6))
9015                                 return true;
9016
9017                         if (bpf_ctx_wide_access_ok(off, size,
9018                                                    struct bpf_sock_addr,
9019                                                    msg_src_ip6))
9020                                 return true;
9021
9022                         if (size != size_default)
9023                                 return false;
9024                 }
9025                 break;
9026         case offsetof(struct bpf_sock_addr, sk):
9027                 if (type != BPF_READ)
9028                         return false;
9029                 if (size != sizeof(__u64))
9030                         return false;
9031                 info->reg_type = PTR_TO_SOCKET;
9032                 break;
9033         default:
9034                 if (type == BPF_READ) {
9035                         if (size != size_default)
9036                                 return false;
9037                 } else {
9038                         return false;
9039                 }
9040         }
9041
9042         return true;
9043 }
9044
9045 static bool sock_ops_is_valid_access(int off, int size,
9046                                      enum bpf_access_type type,
9047                                      const struct bpf_prog *prog,
9048                                      struct bpf_insn_access_aux *info)
9049 {
9050         const int size_default = sizeof(__u32);
9051
9052         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9053                 return false;
9054
9055         /* The verifier guarantees that size > 0. */
9056         if (off % size != 0)
9057                 return false;
9058
9059         if (type == BPF_WRITE) {
9060                 switch (off) {
9061                 case offsetof(struct bpf_sock_ops, reply):
9062                 case offsetof(struct bpf_sock_ops, sk_txhash):
9063                         if (size != size_default)
9064                                 return false;
9065                         break;
9066                 default:
9067                         return false;
9068                 }
9069         } else {
9070                 switch (off) {
9071                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9072                                         bytes_acked):
9073                         if (size != sizeof(__u64))
9074                                 return false;
9075                         break;
9076                 case offsetof(struct bpf_sock_ops, sk):
9077                         if (size != sizeof(__u64))
9078                                 return false;
9079                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
9080                         break;
9081                 case offsetof(struct bpf_sock_ops, skb_data):
9082                         if (size != sizeof(__u64))
9083                                 return false;
9084                         info->reg_type = PTR_TO_PACKET;
9085                         break;
9086                 case offsetof(struct bpf_sock_ops, skb_data_end):
9087                         if (size != sizeof(__u64))
9088                                 return false;
9089                         info->reg_type = PTR_TO_PACKET_END;
9090                         break;
9091                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9092                         bpf_ctx_record_field_size(info, size_default);
9093                         return bpf_ctx_narrow_access_ok(off, size,
9094                                                         size_default);
9095                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9096                         if (size != sizeof(__u64))
9097                                 return false;
9098                         break;
9099                 default:
9100                         if (size != size_default)
9101                                 return false;
9102                         break;
9103                 }
9104         }
9105
9106         return true;
9107 }
9108
9109 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9110                            const struct bpf_prog *prog)
9111 {
9112         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9113 }
9114
9115 static bool sk_skb_is_valid_access(int off, int size,
9116                                    enum bpf_access_type type,
9117                                    const struct bpf_prog *prog,
9118                                    struct bpf_insn_access_aux *info)
9119 {
9120         switch (off) {
9121         case bpf_ctx_range(struct __sk_buff, tc_classid):
9122         case bpf_ctx_range(struct __sk_buff, data_meta):
9123         case bpf_ctx_range(struct __sk_buff, tstamp):
9124         case bpf_ctx_range(struct __sk_buff, wire_len):
9125         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9126                 return false;
9127         }
9128
9129         if (type == BPF_WRITE) {
9130                 switch (off) {
9131                 case bpf_ctx_range(struct __sk_buff, tc_index):
9132                 case bpf_ctx_range(struct __sk_buff, priority):
9133                         break;
9134                 default:
9135                         return false;
9136                 }
9137         }
9138
9139         switch (off) {
9140         case bpf_ctx_range(struct __sk_buff, mark):
9141                 return false;
9142         case bpf_ctx_range(struct __sk_buff, data):
9143                 info->reg_type = PTR_TO_PACKET;
9144                 break;
9145         case bpf_ctx_range(struct __sk_buff, data_end):
9146                 info->reg_type = PTR_TO_PACKET_END;
9147                 break;
9148         }
9149
9150         return bpf_skb_is_valid_access(off, size, type, prog, info);
9151 }
9152
9153 static bool sk_msg_is_valid_access(int off, int size,
9154                                    enum bpf_access_type type,
9155                                    const struct bpf_prog *prog,
9156                                    struct bpf_insn_access_aux *info)
9157 {
9158         if (type == BPF_WRITE)
9159                 return false;
9160
9161         if (off % size != 0)
9162                 return false;
9163
9164         switch (off) {
9165         case offsetof(struct sk_msg_md, data):
9166                 info->reg_type = PTR_TO_PACKET;
9167                 if (size != sizeof(__u64))
9168                         return false;
9169                 break;
9170         case offsetof(struct sk_msg_md, data_end):
9171                 info->reg_type = PTR_TO_PACKET_END;
9172                 if (size != sizeof(__u64))
9173                         return false;
9174                 break;
9175         case offsetof(struct sk_msg_md, sk):
9176                 if (size != sizeof(__u64))
9177                         return false;
9178                 info->reg_type = PTR_TO_SOCKET;
9179                 break;
9180         case bpf_ctx_range(struct sk_msg_md, family):
9181         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9182         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9183         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9184         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9185         case bpf_ctx_range(struct sk_msg_md, remote_port):
9186         case bpf_ctx_range(struct sk_msg_md, local_port):
9187         case bpf_ctx_range(struct sk_msg_md, size):
9188                 if (size != sizeof(__u32))
9189                         return false;
9190                 break;
9191         default:
9192                 return false;
9193         }
9194         return true;
9195 }
9196
9197 static bool flow_dissector_is_valid_access(int off, int size,
9198                                            enum bpf_access_type type,
9199                                            const struct bpf_prog *prog,
9200                                            struct bpf_insn_access_aux *info)
9201 {
9202         const int size_default = sizeof(__u32);
9203
9204         if (off < 0 || off >= sizeof(struct __sk_buff))
9205                 return false;
9206
9207         if (type == BPF_WRITE)
9208                 return false;
9209
9210         switch (off) {
9211         case bpf_ctx_range(struct __sk_buff, data):
9212                 if (size != size_default)
9213                         return false;
9214                 info->reg_type = PTR_TO_PACKET;
9215                 return true;
9216         case bpf_ctx_range(struct __sk_buff, data_end):
9217                 if (size != size_default)
9218                         return false;
9219                 info->reg_type = PTR_TO_PACKET_END;
9220                 return true;
9221         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9222                 if (size != sizeof(__u64))
9223                         return false;
9224                 info->reg_type = PTR_TO_FLOW_KEYS;
9225                 return true;
9226         default:
9227                 return false;
9228         }
9229 }
9230
9231 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9232                                              const struct bpf_insn *si,
9233                                              struct bpf_insn *insn_buf,
9234                                              struct bpf_prog *prog,
9235                                              u32 *target_size)
9236
9237 {
9238         struct bpf_insn *insn = insn_buf;
9239
9240         switch (si->off) {
9241         case offsetof(struct __sk_buff, data):
9242                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9243                                       si->dst_reg, si->src_reg,
9244                                       offsetof(struct bpf_flow_dissector, data));
9245                 break;
9246
9247         case offsetof(struct __sk_buff, data_end):
9248                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9249                                       si->dst_reg, si->src_reg,
9250                                       offsetof(struct bpf_flow_dissector, data_end));
9251                 break;
9252
9253         case offsetof(struct __sk_buff, flow_keys):
9254                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9255                                       si->dst_reg, si->src_reg,
9256                                       offsetof(struct bpf_flow_dissector, flow_keys));
9257                 break;
9258         }
9259
9260         return insn - insn_buf;
9261 }
9262
9263 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9264                                                      struct bpf_insn *insn)
9265 {
9266         __u8 value_reg = si->dst_reg;
9267         __u8 skb_reg = si->src_reg;
9268         /* AX is needed because src_reg and dst_reg could be the same */
9269         __u8 tmp_reg = BPF_REG_AX;
9270
9271         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9272                               SKB_BF_MONO_TC_OFFSET);
9273         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9274                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9275         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9276         *insn++ = BPF_JMP_A(1);
9277         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9278
9279         return insn;
9280 }
9281
9282 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9283                                                   struct bpf_insn *insn)
9284 {
9285         /* si->dst_reg = skb_shinfo(SKB); */
9286 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9287         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9288                               BPF_REG_AX, skb_reg,
9289                               offsetof(struct sk_buff, end));
9290         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9291                               dst_reg, skb_reg,
9292                               offsetof(struct sk_buff, head));
9293         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9294 #else
9295         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9296                               dst_reg, skb_reg,
9297                               offsetof(struct sk_buff, end));
9298 #endif
9299
9300         return insn;
9301 }
9302
9303 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9304                                                 const struct bpf_insn *si,
9305                                                 struct bpf_insn *insn)
9306 {
9307         __u8 value_reg = si->dst_reg;
9308         __u8 skb_reg = si->src_reg;
9309
9310 #ifdef CONFIG_NET_XGRESS
9311         /* If the tstamp_type is read,
9312          * the bpf prog is aware the tstamp could have delivery time.
9313          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9314          */
9315         if (!prog->tstamp_type_access) {
9316                 /* AX is needed because src_reg and dst_reg could be the same */
9317                 __u8 tmp_reg = BPF_REG_AX;
9318
9319                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9320                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9321                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9322                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9323                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9324                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9325                  * read 0 as the (rcv) timestamp.
9326                  */
9327                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9328                 *insn++ = BPF_JMP_A(1);
9329         }
9330 #endif
9331
9332         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9333                               offsetof(struct sk_buff, tstamp));
9334         return insn;
9335 }
9336
9337 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9338                                                  const struct bpf_insn *si,
9339                                                  struct bpf_insn *insn)
9340 {
9341         __u8 value_reg = si->src_reg;
9342         __u8 skb_reg = si->dst_reg;
9343
9344 #ifdef CONFIG_NET_XGRESS
9345         /* If the tstamp_type is read,
9346          * the bpf prog is aware the tstamp could have delivery time.
9347          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9348          * Otherwise, writing at ingress will have to clear the
9349          * mono_delivery_time bit also.
