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                 /* XDP_REDIRECT is not supported AF_XDP yet. */
4350                 if (unlikely(xdp_buff_has_frags(xdp)))
4351                         return -EOPNOTSUPP;
4352
4353                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4354         }
4355
4356         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4357                                        xdp_prog);
4358 }
4359 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4360
4361 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4362                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4363 {
4364         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4365         enum bpf_map_type map_type = ri->map_type;
4366
4367         if (map_type == BPF_MAP_TYPE_XSKMAP)
4368                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4369
4370         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4371 }
4372 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4373
4374 static int xdp_do_generic_redirect_map(struct net_device *dev,
4375                                        struct sk_buff *skb,
4376                                        struct xdp_buff *xdp,
4377                                        struct bpf_prog *xdp_prog,
4378                                        void *fwd,
4379                                        enum bpf_map_type map_type, u32 map_id)
4380 {
4381         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4382         struct bpf_map *map;
4383         int err;
4384
4385         switch (map_type) {
4386         case BPF_MAP_TYPE_DEVMAP:
4387                 fallthrough;
4388         case BPF_MAP_TYPE_DEVMAP_HASH:
4389                 map = READ_ONCE(ri->map);
4390                 if (unlikely(map)) {
4391                         WRITE_ONCE(ri->map, NULL);
4392                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4393                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4394                 } else {
4395                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4396                 }
4397                 if (unlikely(err))
4398                         goto err;
4399                 break;
4400         case BPF_MAP_TYPE_XSKMAP:
4401                 err = xsk_generic_rcv(fwd, xdp);
4402                 if (err)
4403                         goto err;
4404                 consume_skb(skb);
4405                 break;
4406         case BPF_MAP_TYPE_CPUMAP:
4407                 err = cpu_map_generic_redirect(fwd, skb);
4408                 if (unlikely(err))
4409                         goto err;
4410                 break;
4411         default:
4412                 err = -EBADRQC;
4413                 goto err;
4414         }
4415
4416         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4417         return 0;
4418 err:
4419         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4420         return err;
4421 }
4422
4423 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4424                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4425 {
4426         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4427         enum bpf_map_type map_type = ri->map_type;
4428         void *fwd = ri->tgt_value;
4429         u32 map_id = ri->map_id;
4430         int err;
4431
4432         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4433         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4434
4435         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4436                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4437                 if (unlikely(!fwd)) {
4438                         err = -EINVAL;
4439                         goto err;
4440                 }
4441
4442                 err = xdp_ok_fwd_dev(fwd, skb->len);
4443                 if (unlikely(err))
4444                         goto err;
4445
4446                 skb->dev = fwd;
4447                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4448                 generic_xdp_tx(skb, xdp_prog);
4449                 return 0;
4450         }
4451
4452         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4453 err:
4454         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4455         return err;
4456 }
4457
4458 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4459 {
4460         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4461
4462         if (unlikely(flags))
4463                 return XDP_ABORTED;
4464
4465         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4466          * by map_idr) is used for ifindex based XDP redirect.
4467          */
4468         ri->tgt_index = ifindex;
4469         ri->map_id = INT_MAX;
4470         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4471
4472         return XDP_REDIRECT;
4473 }
4474
4475 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4476         .func           = bpf_xdp_redirect,
4477         .gpl_only       = false,
4478         .ret_type       = RET_INTEGER,
4479         .arg1_type      = ARG_ANYTHING,
4480         .arg2_type      = ARG_ANYTHING,
4481 };
4482
4483 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4484            u64, flags)
4485 {
4486         return map->ops->map_redirect(map, key, flags);
4487 }
4488
4489 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4490         .func           = bpf_xdp_redirect_map,
4491         .gpl_only       = false,
4492         .ret_type       = RET_INTEGER,
4493         .arg1_type      = ARG_CONST_MAP_PTR,
4494         .arg2_type      = ARG_ANYTHING,
4495         .arg3_type      = ARG_ANYTHING,
4496 };
4497
4498 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4499                                   unsigned long off, unsigned long len)
4500 {
4501         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4502
4503         if (unlikely(!ptr))
4504                 return len;
4505         if (ptr != dst_buff)
4506                 memcpy(dst_buff, ptr, len);
4507
4508         return 0;
4509 }
4510
4511 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4512            u64, flags, void *, meta, u64, meta_size)
4513 {
4514         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4515
4516         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4517                 return -EINVAL;
4518         if (unlikely(!skb || skb_size > skb->len))
4519                 return -EFAULT;
4520
4521         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4522                                 bpf_skb_copy);
4523 }
4524
4525 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4526         .func           = bpf_skb_event_output,
4527         .gpl_only       = true,
4528         .ret_type       = RET_INTEGER,
4529         .arg1_type      = ARG_PTR_TO_CTX,
4530         .arg2_type      = ARG_CONST_MAP_PTR,
4531         .arg3_type      = ARG_ANYTHING,
4532         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4533         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4534 };
4535
4536 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4537
4538 const struct bpf_func_proto bpf_skb_output_proto = {
4539         .func           = bpf_skb_event_output,
4540         .gpl_only       = true,
4541         .ret_type       = RET_INTEGER,
4542         .arg1_type      = ARG_PTR_TO_BTF_ID,
4543         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4544         .arg2_type      = ARG_CONST_MAP_PTR,
4545         .arg3_type      = ARG_ANYTHING,
4546         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4547         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4548 };
4549
4550 static unsigned short bpf_tunnel_key_af(u64 flags)
4551 {
4552         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4553 }
4554
4555 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4556            u32, size, u64, flags)
4557 {
4558         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4559         u8 compat[sizeof(struct bpf_tunnel_key)];
4560         void *to_orig = to;
4561         int err;
4562
4563         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4564                                          BPF_F_TUNINFO_FLAGS)))) {
4565                 err = -EINVAL;
4566                 goto err_clear;
4567         }
4568         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4569                 err = -EPROTO;
4570                 goto err_clear;
4571         }
4572         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4573                 err = -EINVAL;
4574                 switch (size) {
4575                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4576                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4577                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4578                         goto set_compat;
4579                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4580                         /* Fixup deprecated structure layouts here, so we have
4581                          * a common path later on.