9350          */
9351         if (!prog->tstamp_type_access) {
9352                 __u8 tmp_reg = BPF_REG_AX;
9353
9354                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9355                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9356                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9357                 /* goto <store> */
9358                 *insn++ = BPF_JMP_A(2);
9359                 /* <clear>: mono_delivery_time */
9360                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9361                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9362         }
9363 #endif
9364
9365         /* <store>: skb->tstamp = tstamp */
9366         *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9367                                skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9368         return insn;
9369 }
9370
9371 #define BPF_EMIT_STORE(size, si, off)                                   \
9372         BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM,          \
9373                      (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9374
9375 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9376                                   const struct bpf_insn *si,
9377                                   struct bpf_insn *insn_buf,
9378                                   struct bpf_prog *prog, u32 *target_size)
9379 {
9380         struct bpf_insn *insn = insn_buf;
9381         int off;
9382
9383         switch (si->off) {
9384         case offsetof(struct __sk_buff, len):
9385                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9386                                       bpf_target_off(struct sk_buff, len, 4,
9387                                                      target_size));
9388                 break;
9389
9390         case offsetof(struct __sk_buff, protocol):
9391                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9392                                       bpf_target_off(struct sk_buff, protocol, 2,
9393                                                      target_size));
9394                 break;
9395
9396         case offsetof(struct __sk_buff, vlan_proto):
9397                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9398                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9399                                                      target_size));
9400                 break;
9401
9402         case offsetof(struct __sk_buff, priority):
9403                 if (type == BPF_WRITE)
9404                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9405                                                  bpf_target_off(struct sk_buff, priority, 4,
9406                                                                 target_size));
9407                 else
9408                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9409                                               bpf_target_off(struct sk_buff, priority, 4,
9410                                                              target_size));
9411                 break;
9412
9413         case offsetof(struct __sk_buff, ingress_ifindex):
9414                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9415                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9416                                                      target_size));
9417                 break;
9418
9419         case offsetof(struct __sk_buff, ifindex):
9420                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9421                                       si->dst_reg, si->src_reg,
9422                                       offsetof(struct sk_buff, dev));
9423                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9425                                       bpf_target_off(struct net_device, ifindex, 4,
9426                                                      target_size));
9427                 break;
9428
9429         case offsetof(struct __sk_buff, hash):
9430                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9431                                       bpf_target_off(struct sk_buff, hash, 4,
9432                                                      target_size));
9433                 break;
9434
9435         case offsetof(struct __sk_buff, mark):
9436                 if (type == BPF_WRITE)
9437                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9438                                                  bpf_target_off(struct sk_buff, mark, 4,
9439                                                                 target_size));
9440                 else
9441                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9442                                               bpf_target_off(struct sk_buff, mark, 4,
9443                                                              target_size));
9444                 break;
9445
9446         case offsetof(struct __sk_buff, pkt_type):
9447                 *target_size = 1;
9448                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9449                                       PKT_TYPE_OFFSET);
9450                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9451 #ifdef __BIG_ENDIAN_BITFIELD
9452                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9453 #endif
9454                 break;
9455
9456         case offsetof(struct __sk_buff, queue_mapping):
9457                 if (type == BPF_WRITE) {
9458                         u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9459
9460                         if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9461                                 *insn++ = BPF_JMP_A(0); /* noop */
9462                                 break;
9463                         }
9464
9465                         if (BPF_CLASS(si->code) == BPF_STX)
9466                                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9467                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9468                 } else {
9469                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9470                                               bpf_target_off(struct sk_buff,
9471                                                              queue_mapping,
9472                                                              2, target_size));
9473                 }
9474                 break;
9475
9476         case offsetof(struct __sk_buff, vlan_present):
9477                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9478                                       bpf_target_off(struct sk_buff,
9479                                                      vlan_all, 4, target_size));
9480                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9481                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9482                 break;
9483
9484         case offsetof(struct __sk_buff, vlan_tci):
9485                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9486                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9487                                                      target_size));
9488                 break;
9489
9490         case offsetof(struct __sk_buff, cb[0]) ...
9491              offsetofend(struct __sk_buff, cb[4]) - 1:
9492                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9493                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9494                               offsetof(struct qdisc_skb_cb, data)) %
9495                              sizeof(__u64));
9496
9497                 prog->cb_access = 1;
9498                 off  = si->off;
9499                 off -= offsetof(struct __sk_buff, cb[0]);
9500                 off += offsetof(struct sk_buff, cb);
9501                 off += offsetof(struct qdisc_skb_cb, data);
9502                 if (type == BPF_WRITE)
9503                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9504                 else
9505                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9506                                               si->src_reg, off);
9507                 break;
9508
9509         case offsetof(struct __sk_buff, tc_classid):
9510                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9511
9512                 off  = si->off;
9513                 off -= offsetof(struct __sk_buff, tc_classid);
9514                 off += offsetof(struct sk_buff, cb);
9515                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9516                 *target_size = 2;
9517                 if (type == BPF_WRITE)
9518                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9519                 else
9520                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9521                                               si->src_reg, off);
9522                 break;
9523
9524         case offsetof(struct __sk_buff, data):
9525                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9526                                       si->dst_reg, si->src_reg,
9527                                       offsetof(struct sk_buff, data));
9528                 break;
9529
9530         case offsetof(struct __sk_buff, data_meta):
9531                 off  = si->off;
9532                 off -= offsetof(struct __sk_buff, data_meta);
9533                 off += offsetof(struct sk_buff, cb);
9534                 off += offsetof(struct bpf_skb_data_end, data_meta);
9535                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9536                                       si->src_reg, off);
9537                 break;
9538
9539         case offsetof(struct __sk_buff, data_end):
9540                 off  = si->off;
9541                 off -= offsetof(struct __sk_buff, data_end);
9542                 off += offsetof(struct sk_buff, cb);
9543                 off += offsetof(struct bpf_skb_data_end, data_end);
9544                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9545                                       si->src_reg, off);
9546                 break;
9547
9548         case offsetof(struct __sk_buff, tc_index):
9549 #ifdef CONFIG_NET_SCHED
9550                 if (type == BPF_WRITE)
9551                         *insn++ = BPF_EMIT_STORE(BPF_H, si,
9552                                                  bpf_target_off(struct sk_buff, tc_index, 2,
9553                                                                 target_size));
9554                 else
9555                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9556                                               bpf_target_off(struct sk_buff, tc_index, 2,
9557                                                              target_size));
9558 #else
9559                 *target_size = 2;
9560                 if (type == BPF_WRITE)
9561                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9562                 else
9563                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9564 #endif
9565                 break;
9566
9567         case offsetof(struct __sk_buff, napi_id):
9568 #if defined(CONFIG_NET_RX_BUSY_POLL)
9569                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9570                                       bpf_target_off(struct sk_buff, napi_id, 4,
9571                                                      target_size));
9572                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9573                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9574 #else
9575                 *target_size = 4;
9576                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9577 #endif
9578                 break;
9579         case offsetof(struct __sk_buff, family):
9580                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9581
9582                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9583                                       si->dst_reg, si->src_reg,
9584                                       offsetof(struct sk_buff, sk));
9585                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9586                                       bpf_target_off(struct sock_common,
9587                                                      skc_family,
9588                                                      2, target_size));
9589                 break;
9590         case offsetof(struct __sk_buff, remote_ip4):
9591                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9592
9593                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9594                                       si->dst_reg, si->src_reg,
9595                                       offsetof(struct sk_buff, sk));
9596                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9597                                       bpf_target_off(struct sock_common,
9598                                                      skc_daddr,
9599                                                      4, target_size));
9600                 break;
9601         case offsetof(struct __sk_buff, local_ip4):
9602                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9603                                           skc_rcv_saddr) != 4);
9604
9605                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9606                                       si->dst_reg, si->src_reg,
9607                                       offsetof(struct sk_buff, sk));
9608                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9609                                       bpf_target_off(struct sock_common,
9610                                                      skc_rcv_saddr,
9611                                                      4, target_size));
9612                 break;
9613         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9614              offsetof(struct __sk_buff, remote_ip6[3]):
9615 #if IS_ENABLED(CONFIG_IPV6)
9616                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9617                                           skc_v6_daddr.s6_addr32[0]) != 4);
9618
9619                 off = si->off;
9620                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9621
9622                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9623                                       si->dst_reg, si->src_reg,
9624                                       offsetof(struct sk_buff, sk));
9625                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9626                                       offsetof(struct sock_common,
9627                                                skc_v6_daddr.s6_addr32[0]) +
9628                                       off);
9629 #else
9630                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9631 #endif
9632                 break;
9633         case offsetof(struct __sk_buff, local_ip6[0]) ...
9634              offsetof(struct __sk_buff, local_ip6[3]):
9635 #if IS_ENABLED(CONFIG_IPV6)
9636                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9637                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9638
9639                 off = si->off;
9640                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9641
9642                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9643                                       si->dst_reg, si->src_reg,
9644                                       offsetof(struct sk_buff, sk));
9645                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9646                                       offsetof(struct sock_common,
9647                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9648                                       off);
9649 #else
9650                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9651 #endif
9652                 break;
9653
9654         case offsetof(struct __sk_buff, remote_port):
9655                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9656
9657                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9658                                       si->dst_reg, si->src_reg,
9659                                       offsetof(struct sk_buff, sk));
9660                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9661                                       bpf_target_off(struct sock_common,
9662                                                      skc_dport,
9663                                                      2, target_size));
9664 #ifndef __BIG_ENDIAN_BITFIELD
9665                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9666 #endif
9667                 break;
9668
9669         case offsetof(struct __sk_buff, local_port):
9670                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9671
9672                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9673                                       si->dst_reg, si->src_reg,
9674                                       offsetof(struct sk_buff, sk));
9675                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9676                                       bpf_target_off(struct sock_common,
9677                                                      skc_num, 2, target_size));
9678                 break;
9679
9680         case offsetof(struct __sk_buff, tstamp):
9681                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9682
9683                 if (type == BPF_WRITE)
9684                         insn = bpf_convert_tstamp_write(prog, si, insn);
9685                 else
9686                         insn = bpf_convert_tstamp_read(prog, si, insn);
9687                 break;
9688
9689         case offsetof(struct __sk_buff, tstamp_type):
9690                 insn = bpf_convert_tstamp_type_read(si, insn);
9691                 break;
9692
9693         case offsetof(struct __sk_buff, gso_segs):
9694                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9695                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9696                                       si->dst_reg, si->dst_reg,
9697                                       bpf_target_off(struct skb_shared_info,
9698                                                      gso_segs, 2,
9699                                                      target_size));
9700                 break;
9701         case offsetof(struct __sk_buff, gso_size):
9702                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9703                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9704                                       si->dst_reg, si->dst_reg,
9705                                       bpf_target_off(struct skb_shared_info,
9706                                                      gso_size, 2,
9707                                                      target_size));
9708                 break;
9709         case offsetof(struct __sk_buff, wire_len):
9710                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9711
9712                 off = si->off;
9713                 off -= offsetof(struct __sk_buff, wire_len);
9714                 off += offsetof(struct sk_buff, cb);
9715                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9716                 *target_size = 4;
9717                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9718                 break;
9719
9720         case offsetof(struct __sk_buff, sk):
9721                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9722                                       si->dst_reg, si->src_reg,
9723                                       offsetof(struct sk_buff, sk));
9724                 break;
9725         case offsetof(struct __sk_buff, hwtstamp):
9726                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9727                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9728
9729                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9730                 *insn++ = BPF_LDX_MEM(BPF_DW,
9731                                       si->dst_reg, si->dst_reg,
9732                                       bpf_target_off(struct skb_shared_info,
9733                                                      hwtstamps, 8,
9734                                                      target_size));
9735                 break;
9736         }
9737
9738         return insn - insn_buf;
9739 }
9740
9741 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9742                                 const struct bpf_insn *si,
9743                                 struct bpf_insn *insn_buf,
9744                                 struct bpf_prog *prog, u32 *target_size)
9745 {
9746         struct bpf_insn *insn = insn_buf;
9747         int off;
9748
9749         switch (si->off) {
9750         case offsetof(struct bpf_sock, bound_dev_if):
9751                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9752
9753                 if (type == BPF_WRITE)
9754                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9755                                                  offsetof(struct sock, sk_bound_dev_if));
9756                 else
9757                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9758                                       offsetof(struct sock, sk_bound_dev_if));
9759                 break;
9760
9761         case offsetof(struct bpf_sock, mark):
9762                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9763
9764                 if (type == BPF_WRITE)
9765                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9766                                                  offsetof(struct sock, sk_mark));
9767                 else
9768                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9769                                       offsetof(struct sock, sk_mark));
9770                 break;
9771
9772         case offsetof(struct bpf_sock, priority):
9773                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9774
9775                 if (type == BPF_WRITE)
9776                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9777                                                  offsetof(struct sock, sk_priority));
9778                 else
9779                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9780                                       offsetof(struct sock, sk_priority));
9781                 break;
9782
9783         case offsetof(struct bpf_sock, family):
9784                 *insn++ = BPF_LDX_MEM(
9785                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9786                         si->dst_reg, si->src_reg,
9787                         bpf_target_off(struct sock_common,
9788                                        skc_family,
9789                                        sizeof_field(struct sock_common,
9790                                                     skc_family),
9791                                        target_size));
9792                 break;
9793
9794         case offsetof(struct bpf_sock, type):
9795                 *insn++ = BPF_LDX_MEM(
9796                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9797                         si->dst_reg, si->src_reg,
9798                         bpf_target_off(struct sock, sk_type,
9799                                        sizeof_field(struct sock, sk_type),
9800                                        target_size));
9801                 break;
9802
9803         case offsetof(struct bpf_sock, protocol):
9804                 *insn++ = BPF_LDX_MEM(
9805                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9806                         si->dst_reg, si->src_reg,
9807                         bpf_target_off(struct sock, sk_protocol,
9808                                        sizeof_field(struct sock, sk_protocol),
9809                                        target_size));
9810                 break;
9811
9812         case offsetof(struct bpf_sock, src_ip4):
9813                 *insn++ = BPF_LDX_MEM(
9814                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9815                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9816                                        sizeof_field(struct sock_common,
9817                                                     skc_rcv_saddr),
9818                                        target_size));
9819                 break;
9820
9821         case offsetof(struct bpf_sock, dst_ip4):
9822                 *insn++ = BPF_LDX_MEM(
9823                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9824                         bpf_target_off(struct sock_common, skc_daddr,
9825                                        sizeof_field(struct sock_common,
9826                                                     skc_daddr),
9827                                        target_size));
9828                 break;
9829
9830         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9831 #if IS_ENABLED(CONFIG_IPV6)
9832                 off = si->off;
9833                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9834                 *insn++ = BPF_LDX_MEM(
9835                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9836                         bpf_target_off(
9837                                 struct sock_common,
9838                                 skc_v6_rcv_saddr.s6_addr32[0],
9839                                 sizeof_field(struct sock_common,
9840                                              skc_v6_rcv_saddr.s6_addr32[0]),
9841                                 target_size) + off);
9842 #else
9843                 (void)off;
9844                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9845 #endif
9846                 break;
9847
9848         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9849 #if IS_ENABLED(CONFIG_IPV6)
9850                 off = si->off;
9851                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9852                 *insn++ = BPF_LDX_MEM(
9853                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9854                         bpf_target_off(struct sock_common,
9855                                        skc_v6_daddr.s6_addr32[0],
9856                                        sizeof_field(struct sock_common,
9857                                                     skc_v6_daddr.s6_addr32[0]),
9858                                        target_size) + off);
9859 #else
9860                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9861                 *target_size = 4;
9862 #endif
9863                 break;
9864
9865         case offsetof(struct bpf_sock, src_port):
9866                 *insn++ = BPF_LDX_MEM(
9867                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9868                         si->dst_reg, si->src_reg,
9869                         bpf_target_off(struct sock_common, skc_num,
9870                                        sizeof_field(struct sock_common,
9871                                                     skc_num),
9872                                        target_size));
9873                 break;
9874
9875         case offsetof(struct bpf_sock, dst_port):
9876                 *insn++ = BPF_LDX_MEM(
9877                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9878                         si->dst_reg, si->src_reg,
9879                         bpf_target_off(struct sock_common, skc_dport,
9880                                        sizeof_field(struct sock_common,
9881                                                     skc_dport),
9882                                        target_size));
9883                 break;
9884
9885         case offsetof(struct bpf_sock, state):
9886                 *insn++ = BPF_LDX_MEM(
9887                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9888                         si->dst_reg, si->src_reg,
9889                         bpf_target_off(struct sock_common, skc_state,
9890                                        sizeof_field(struct sock_common,
9891                                                     skc_state),
9892                                        target_size));
9893                 break;
9894         case offsetof(struct bpf_sock, rx_queue_mapping):
9895 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9896                 *insn++ = BPF_LDX_MEM(
9897                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9898                         si->dst_reg, si->src_reg,
9899                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9900                                        sizeof_field(struct sock,
9901                                                     sk_rx_queue_mapping),
9902                                        target_size));
9903                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9904                                       1);
9905                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9906 #else
9907                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9908                 *target_size = 2;
9909 #endif
9910                 break;
9911         }
9912
9913         return insn - insn_buf;
9914 }
9915
9916 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9917                                          const struct bpf_insn *si,
9918                                          struct bpf_insn *insn_buf,
9919                                          struct bpf_prog *prog, u32 *target_size)
9920 {
9921         struct bpf_insn *insn = insn_buf;
9922
9923         switch (si->off) {
9924         case offsetof(struct __sk_buff, ifindex):
9925                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9926                                       si->dst_reg, si->src_reg,
9927                                       offsetof(struct sk_buff, dev));
9928                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9929                                       bpf_target_off(struct net_device, ifindex, 4,
9930                                                      target_size));
9931                 break;
9932         default:
9933                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9934                                               target_size);
9935         }
9936
9937         return insn - insn_buf;
9938 }
9939
9940 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9941                                   const struct bpf_insn *si,
9942                                   struct bpf_insn *insn_buf,
9943                                   struct bpf_prog *prog, u32 *target_size)
9944 {
9945         struct bpf_insn *insn = insn_buf;
9946
9947         switch (si->off) {
9948         case offsetof(struct xdp_md, data):
9949                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9950                                       si->dst_reg, si->src_reg,
9951                                       offsetof(struct xdp_buff, data));
9952                 break;
9953         case offsetof(struct xdp_md, data_meta):
9954                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9955                                       si->dst_reg, si->src_reg,
9956                                       offsetof(struct xdp_buff, data_meta));
9957                 break;
9958         case offsetof(struct xdp_md, data_end):
9959                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9960                                       si->dst_reg, si->src_reg,
9961                                       offsetof(struct xdp_buff, data_end));
9962                 break;
9963         case offsetof(struct xdp_md, ingress_ifindex):
9964                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9965                                       si->dst_reg, si->src_reg,
9966                                       offsetof(struct xdp_buff, rxq));
9967                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9968                                       si->dst_reg, si->dst_reg,
9969                                       offsetof(struct xdp_rxq_info, dev));
9970                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9971                                       offsetof(struct net_device, ifindex));
9972                 break;
9973         case offsetof(struct xdp_md, rx_queue_index):
9974                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9975                                       si->dst_reg, si->src_reg,
9976                                       offsetof(struct xdp_buff, rxq));
9977                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9978                                       offsetof(struct xdp_rxq_info,
9979                                                queue_index));
9980                 break;
9981         case offsetof(struct xdp_md, egress_ifindex):
9982                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9983                                       si->dst_reg, si->src_reg,
9984                                       offsetof(struct xdp_buff, txq));
9985                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9986                                       si->dst_reg, si->dst_reg,
9987                                       offsetof(struct xdp_txq_info, dev));
9988                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9989                                       offsetof(struct net_device, ifindex));
9990                 break;
9991         }
9992
9993         return insn - insn_buf;
9994 }
9995
9996 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9997  * context Structure, F is Field in context structure that contains a pointer
9998  * to Nested Structure of type NS that has the field NF.
9999  *
10000  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10001  * sure that SIZE is not greater than actual size of S.F.NF.
10002  *
10003  * If offset OFF is provided, the load happens from that offset relative to
10004  * offset of NF.
10005  */
10006 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
10007         do {                                                                   \
10008                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
10009                                       si->src_reg, offsetof(S, F));            \
10010                 *insn++ = BPF_LDX_MEM(                                         \
10011                         SIZE, si->dst_reg, si->dst_reg,                        \
10012                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10013                                        target_size)                            \
10014                                 + OFF);                                        \
10015         } while (0)
10016
10017 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
10018         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
10019                                              BPF_FIELD_SIZEOF(NS, NF), 0)
10020
10021 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10022  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10023  *
10024  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10025  * "register" since two registers available in convert_ctx_access are not
10026  * enough: we can't override neither SRC, since it contains value to store, nor
10027  * DST since it contains pointer to context that may be used by later
10028  * instructions. But we need a temporary place to save pointer to nested
10029  * structure whose field we want to store to.
10030  */
10031 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
10032         do {                                                                   \
10033                 int tmp_reg = BPF_REG_9;                                       \
10034                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10035                         --tmp_reg;                                             \
10036                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10037                         --tmp_reg;                                             \
10038                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
10039                                       offsetof(S, TF));                        \
10040                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
10041                                       si->dst_reg, offsetof(S, F));            \
10042                 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code),   \
10043                                        tmp_reg, si->src_reg,                   \
10044                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10045                                        target_size)                            \
10046                                        + OFF,                                  \
10047                                        si->imm);                               \
10048                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
10049                                       offsetof(S, TF));                        \
10050         } while (0)
10051
10052 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10053                                                       TF)                      \
10054         do {                                                                   \
10055                 if (type == BPF_WRITE) {                                       \
10056                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
10057                                                          OFF, TF);             \
10058                 } else {                                                       \
10059                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
10060                                 S, NS, F, NF, SIZE, OFF);  \
10061                 }                                                              \
10062         } while (0)
10063
10064 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
10065         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
10066                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10067
10068 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10069                                         const struct bpf_insn *si,
10070                                         struct bpf_insn *insn_buf,
10071                                         struct bpf_prog *prog, u32 *target_size)
10072 {
10073         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10074         struct bpf_insn *insn = insn_buf;
10075
10076         switch (si->off) {
10077         case offsetof(struct bpf_sock_addr, user_family):
10078                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10079                                             struct sockaddr, uaddr, sa_family);
10080                 break;
10081
10082         case offsetof(struct bpf_sock_addr, user_ip4):
10083                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10084                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10085                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10086                 break;
10087
10088         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10089                 off = si->off;
10090                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10091                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10092                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10093                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10094                         tmp_reg);
10095                 break;
10096
10097         case offsetof(struct bpf_sock_addr, user_port):
10098                 /* To get port we need to know sa_family first and then treat
10099                  * sockaddr as either sockaddr_in or sockaddr_in6.
10100                  * Though we can simplify since port field has same offset and
10101                  * size in both structures.
10102                  * Here we check this invariant and use just one of the
10103                  * structures if it's true.
10104                  */
10105                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10106                              offsetof(struct sockaddr_in6, sin6_port));
10107                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10108                              sizeof_field(struct sockaddr_in6, sin6_port));
10109                 /* Account for sin6_port being smaller than user_port. */
10110                 port_size = min(port_size, BPF_LDST_BYTES(si));
10111                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10112                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10113                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10114                 break;
10115
10116         case offsetof(struct bpf_sock_addr, family):
10117                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10118                                             struct sock, sk, sk_family);
10119                 break;
10120
10121         case offsetof(struct bpf_sock_addr, type):
10122                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10123                                             struct sock, sk, sk_type);
10124                 break;
10125
10126         case offsetof(struct bpf_sock_addr, protocol):
10127                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10128                                             struct sock, sk, sk_protocol);
10129                 break;
10130
10131         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10132                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10133                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10134                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10135                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10136                 break;
10137
10138         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10139                                 msg_src_ip6[3]):
10140                 off = si->off;
10141                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10142                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10143                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10144                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10145                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10146                 break;
10147         case offsetof(struct bpf_sock_addr, sk):
10148                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10149                                       si->dst_reg, si->src_reg,
10150                                       offsetof(struct bpf_sock_addr_kern, sk));
10151                 break;
10152         }
10153
10154         return insn - insn_buf;
10155 }
10156
10157 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10158                                        const struct bpf_insn *si,
10159                                        struct bpf_insn *insn_buf,
10160                                        struct bpf_prog *prog,
10161                                        u32 *target_size)
10162 {
10163         struct bpf_insn *insn = insn_buf;
10164         int off;
10165
10166 /* Helper macro for adding read access to tcp_sock or sock fields. */
10167 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10168         do {                                                                  \
10169                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10170                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10171                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10172                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10173                         reg--;                                                \
10174                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10175                         reg--;                                                \
10176                 if (si->dst_reg == si->src_reg) {                             \
10177                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10178                                           offsetof(struct bpf_sock_ops_kern,  \
10179                                           temp));                             \
10180                         fullsock_reg = reg;                                   \
10181                         jmp += 2;                                             \
10182                 }                                                             \
10183                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10184                                                 struct bpf_sock_ops_kern,     \
10185                                                 is_fullsock),                 \
10186                                       fullsock_reg, si->src_reg,              \
10187                                       offsetof(struct bpf_sock_ops_kern,      \
10188                                                is_fullsock));                 \
10189                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10190                 if (si->dst_reg == si->src_reg)                               \
10191                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10192                                       offsetof(struct bpf_sock_ops_kern,      \
10193                                       temp));                                 \
10194                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10195                                                 struct bpf_sock_ops_kern, sk),\
10196                                       si->dst_reg, si->src_reg,               \
10197                                       offsetof(struct bpf_sock_ops_kern, sk));\
10198                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10199                                                        OBJ_FIELD),            \
10200                                       si->dst_reg, si->dst_reg,               \
10201                                       offsetof(OBJ, OBJ_FIELD));              \
10202                 if (si->dst_reg == si->src_reg) {                             \
10203                         *insn++ = BPF_JMP_A(1);                               \
10204                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10205                                       offsetof(struct bpf_sock_ops_kern,      \
10206                                       temp));                                 \
10207                 }                                                             \
10208         } while (0)
10209
10210 #define SOCK_OPS_GET_SK()                                                             \
10211         do {                                                                  \
10212                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10213                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10214                         reg--;                                                \
10215                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10216                         reg--;                                                \
10217                 if (si->dst_reg == si->src_reg) {                             \
10218                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10219                                           offsetof(struct bpf_sock_ops_kern,  \
10220                                           temp));                             \
10221                         fullsock_reg = reg;                                   \
10222                         jmp += 2;                                             \
10223                 }                                                             \
10224                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10225                                                 struct bpf_sock_ops_kern,     \
10226                                                 is_fullsock),                 \
10227                                       fullsock_reg, si->src_reg,              \
10228                                       offsetof(struct bpf_sock_ops_kern,      \
10229                                                is_fullsock));                 \
10230                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10231                 if (si->dst_reg == si->src_reg)                               \
10232                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10233                                       offsetof(struct bpf_sock_ops_kern,      \
10234                                       temp));                                 \
10235                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10236                                                 struct bpf_sock_ops_kern, sk),\
10237                                       si->dst_reg, si->src_reg,               \
10238                                       offsetof(struct bpf_sock_ops_kern, sk));\
10239                 if (si->dst_reg == si->src_reg) {                             \
10240                         *insn++ = BPF_JMP_A(1);                               \
10241                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10242                                       offsetof(struct bpf_sock_ops_kern,      \
10243                                       temp));                                 \
10244                 }                                                             \
10245         } while (0)
10246
10247 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10248                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10249
10250 /* Helper macro for adding write access to tcp_sock or sock fields.
10251  * The macro is called with two registers, dst_reg which contains a pointer
10252  * to ctx (context) and src_reg which contains the value that should be
10253  * stored. However, we need an additional register since we cannot overwrite
10254  * dst_reg because it may be used later in the program.
10255  * Instead we "borrow" one of the other register. We first save its value
10256  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10257  * it at the end of the macro.
10258  */
10259 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10260         do {                                                                  \
10261                 int reg = BPF_REG_9;                                          \
10262                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10263                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10264                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10265                         reg--;                                                \
10266                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10267                         reg--;                                                \
10268                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10269                                       offsetof(struct bpf_sock_ops_kern,      \
10270                                                temp));                        \
10271                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10272                                                 struct bpf_sock_ops_kern,     \
10273                                                 is_fullsock),                 \
10274                                       reg, si->dst_reg,                       \
10275                                       offsetof(struct bpf_sock_ops_kern,      \
10276                                                is_fullsock));                 \
10277                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10278                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10279                                                 struct bpf_sock_ops_kern, sk),\
10280                                       reg, si->dst_reg,                       \
10281                                       offsetof(struct bpf_sock_ops_kern, sk));\
10282                 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) |     \
10283                                        BPF_MEM | BPF_CLASS(si->code),         \
10284                                        reg, si->src_reg,                      \
10285                                        offsetof(OBJ, OBJ_FIELD),              \
10286                                        si->imm);                              \
10287                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10288                                       offsetof(struct bpf_sock_ops_kern,      \
10289                                                temp));                        \
10290         } while (0)
10291
10292 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10293         do {                                                                  \
10294                 if (TYPE == BPF_WRITE)                                        \
10295                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10296                 else                                                          \
10297                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10298         } while (0)
10299
10300         switch (si->off) {
10301         case offsetof(struct bpf_sock_ops, op):
10302                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10303                                                        op),
10304                                       si->dst_reg, si->src_reg,
10305                                       offsetof(struct bpf_sock_ops_kern, op));
10306                 break;
10307
10308         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10309              offsetof(struct bpf_sock_ops, replylong[3]):
10310                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10311                              sizeof_field(struct bpf_sock_ops_kern, reply));
10312                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10313                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10314                 off = si->off;
10315                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10316                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10317                 if (type == BPF_WRITE)
10318                         *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10319                 else
10320                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10321                                               off);
10322                 break;
10323
10324         case offsetof(struct bpf_sock_ops, family):
10325                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10326
10327                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10328                                               struct bpf_sock_ops_kern, sk),
10329                                       si->dst_reg, si->src_reg,
10330                                       offsetof(struct bpf_sock_ops_kern, sk));
10331                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10332                                       offsetof(struct sock_common, skc_family));
10333                 break;
10334
10335         case offsetof(struct bpf_sock_ops, remote_ip4):
10336                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10337
10338                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10339                                                 struct bpf_sock_ops_kern, sk),
10340                                       si->dst_reg, si->src_reg,
10341                                       offsetof(struct bpf_sock_ops_kern, sk));
10342                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10343                                       offsetof(struct sock_common, skc_daddr));
10344                 break;
10345
10346         case offsetof(struct bpf_sock_ops, local_ip4):
10347                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10348                                           skc_rcv_saddr) != 4);
10349
10350                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10351                                               struct bpf_sock_ops_kern, sk),
10352                                       si->dst_reg, si->src_reg,
10353                                       offsetof(struct bpf_sock_ops_kern, sk));
10354                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10355                                       offsetof(struct sock_common,
10356                                                skc_rcv_saddr));
10357                 break;
10358
10359         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10360              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10361 #if IS_ENABLED(CONFIG_IPV6)
10362                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10363                                           skc_v6_daddr.s6_addr32[0]) != 4);
10364
10365                 off = si->off;
10366                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10367                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10368                                                 struct bpf_sock_ops_kern, sk),
10369                                       si->dst_reg, si->src_reg,
10370                                       offsetof(struct bpf_sock_ops_kern, sk));
10371                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10372                                       offsetof(struct sock_common,
10373                                                skc_v6_daddr.s6_addr32[0]) +
10374                                       off);
10375 #else
10376                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10377 #endif
10378                 break;
10379
10380         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10381              offsetof(struct bpf_sock_ops, local_ip6[3]):
10382 #if IS_ENABLED(CONFIG_IPV6)
10383                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10384                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10385
10386                 off = si->off;
10387                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10388                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10389                                                 struct bpf_sock_ops_kern, sk),
10390                                       si->dst_reg, si->src_reg,
10391                                       offsetof(struct bpf_sock_ops_kern, sk));
10392                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10393                                       offsetof(struct sock_common,
10394                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10395                                       off);
10396 #else
10397                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10398 #endif
10399                 break;
10400
10401         case offsetof(struct bpf_sock_ops, remote_port):
10402                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10403
10404                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10405                                                 struct bpf_sock_ops_kern, sk),
10406                                       si->dst_reg, si->src_reg,
10407                                       offsetof(struct bpf_sock_ops_kern, sk));
10408                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10409                                       offsetof(struct sock_common, skc_dport));
10410 #ifndef __BIG_ENDIAN_BITFIELD
10411                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10412 #endif
10413                 break;
10414
10415         case offsetof(struct bpf_sock_ops, local_port):
10416                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 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_num));
10424                 break;
10425
10426         case offsetof(struct bpf_sock_ops, is_fullsock):
10427                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10428                                                 struct bpf_sock_ops_kern,
10429                                                 is_fullsock),
10430                                       si->dst_reg, si->src_reg,
10431                                       offsetof(struct bpf_sock_ops_kern,
10432                                                is_fullsock));
10433                 break;
10434
10435         case offsetof(struct bpf_sock_ops, state):
10436                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10437
10438                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10439                                                 struct bpf_sock_ops_kern, sk),
10440                                       si->dst_reg, si->src_reg,
10441                                       offsetof(struct bpf_sock_ops_kern, sk));
10442                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10443                                       offsetof(struct sock_common, skc_state));
10444                 break;
10445
10446         case offsetof(struct bpf_sock_ops, rtt_min):
10447                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10448                              sizeof(struct minmax));
10449                 BUILD_BUG_ON(sizeof(struct minmax) <
10450                              sizeof(struct minmax_sample));
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_W, si->dst_reg, si->dst_reg,
10457                                       offsetof(struct tcp_sock, rtt_min) +
10458                                       sizeof_field(struct minmax_sample, t));
10459                 break;
10460
10461         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10462                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10463                                    struct tcp_sock);
10464                 break;
10465
10466         case offsetof(struct bpf_sock_ops, sk_txhash):
10467                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10468                                           struct sock, type);
10469                 break;
10470         case offsetof(struct bpf_sock_ops, snd_cwnd):
10471                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10472                 break;
10473         case offsetof(struct bpf_sock_ops, srtt_us):
10474                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10475                 break;
10476         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10477                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10478                 break;
10479         case offsetof(struct bpf_sock_ops, rcv_nxt):
10480                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10481                 break;
10482         case offsetof(struct bpf_sock_ops, snd_nxt):
10483                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10484                 break;
10485         case offsetof(struct bpf_sock_ops, snd_una):
10486                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10487                 break;
10488         case offsetof(struct bpf_sock_ops, mss_cache):
10489                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10490                 break;
10491         case offsetof(struct bpf_sock_ops, ecn_flags):
10492                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10493                 break;
10494         case offsetof(struct bpf_sock_ops, rate_delivered):
10495                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10496                 break;
10497         case offsetof(struct bpf_sock_ops, rate_interval_us):
10498                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10499                 break;
10500         case offsetof(struct bpf_sock_ops, packets_out):
10501                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10502                 break;
10503         case offsetof(struct bpf_sock_ops, retrans_out):
10504                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10505                 break;
10506         case offsetof(struct bpf_sock_ops, total_retrans):
10507                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10508                 break;
10509         case offsetof(struct bpf_sock_ops, segs_in):
10510                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10511                 break;
10512         case offsetof(struct bpf_sock_ops, data_segs_in):
10513                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10514                 break;
10515         case offsetof(struct bpf_sock_ops, segs_out):
10516                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10517                 break;
10518         case offsetof(struct bpf_sock_ops, data_segs_out):
10519                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10520                 break;
10521         case offsetof(struct bpf_sock_ops, lost_out):
10522                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10523                 break;
10524         case offsetof(struct bpf_sock_ops, sacked_out):
10525                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10526                 break;
10527         case offsetof(struct bpf_sock_ops, bytes_received):
10528                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10529                 break;
10530         case offsetof(struct bpf_sock_ops, bytes_acked):
10531                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10532                 break;
10533         case offsetof(struct bpf_sock_ops, sk):
10534                 SOCK_OPS_GET_SK();
10535                 break;
10536         case offsetof(struct bpf_sock_ops, skb_data_end):
10537                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10538                                                        skb_data_end),
10539                                       si->dst_reg, si->src_reg,
10540                                       offsetof(struct bpf_sock_ops_kern,
10541                                                skb_data_end));
10542                 break;
10543         case offsetof(struct bpf_sock_ops, skb_data):
10544                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10545                                                        skb),
10546                                       si->dst_reg, si->src_reg,
10547                                       offsetof(struct bpf_sock_ops_kern,
10548                                                skb));
10549                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10550                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10551                                       si->dst_reg, si->dst_reg,
10552                                       offsetof(struct sk_buff, data));
10553                 break;
10554         case offsetof(struct bpf_sock_ops, skb_len):
10555                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10556                                                        skb),
10557                                       si->dst_reg, si->src_reg,
10558                                       offsetof(struct bpf_sock_ops_kern,
10559                                                skb));
10560                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10561                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10562                                       si->dst_reg, si->dst_reg,
10563                                       offsetof(struct sk_buff, len));
10564                 break;
10565         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10566                 off = offsetof(struct sk_buff, cb);
10567                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10568                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
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 tcp_skb_cb,
10576                                                        tcp_flags),
10577                                       si->dst_reg, si->dst_reg, off);
10578                 break;
10579         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10580                 struct bpf_insn *jmp_on_null_skb;
10581
10582                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10583                                                        skb),
10584                                       si->dst_reg, si->src_reg,
10585                                       offsetof(struct bpf_sock_ops_kern,
10586                                                skb));
10587                 /* Reserve one insn to test skb == NULL */
10588                 jmp_on_null_skb = insn++;
10589                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10590                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10591                                       bpf_target_off(struct skb_shared_info,
10592                                                      hwtstamps, 8,
10593                                                      target_size));
10594                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10595                                                insn - jmp_on_null_skb - 1);
10596                 break;
10597         }
10598         }
10599         return insn - insn_buf;
10600 }
10601
10602 /* data_end = skb->data + skb_headlen() */
10603 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10604                                                     struct bpf_insn *insn)
10605 {
10606         int reg;
10607         int temp_reg_off = offsetof(struct sk_buff, cb) +
10608                            offsetof(struct sk_skb_cb, temp_reg);
10609
10610         if (si->src_reg == si->dst_reg) {
10611                 /* We need an extra register, choose and save a register. */
10612                 reg = BPF_REG_9;
10613                 if (si->src_reg == reg || si->dst_reg == reg)
10614                         reg--;
10615                 if (si->src_reg == reg || si->dst_reg == reg)
10616                         reg--;
10617                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10618         } else {
10619                 reg = si->dst_reg;
10620         }
10621
10622         /* reg = skb->data */
10623         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10624                               reg, si->src_reg,
10625                               offsetof(struct sk_buff, data));
10626         /* AX = skb->len */
10627         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10628                               BPF_REG_AX, si->src_reg,
10629                               offsetof(struct sk_buff, len));
10630         /* reg = skb->data + skb->len */
10631         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10632         /* AX = skb->data_len */
10633         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10634                               BPF_REG_AX, si->src_reg,
10635                               offsetof(struct sk_buff, data_len));
10636
10637         /* reg = skb->data + skb->len - skb->data_len */
10638         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10639
10640         if (si->src_reg == si->dst_reg) {
10641                 /* Restore the saved register */
10642                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10643                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10644                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10645         }
10646
10647         return insn;
10648 }
10649
10650 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10651                                      const struct bpf_insn *si,
10652                                      struct bpf_insn *insn_buf,
10653                                      struct bpf_prog *prog, u32 *target_size)
10654 {
10655         struct bpf_insn *insn = insn_buf;
10656         int off;
10657
10658         switch (si->off) {
10659         case offsetof(struct __sk_buff, data_end):
10660                 insn = bpf_convert_data_end_access(si, insn);
10661                 break;
10662         case offsetof(struct __sk_buff, cb[0]) ...
10663              offsetofend(struct __sk_buff, cb[4]) - 1:
10664                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10665                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10666                               offsetof(struct sk_skb_cb, data)) %
10667                              sizeof(__u64));
10668
10669                 prog->cb_access = 1;
10670                 off  = si->off;
10671                 off -= offsetof(struct __sk_buff, cb[0]);
10672                 off += offsetof(struct sk_buff, cb);
10673                 off += offsetof(struct sk_skb_cb, data);
10674                 if (type == BPF_WRITE)
10675                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10676                 else
10677                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10678                                               si->src_reg, off);
10679                 break;
10680
10681
10682         default:
10683                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10684                                               target_size);
10685         }
10686
10687         return insn - insn_buf;
10688 }
10689
10690 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10691                                      const struct bpf_insn *si,
10692                                      struct bpf_insn *insn_buf,
10693                                      struct bpf_prog *prog, u32 *target_size)
10694 {
10695         struct bpf_insn *insn = insn_buf;
10696 #if IS_ENABLED(CONFIG_IPV6)
10697         int off;
10698 #endif
10699
10700         /* convert ctx uses the fact sg element is first in struct */
10701         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10702
10703         switch (si->off) {
10704         case offsetof(struct sk_msg_md, data):
10705                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10706                                       si->dst_reg, si->src_reg,
10707                                       offsetof(struct sk_msg, data));
10708                 break;
10709         case offsetof(struct sk_msg_md, data_end):
10710                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10711                                       si->dst_reg, si->src_reg,
10712                                       offsetof(struct sk_msg, data_end));
10713                 break;
10714         case offsetof(struct sk_msg_md, family):
10715                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10716
10717                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10718                                               struct sk_msg, sk),
10719                                       si->dst_reg, si->src_reg,
10720                                       offsetof(struct sk_msg, sk));
10721                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10722                                       offsetof(struct sock_common, skc_family));
10723                 break;
10724
10725         case offsetof(struct sk_msg_md, remote_ip4):
10726                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10727
10728                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10729                                                 struct sk_msg, sk),
10730                                       si->dst_reg, si->src_reg,
10731                                       offsetof(struct sk_msg, sk));
10732                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10733                                       offsetof(struct sock_common, skc_daddr));
10734                 break;
10735
10736         case offsetof(struct sk_msg_md, local_ip4):
10737                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10738                                           skc_rcv_saddr) != 4);
10739
10740                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10741                                               struct sk_msg, sk),
10742                                       si->dst_reg, si->src_reg,
10743                                       offsetof(struct sk_msg, sk));
10744                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10745                                       offsetof(struct sock_common,
10746                                                skc_rcv_saddr));
10747                 break;
10748
10749         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10750              offsetof(struct sk_msg_md, remote_ip6[3]):
10751 #if IS_ENABLED(CONFIG_IPV6)
10752                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10753                                           skc_v6_daddr.s6_addr32[0]) != 4);
10754
10755                 off = si->off;
10756                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10757                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10758                                                 struct sk_msg, sk),
10759                                       si->dst_reg, si->src_reg,
10760                                       offsetof(struct sk_msg, sk));
10761                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10762                                       offsetof(struct sock_common,
10763                                                skc_v6_daddr.s6_addr32[0]) +
10764                                       off);
10765 #else
10766                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10767 #endif
10768                 break;
10769
10770         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10771              offsetof(struct sk_msg_md, local_ip6[3]):
10772 #if IS_ENABLED(CONFIG_IPV6)
10773                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10774                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10775
10776                 off = si->off;
10777                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10778                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10779                                                 struct sk_msg, sk),
10780                                       si->dst_reg, si->src_reg,
10781                                       offsetof(struct sk_msg, sk));
10782                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10783                                       offsetof(struct sock_common,
10784                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10785                                       off);
10786 #else
10787                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10788 #endif
10789                 break;
10790
10791         case offsetof(struct sk_msg_md, remote_port):
10792                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10793
10794                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10795                                                 struct sk_msg, sk),
10796                                       si->dst_reg, si->src_reg,
10797                                       offsetof(struct sk_msg, sk));
10798                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10799                                       offsetof(struct sock_common, skc_dport));
10800 #ifndef __BIG_ENDIAN_BITFIELD
10801                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10802 #endif
10803                 break;
10804
10805         case offsetof(struct sk_msg_md, local_port):
10806                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 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_num));
10814                 break;
10815
10816         case offsetof(struct sk_msg_md, size):
10817                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10818                                       si->dst_reg, si->src_reg,
10819                                       offsetof(struct sk_msg_sg, size));
10820                 break;
10821
10822         case offsetof(struct sk_msg_md, sk):
10823                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10824                                       si->dst_reg, si->src_reg,
10825                                       offsetof(struct sk_msg, sk));
10826                 break;
10827         }
10828
10829         return insn - insn_buf;
10830 }
10831
10832 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10833         .get_func_proto         = sk_filter_func_proto,
10834         .is_valid_access        = sk_filter_is_valid_access,
10835         .convert_ctx_access     = bpf_convert_ctx_access,
10836         .gen_ld_abs             = bpf_gen_ld_abs,
10837 };
10838
10839 const struct bpf_prog_ops sk_filter_prog_ops = {
10840         .test_run               = bpf_prog_test_run_skb,
10841 };
10842
10843 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10844         .get_func_proto         = tc_cls_act_func_proto,
10845         .is_valid_access        = tc_cls_act_is_valid_access,
10846         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10847         .gen_prologue           = tc_cls_act_prologue,
10848         .gen_ld_abs             = bpf_gen_ld_abs,
10849         .btf_struct_access      = tc_cls_act_btf_struct_access,
10850 };
10851
10852 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10853         .test_run               = bpf_prog_test_run_skb,
10854 };
10855
10856 const struct bpf_verifier_ops xdp_verifier_ops = {
10857         .get_func_proto         = xdp_func_proto,
10858         .is_valid_access        = xdp_is_valid_access,
10859         .convert_ctx_access     = xdp_convert_ctx_access,
10860         .gen_prologue           = bpf_noop_prologue,
10861         .btf_struct_access      = xdp_btf_struct_access,
10862 };
10863
10864 const struct bpf_prog_ops xdp_prog_ops = {
10865         .test_run               = bpf_prog_test_run_xdp,
10866 };
10867
10868 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10869         .get_func_proto         = cg_skb_func_proto,
10870         .is_valid_access        = cg_skb_is_valid_access,
10871         .convert_ctx_access     = bpf_convert_ctx_access,
10872 };
10873
10874 const struct bpf_prog_ops cg_skb_prog_ops = {
10875         .test_run               = bpf_prog_test_run_skb,
10876 };
10877
10878 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10879         .get_func_proto         = lwt_in_func_proto,
10880         .is_valid_access        = lwt_is_valid_access,
10881         .convert_ctx_access     = bpf_convert_ctx_access,
10882 };
10883
10884 const struct bpf_prog_ops lwt_in_prog_ops = {
10885         .test_run               = bpf_prog_test_run_skb,
10886 };
10887
10888 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10889         .get_func_proto         = lwt_out_func_proto,
10890         .is_valid_access        = lwt_is_valid_access,
10891         .convert_ctx_access     = bpf_convert_ctx_access,
10892 };
10893
10894 const struct bpf_prog_ops lwt_out_prog_ops = {
10895         .test_run               = bpf_prog_test_run_skb,
10896 };
10897
10898 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10899         .get_func_proto         = lwt_xmit_func_proto,
10900         .is_valid_access        = lwt_is_valid_access,
10901         .convert_ctx_access     = bpf_convert_ctx_access,
10902         .gen_prologue           = tc_cls_act_prologue,
10903 };
10904
10905 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10906         .test_run               = bpf_prog_test_run_skb,
10907 };
10908
10909 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10910         .get_func_proto         = lwt_seg6local_func_proto,
10911         .is_valid_access        = lwt_is_valid_access,
10912         .convert_ctx_access     = bpf_convert_ctx_access,
10913 };
10914
10915 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10916         .test_run               = bpf_prog_test_run_skb,
10917 };
10918
10919 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10920         .get_func_proto         = sock_filter_func_proto,
10921         .is_valid_access        = sock_filter_is_valid_access,
10922         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10923 };
10924
10925 const struct bpf_prog_ops cg_sock_prog_ops = {
10926 };
10927
10928 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10929         .get_func_proto         = sock_addr_func_proto,
10930         .is_valid_access        = sock_addr_is_valid_access,
10931         .convert_ctx_access     = sock_addr_convert_ctx_access,
10932 };
10933
10934 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10935 };
10936
10937 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10938         .get_func_proto         = sock_ops_func_proto,
10939         .is_valid_access        = sock_ops_is_valid_access,
10940         .convert_ctx_access     = sock_ops_convert_ctx_access,
10941 };
10942
10943 const struct bpf_prog_ops sock_ops_prog_ops = {
10944 };
10945
10946 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10947         .get_func_proto         = sk_skb_func_proto,
10948         .is_valid_access        = sk_skb_is_valid_access,
10949         .convert_ctx_access     = sk_skb_convert_ctx_access,
10950         .gen_prologue           = sk_skb_prologue,
10951 };
10952
10953 const struct bpf_prog_ops sk_skb_prog_ops = {
10954 };
10955
10956 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10957         .get_func_proto         = sk_msg_func_proto,
10958         .is_valid_access        = sk_msg_is_valid_access,
10959         .convert_ctx_access     = sk_msg_convert_ctx_access,
10960         .gen_prologue           = bpf_noop_prologue,
10961 };
10962
10963 const struct bpf_prog_ops sk_msg_prog_ops = {
10964 };
10965
10966 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10967         .get_func_proto         = flow_dissector_func_proto,
10968         .is_valid_access        = flow_dissector_is_valid_access,
10969         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10970 };
10971
10972 const struct bpf_prog_ops flow_dissector_prog_ops = {
10973         .test_run               = bpf_prog_test_run_flow_dissector,
10974 };
10975
10976 int sk_detach_filter(struct sock *sk)
10977 {
10978         int ret = -ENOENT;
10979         struct sk_filter *filter;
10980
10981         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10982                 return -EPERM;
10983
10984         filter = rcu_dereference_protected(sk->sk_filter,
10985                                            lockdep_sock_is_held(sk));
10986         if (filter) {
10987                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10988                 sk_filter_uncharge(sk, filter);
10989                 ret = 0;
10990         }
10991
10992         return ret;
10993 }
10994 EXPORT_SYMBOL_GPL(sk_detach_filter);
10995
10996 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10997 {
10998         struct sock_fprog_kern *fprog;
10999         struct sk_filter *filter;
11000         int ret = 0;
11001
11002         sockopt_lock_sock(sk);
11003         filter = rcu_dereference_protected(sk->sk_filter,
11004                                            lockdep_sock_is_held(sk));
11005         if (!filter)
11006                 goto out;
11007
11008         /* We're copying the filter that has been originally attached,
11009          * so no conversion/decode needed anymore. eBPF programs that
11010          * have no original program cannot be dumped through this.
11011          */
11012         ret = -EACCES;
11013         fprog = filter->prog->orig_prog;
11014         if (!fprog)
11015                 goto out;
11016
11017         ret = fprog->len;
11018         if (!len)
11019                 /* User space only enquires number of filter blocks. */
11020                 goto out;
11021
11022         ret = -EINVAL;
11023         if (len < fprog->len)
11024                 goto out;
11025
11026         ret = -EFAULT;
11027         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11028                 goto out;
11029
11030         /* Instead of bytes, the API requests to return the number
11031          * of filter blocks.
11032          */
11033         ret = fprog->len;
11034 out:
11035         sockopt_release_sock(sk);
11036         return ret;
11037 }
11038
11039 #ifdef CONFIG_INET
11040 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11041                                     struct sock_reuseport *reuse,
11042                                     struct sock *sk, struct sk_buff *skb,
11043                                     struct sock *migrating_sk,
11044                                     u32 hash)
11045 {
11046         reuse_kern->skb = skb;
11047         reuse_kern->sk = sk;
11048         reuse_kern->selected_sk = NULL;
11049         reuse_kern->migrating_sk = migrating_sk;
11050         reuse_kern->data_end = skb->data + skb_headlen(skb);
11051         reuse_kern->hash = hash;
11052         reuse_kern->reuseport_id = reuse->reuseport_id;
11053         reuse_kern->bind_inany = reuse->bind_inany;
11054 }
11055
11056 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11057                                   struct bpf_prog *prog, struct sk_buff *skb,
11058                                   struct sock *migrating_sk,
11059                                   u32 hash)
11060 {
11061         struct sk_reuseport_kern reuse_kern;
11062         enum sk_action action;
11063
11064         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11065         action = bpf_prog_run(prog, &reuse_kern);
11066
11067         if (action == SK_PASS)
11068                 return reuse_kern.selected_sk;
11069         else
11070                 return ERR_PTR(-ECONNREFUSED);
11071 }
11072
11073 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11074            struct bpf_map *, map, void *, key, u32, flags)
11075 {
11076         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11077         struct sock_reuseport *reuse;
11078         struct sock *selected_sk;
11079
11080         selected_sk = map->ops->map_lookup_elem(map, key);
11081         if (!selected_sk)
11082                 return -ENOENT;
11083
11084         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11085         if (!reuse) {
11086                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11087                 if (sk_is_refcounted(selected_sk))
11088                         sock_put(selected_sk);
11089
11090                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11091                  * The only (!reuse) case here is - the sk has already been
11092                  * unhashed (e.g. by close()), so treat it as -ENOENT.
11093                  *
11094                  * Other maps (e.g. sock_map) do not provide this guarantee and
11095                  * the sk may never be in the reuseport group to begin with.
11096                  */
11097                 return is_sockarray ? -ENOENT : -EINVAL;
11098         }
11099
11100         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11101                 struct sock *sk = reuse_kern->sk;
11102
11103                 if (sk->sk_protocol != selected_sk->sk_protocol)
11104                         return -EPROTOTYPE;
11105                 else if (sk->sk_family != selected_sk->sk_family)
11106                         return -EAFNOSUPPORT;
11107
11108                 /* Catch all. Likely bound to a different sockaddr. */
11109                 return -EBADFD;
11110         }
11111
11112         reuse_kern->selected_sk = selected_sk;
11113
11114         return 0;
11115 }
11116
11117 static const struct bpf_func_proto sk_select_reuseport_proto = {
11118         .func           = sk_select_reuseport,
11119         .gpl_only       = false,
11120         .ret_type       = RET_INTEGER,
11121         .arg1_type      = ARG_PTR_TO_CTX,
11122         .arg2_type      = ARG_CONST_MAP_PTR,
11123         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11124         .arg4_type      = ARG_ANYTHING,
11125 };
11126
11127 BPF_CALL_4(sk_reuseport_load_bytes,
11128            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11129            void *, to, u32, len)
11130 {
11131         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11132 }
11133
11134 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11135         .func           = sk_reuseport_load_bytes,
11136         .gpl_only       = false,
11137         .ret_type       = RET_INTEGER,
11138         .arg1_type      = ARG_PTR_TO_CTX,
11139         .arg2_type      = ARG_ANYTHING,
11140         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11141         .arg4_type      = ARG_CONST_SIZE,
11142 };
11143
11144 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11145            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11146            void *, to, u32, len, u32, start_header)
11147 {
11148         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11149                                                len, start_header);
11150 }
11151
11152 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11153         .func           = sk_reuseport_load_bytes_relative,
11154         .gpl_only       = false,
11155         .ret_type       = RET_INTEGER,
11156         .arg1_type      = ARG_PTR_TO_CTX,
11157         .arg2_type      = ARG_ANYTHING,
11158         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11159         .arg4_type      = ARG_CONST_SIZE,
11160         .arg5_type      = ARG_ANYTHING,
11161 };
11162
11163 static const struct bpf_func_proto *
11164 sk_reuseport_func_proto(enum bpf_func_id func_id,
11165                         const struct bpf_prog *prog)
11166 {
11167         switch (func_id) {
11168         case BPF_FUNC_sk_select_reuseport:
11169                 return &sk_select_reuseport_proto;
11170         case BPF_FUNC_skb_load_bytes:
11171                 return &sk_reuseport_load_bytes_proto;
11172         case BPF_FUNC_skb_load_bytes_relative:
11173                 return &sk_reuseport_load_bytes_relative_proto;
11174         case BPF_FUNC_get_socket_cookie:
11175                 return &bpf_get_socket_ptr_cookie_proto;
11176         case BPF_FUNC_ktime_get_coarse_ns:
11177                 return &bpf_ktime_get_coarse_ns_proto;
11178         default:
11179                 return bpf_base_func_proto(func_id);
11180         }
11181 }
11182
11183 static bool
11184 sk_reuseport_is_valid_access(int off, int size,
11185                              enum bpf_access_type type,
11186                              const struct bpf_prog *prog,
11187                              struct bpf_insn_access_aux *info)
11188 {
11189         const u32 size_default = sizeof(__u32);
11190
11191         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11192             off % size || type != BPF_READ)
11193                 return false;
11194
11195         switch (off) {
11196         case offsetof(struct sk_reuseport_md, data):
11197                 info->reg_type = PTR_TO_PACKET;
11198                 return size == sizeof(__u64);
11199
11200         case offsetof(struct sk_reuseport_md, data_end):
11201                 info->reg_type = PTR_TO_PACKET_END;
11202                 return size == sizeof(__u64);
11203
11204         case offsetof(struct sk_reuseport_md, hash):
11205                 return size == size_default;
11206
11207         case offsetof(struct sk_reuseport_md, sk):
11208                 info->reg_type = PTR_TO_SOCKET;
11209                 return size == sizeof(__u64);
11210
11211         case offsetof(struct sk_reuseport_md, migrating_sk):
11212                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11213                 return size == sizeof(__u64);
11214
11215         /* Fields that allow narrowing */
11216         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11217                 if (size < sizeof_field(struct sk_buff, protocol))
11218                         return false;
11219                 fallthrough;
11220         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11221         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11222         case bpf_ctx_range(struct sk_reuseport_md, len):
11223                 bpf_ctx_record_field_size(info, size_default);
11224                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11225
11226         default:
11227                 return false;
11228         }
11229 }
11230
11231 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11232         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11233                               si->dst_reg, si->src_reg,                 \
11234                               bpf_target_off(struct sk_reuseport_kern, F, \
11235                                              sizeof_field(struct sk_reuseport_kern, F), \
11236                                              target_size));             \
11237         })
11238
11239 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11240         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11241                                     struct sk_buff,                     \
11242                                     skb,                                \
11243                                     SKB_FIELD)
11244
11245 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11246         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11247                                     struct sock,                        \
11248                                     sk,                                 \
11249                                     SK_FIELD)
11250
11251 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11252                                            const struct bpf_insn *si,
11253                                            struct bpf_insn *insn_buf,
11254                                            struct bpf_prog *prog,
11255                                            u32 *target_size)
11256 {
11257         struct bpf_insn *insn = insn_buf;
11258
11259         switch (si->off) {
11260         case offsetof(struct sk_reuseport_md, data):
11261                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11262                 break;
11263
11264         case offsetof(struct sk_reuseport_md, len):
11265                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11266                 break;
11267
11268         case offsetof(struct sk_reuseport_md, eth_protocol):
11269                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11270                 break;
11271
11272         case offsetof(struct sk_reuseport_md, ip_protocol):
11273                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11274                 break;
11275
11276         case offsetof(struct sk_reuseport_md, data_end):
11277                 SK_REUSEPORT_LOAD_FIELD(data_end);
11278                 break;
11279
11280         case offsetof(struct sk_reuseport_md, hash):
11281                 SK_REUSEPORT_LOAD_FIELD(hash);
11282                 break;
11283
11284         case offsetof(struct sk_reuseport_md, bind_inany):
11285                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11286                 break;
11287
11288         case offsetof(struct sk_reuseport_md, sk):
11289                 SK_REUSEPORT_LOAD_FIELD(sk);
11290                 break;
11291
11292         case offsetof(struct sk_reuseport_md, migrating_sk):
11293                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11294                 break;
11295         }
11296
11297         return insn - insn_buf;
11298 }
11299
11300 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11301         .get_func_proto         = sk_reuseport_func_proto,
11302         .is_valid_access        = sk_reuseport_is_valid_access,
11303         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11304 };
11305
11306 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11307 };
11308
11309 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11310 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11311
11312 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11313            struct sock *, sk, u64, flags)
11314 {
11315         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11316                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11317                 return -EINVAL;
11318         if (unlikely(sk && sk_is_refcounted(sk)))
11319                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11320         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11321                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11322         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11323                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11324
11325         /* Check if socket is suitable for packet L3/L4 protocol */
11326         if (sk && sk->sk_protocol != ctx->protocol)
11327                 return -EPROTOTYPE;
11328         if (sk && sk->sk_family != ctx->family &&
11329             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11330                 return -EAFNOSUPPORT;
11331
11332         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11333                 return -EEXIST;
11334
11335         /* Select socket as lookup result */
11336         ctx->selected_sk = sk;
11337         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11338         return 0;
11339 }
11340
11341 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11342         .func           = bpf_sk_lookup_assign,
11343         .gpl_only       = false,
11344         .ret_type       = RET_INTEGER,
11345         .arg1_type      = ARG_PTR_TO_CTX,
11346         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11347         .arg3_type      = ARG_ANYTHING,
11348 };
11349
11350 static const struct bpf_func_proto *
11351 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11352 {
11353         switch (func_id) {
11354         case BPF_FUNC_perf_event_output:
11355                 return &bpf_event_output_data_proto;
11356         case BPF_FUNC_sk_assign:
11357                 return &bpf_sk_lookup_assign_proto;
11358         case BPF_FUNC_sk_release:
11359                 return &bpf_sk_release_proto;
11360         default:
11361                 return bpf_sk_base_func_proto(func_id);
11362         }
11363 }
11364
11365 static bool sk_lookup_is_valid_access(int off, int size,
11366                                       enum bpf_access_type type,
11367                                       const struct bpf_prog *prog,
11368                                       struct bpf_insn_access_aux *info)
11369 {
11370         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11371                 return false;
11372         if (off % size != 0)
11373                 return false;
11374         if (type != BPF_READ)
11375                 return false;
11376
11377         switch (off) {
11378         case offsetof(struct bpf_sk_lookup, sk):
11379                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11380                 return size == sizeof(__u64);
11381
11382         case bpf_ctx_range(struct bpf_sk_lookup, family):
11383         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11384         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11385         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11386         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11387         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11388         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11389         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11390                 bpf_ctx_record_field_size(info, sizeof(__u32));
11391                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11392
11393         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11394                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11395                 if (size == sizeof(__u32))
11396                         return true;
11397                 bpf_ctx_record_field_size(info, sizeof(__be16));
11398                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11399
11400         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11401              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11402                 /* Allow access to zero padding for backward compatibility */
11403                 bpf_ctx_record_field_size(info, sizeof(__u16));
11404                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11405
11406         default:
11407                 return false;
11408         }
11409 }
11410
11411 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11412                                         const struct bpf_insn *si,
11413                                         struct bpf_insn *insn_buf,
11414                                         struct bpf_prog *prog,
11415                                         u32 *target_size)
11416 {
11417         struct bpf_insn *insn = insn_buf;
11418
11419         switch (si->off) {
11420         case offsetof(struct bpf_sk_lookup, sk):
11421                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11422                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11423                 break;
11424
11425         case offsetof(struct bpf_sk_lookup, family):
11426                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11427                                       bpf_target_off(struct bpf_sk_lookup_kern,
11428                                                      family, 2, target_size));
11429                 break;
11430
11431         case offsetof(struct bpf_sk_lookup, protocol):
11432                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11433                                       bpf_target_off(struct bpf_sk_lookup_kern,
11434                                                      protocol, 2, target_size));
11435                 break;
11436
11437         case offsetof(struct bpf_sk_lookup, remote_ip4):
11438                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11439                                       bpf_target_off(struct bpf_sk_lookup_kern,
11440                                                      v4.saddr, 4, target_size));
11441                 break;
11442
11443         case offsetof(struct bpf_sk_lookup, local_ip4):
11444                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11445                                       bpf_target_off(struct bpf_sk_lookup_kern,
11446                                                      v4.daddr, 4, target_size));
11447                 break;
11448
11449         case bpf_ctx_range_till(struct bpf_sk_lookup,
11450                                 remote_ip6[0], remote_ip6[3]): {
11451 #if IS_ENABLED(CONFIG_IPV6)
11452                 int off = si->off;
11453
11454                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11455                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11456                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11457                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11458                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11459                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11460 #else
11461                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11462 #endif
11463                 break;
11464         }
11465         case bpf_ctx_range_till(struct bpf_sk_lookup,
11466                                 local_ip6[0], local_ip6[3]): {
11467 #if IS_ENABLED(CONFIG_IPV6)
11468                 int off = si->off;
11469
11470                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11471                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11472                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11473                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11474                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11475                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11476 #else
11477                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11478 #endif
11479                 break;
11480         }
11481         case offsetof(struct bpf_sk_lookup, remote_port):
11482                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11483                                       bpf_target_off(struct bpf_sk_lookup_kern,
11484                                                      sport, 2, target_size));
11485                 break;
11486
11487         case offsetofend(struct bpf_sk_lookup, remote_port):
11488                 *target_size = 2;
11489                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11490                 break;
11491
11492         case offsetof(struct bpf_sk_lookup, local_port):
11493                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11494                                       bpf_target_off(struct bpf_sk_lookup_kern,
11495                                                      dport, 2, target_size));
11496                 break;
11497
11498         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11499                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11500                                       bpf_target_off(struct bpf_sk_lookup_kern,
11501                                                      ingress_ifindex, 4, target_size));
11502                 break;
11503         }
11504
11505         return insn - insn_buf;
11506 }
11507
11508 const struct bpf_prog_ops sk_lookup_prog_ops = {
11509         .test_run = bpf_prog_test_run_sk_lookup,
11510 };
11511
11512 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11513         .get_func_proto         = sk_lookup_func_proto,
11514         .is_valid_access        = sk_lookup_is_valid_access,
11515         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11516 };
11517
11518 #endif /* CONFIG_INET */
11519
11520 DEFINE_BPF_DISPATCHER(xdp)
11521
11522 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11523 {
11524         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11525 }
11526
11527 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11528 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11529 BTF_SOCK_TYPE_xxx
11530 #undef BTF_SOCK_TYPE
11531
11532 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11533 {
11534         /* tcp6_sock type is not generated in dwarf and hence btf,
11535          * trigger an explicit type generation here.
11536          */
11537         BTF_TYPE_EMIT(struct tcp6_sock);
11538         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11539             sk->sk_family == AF_INET6)
11540                 return (unsigned long)sk;
11541
11542         return (unsigned long)NULL;
11543 }
11544
11545 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11546         .func                   = bpf_skc_to_tcp6_sock,
11547         .gpl_only               = false,
11548         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11549         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11550         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11551 };
11552
11553 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11554 {
11555         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11556                 return (unsigned long)sk;
11557
11558         return (unsigned long)NULL;
11559 }
11560
11561 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11562         .func                   = bpf_skc_to_tcp_sock,
11563         .gpl_only               = false,
11564         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11565         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11566         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11567 };
11568
11569 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11570 {
11571         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11572          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11573          */
11574         BTF_TYPE_EMIT(struct inet_timewait_sock);
11575         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11576
11577 #ifdef CONFIG_INET
11578         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11579                 return (unsigned long)sk;
11580 #endif
11581
11582 #if IS_BUILTIN(CONFIG_IPV6)
11583         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11584                 return (unsigned long)sk;
11585 #endif
11586
11587         return (unsigned long)NULL;
11588 }
11589
11590 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11591         .func                   = bpf_skc_to_tcp_timewait_sock,
11592         .gpl_only               = false,
11593         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11594         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11595         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11596 };
11597
11598 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11599 {
11600 #ifdef CONFIG_INET
11601         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11602                 return (unsigned long)sk;
11603 #endif
11604
11605 #if IS_BUILTIN(CONFIG_IPV6)
11606         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11607                 return (unsigned long)sk;
11608 #endif
11609
11610         return (unsigned long)NULL;
11611 }
11612
11613 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11614         .func                   = bpf_skc_to_tcp_request_sock,
11615         .gpl_only               = false,
11616         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11617         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11618         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11619 };
11620
11621 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11622 {
11623         /* udp6_sock type is not generated in dwarf and hence btf,
11624          * trigger an explicit type generation here.
11625          */
11626         BTF_TYPE_EMIT(struct udp6_sock);
11627         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11628             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11629                 return (unsigned long)sk;
11630
11631         return (unsigned long)NULL;
11632 }
11633
11634 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11635         .func                   = bpf_skc_to_udp6_sock,
11636         .gpl_only               = false,
11637         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11638         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11639         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11640 };
11641
11642 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11643 {
11644         /* unix_sock type is not generated in dwarf and hence btf,
11645          * trigger an explicit type generation here.
11646          */
11647         BTF_TYPE_EMIT(struct unix_sock);
11648         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11649                 return (unsigned long)sk;
11650
11651         return (unsigned long)NULL;
11652 }
11653
11654 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11655         .func                   = bpf_skc_to_unix_sock,
11656         .gpl_only               = false,
11657         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11658         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11659         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11660 };
11661
11662 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11663 {
11664         BTF_TYPE_EMIT(struct mptcp_sock);
11665         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11666 }
11667
11668 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11669         .func           = bpf_skc_to_mptcp_sock,
11670         .gpl_only       = false,
11671         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11672         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11673         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11674 };
11675
11676 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11677 {
11678         return (unsigned long)sock_from_file(file);
11679 }
11680
11681 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11682 BTF_ID(struct, socket)
11683 BTF_ID(struct, file)
11684
11685 const struct bpf_func_proto bpf_sock_from_file_proto = {
11686         .func           = bpf_sock_from_file,
11687         .gpl_only       = false,
11688         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11689         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11690         .arg1_type      = ARG_PTR_TO_BTF_ID,
11691         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11692 };
11693
11694 static const struct bpf_func_proto *
11695 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11696 {
11697         const struct bpf_func_proto *func;
11698
11699         switch (func_id) {
11700         case BPF_FUNC_skc_to_tcp6_sock:
11701                 func = &bpf_skc_to_tcp6_sock_proto;
11702                 break;
11703         case BPF_FUNC_skc_to_tcp_sock:
11704                 func = &bpf_skc_to_tcp_sock_proto;
11705                 break;
11706         case BPF_FUNC_skc_to_tcp_timewait_sock:
11707                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11708                 break;
11709         case BPF_FUNC_skc_to_tcp_request_sock:
11710                 func = &bpf_skc_to_tcp_request_sock_proto;
11711                 break;
11712         case BPF_FUNC_skc_to_udp6_sock:
11713                 func = &bpf_skc_to_udp6_sock_proto;
11714                 break;
11715         case BPF_FUNC_skc_to_unix_sock:
11716                 func = &bpf_skc_to_unix_sock_proto;
11717                 break;
11718         case BPF_FUNC_skc_to_mptcp_sock:
11719                 func = &bpf_skc_to_mptcp_sock_proto;
11720                 break;
11721         case BPF_FUNC_ktime_get_coarse_ns:
11722                 return &bpf_ktime_get_coarse_ns_proto;
11723         default:
11724                 return bpf_base_func_proto(func_id);
11725         }
11726
11727         if (!perfmon_capable())
11728                 return NULL;
11729
11730         return func;
11731 }
11732
11733 __diag_push();
11734 __diag_ignore_all("-Wmissing-prototypes",
11735                   "Global functions as their definitions will be in vmlinux BTF");
11736 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11737                                     struct bpf_dynptr_kern *ptr__uninit)
11738 {
11739         if (flags) {
11740                 bpf_dynptr_set_null(ptr__uninit);
11741                 return -EINVAL;
11742         }
11743
11744         bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11745
11746         return 0;
11747 }
11748
11749 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11750                                     struct bpf_dynptr_kern *ptr__uninit)
11751 {
11752         if (flags) {
11753                 bpf_dynptr_set_null(ptr__uninit);
11754                 return -EINVAL;
11755         }
11756
11757         bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11758
11759         return 0;
11760 }
11761 __diag_pop();
11762
11763 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11764                                struct bpf_dynptr_kern *ptr__uninit)
11765 {
11766         int err;
11767
11768         err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11769         if (err)
11770                 return err;
11771
11772         bpf_dynptr_set_rdonly(ptr__uninit);
11773
11774         return 0;
11775 }
11776
11777 BTF_SET8_START(bpf_kfunc_check_set_skb)
11778 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11779 BTF_SET8_END(bpf_kfunc_check_set_skb)
11780
11781 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11782 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11783 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11784
11785 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11786         .owner = THIS_MODULE,
11787         .set = &bpf_kfunc_check_set_skb,
11788 };
11789
11790 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11791         .owner = THIS_MODULE,
11792         .set = &bpf_kfunc_check_set_xdp,
11793 };
11794
11795 static int __init bpf_kfunc_init(void)
11796 {
11797         int ret;
11798
11799         ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11800         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11801         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11802         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11803         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11804         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11805         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11806         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11807         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11808         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11809         return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11810 }
11811 late_initcall(bpf_kfunc_init);
11812
11813 /* Disables missing prototype warnings */
11814 __diag_push();
11815 __diag_ignore_all("-Wmissing-prototypes",
11816                   "Global functions as their definitions will be in vmlinux BTF");
11817
11818 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11819  *
11820  * The function expects a non-NULL pointer to a socket, and invokes the
11821  * protocol specific socket destroy handlers.
11822  *
11823  * The helper can only be called from BPF contexts that have acquired the socket
11824  * locks.
11825  *
11826  * Parameters:
11827  * @sock: Pointer to socket to be destroyed
11828  *
11829  * Return:
11830  * On error, may return EPROTONOSUPPORT, EINVAL.
11831  * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11832  * 0 otherwise
11833  */
11834 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11835 {
11836         struct sock *sk = (struct sock *)sock;
11837
11838         /* The locking semantics that allow for synchronous execution of the
11839          * destroy handlers are only supported for TCP and UDP.
11840          * Supporting protocols will need to acquire sock lock in the BPF context
11841          * prior to invoking this kfunc.
11842          */
11843         if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11844                                            sk->sk_protocol != IPPROTO_UDP))
11845                 return -EOPNOTSUPP;
11846
11847         return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11848 }
11849
11850 __diag_pop()
11851
11852 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11853 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11854 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11855
11856 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11857 {
11858         if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11859             prog->expected_attach_type != BPF_TRACE_ITER)
11860                 return -EACCES;
11861         return 0;
11862 }
11863
11864 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11865         .owner = THIS_MODULE,
11866         .set   = &bpf_sk_iter_kfunc_ids,
11867         .filter = tracing_iter_filter,
11868 };
11869
11870 static int init_subsystem(void)
11871 {
11872         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11873 }
11874 late_initcall(init_subsystem);