4582                          */
4583                         if (ip_tunnel_info_af(info) != AF_INET)
4584                                 goto err_clear;
4585 set_compat:
4586                         to = (struct bpf_tunnel_key *)compat;
4587                         break;
4588                 default:
4589                         goto err_clear;
4590                 }
4591         }
4592
4593         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4594         to->tunnel_tos = info->key.tos;
4595         to->tunnel_ttl = info->key.ttl;
4596         if (flags & BPF_F_TUNINFO_FLAGS)
4597                 to->tunnel_flags = info->key.tun_flags;
4598         else
4599                 to->tunnel_ext = 0;
4600
4601         if (flags & BPF_F_TUNINFO_IPV6) {
4602                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4603                        sizeof(to->remote_ipv6));
4604                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4605                        sizeof(to->local_ipv6));
4606                 to->tunnel_label = be32_to_cpu(info->key.label);
4607         } else {
4608                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4609                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4610                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4611                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4612                 to->tunnel_label = 0;
4613         }
4614
4615         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4616                 memcpy(to_orig, to, size);
4617
4618         return 0;
4619 err_clear:
4620         memset(to_orig, 0, size);
4621         return err;
4622 }
4623
4624 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4625         .func           = bpf_skb_get_tunnel_key,
4626         .gpl_only       = false,
4627         .ret_type       = RET_INTEGER,
4628         .arg1_type      = ARG_PTR_TO_CTX,
4629         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4630         .arg3_type      = ARG_CONST_SIZE,
4631         .arg4_type      = ARG_ANYTHING,
4632 };
4633
4634 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4635 {
4636         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4637         int err;
4638
4639         if (unlikely(!info ||
4640                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4641                 err = -ENOENT;
4642                 goto err_clear;
4643         }
4644         if (unlikely(size < info->options_len)) {
4645                 err = -ENOMEM;
4646                 goto err_clear;
4647         }
4648
4649         ip_tunnel_info_opts_get(to, info);
4650         if (size > info->options_len)
4651                 memset(to + info->options_len, 0, size - info->options_len);
4652
4653         return info->options_len;
4654 err_clear:
4655         memset(to, 0, size);
4656         return err;
4657 }
4658
4659 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4660         .func           = bpf_skb_get_tunnel_opt,
4661         .gpl_only       = false,
4662         .ret_type       = RET_INTEGER,
4663         .arg1_type      = ARG_PTR_TO_CTX,
4664         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4665         .arg3_type      = ARG_CONST_SIZE,
4666 };
4667
4668 static struct metadata_dst __percpu *md_dst;
4669
4670 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4671            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4672 {
4673         struct metadata_dst *md = this_cpu_ptr(md_dst);
4674         u8 compat[sizeof(struct bpf_tunnel_key)];
4675         struct ip_tunnel_info *info;
4676
4677         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4678                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4679                                BPF_F_NO_TUNNEL_KEY)))
4680                 return -EINVAL;
4681         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4682                 switch (size) {
4683                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4684                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4685                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4686                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4687                         /* Fixup deprecated structure layouts here, so we have
4688                          * a common path later on.
4689                          */
4690                         memcpy(compat, from, size);
4691                         memset(compat + size, 0, sizeof(compat) - size);
4692                         from = (const struct bpf_tunnel_key *) compat;
4693                         break;
4694                 default:
4695                         return -EINVAL;
4696                 }
4697         }
4698         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4699                      from->tunnel_ext))
4700                 return -EINVAL;
4701
4702         skb_dst_drop(skb);
4703         dst_hold((struct dst_entry *) md);
4704         skb_dst_set(skb, (struct dst_entry *) md);
4705
4706         info = &md->u.tun_info;
4707         memset(info, 0, sizeof(*info));
4708         info->mode = IP_TUNNEL_INFO_TX;
4709
4710         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4711         if (flags & BPF_F_DONT_FRAGMENT)
4712                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4713         if (flags & BPF_F_ZERO_CSUM_TX)
4714                 info->key.tun_flags &= ~TUNNEL_CSUM;
4715         if (flags & BPF_F_SEQ_NUMBER)
4716                 info->key.tun_flags |= TUNNEL_SEQ;
4717         if (flags & BPF_F_NO_TUNNEL_KEY)
4718                 info->key.tun_flags &= ~TUNNEL_KEY;
4719
4720         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4721         info->key.tos = from->tunnel_tos;
4722         info->key.ttl = from->tunnel_ttl;
4723
4724         if (flags & BPF_F_TUNINFO_IPV6) {
4725                 info->mode |= IP_TUNNEL_INFO_IPV6;
4726                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4727                        sizeof(from->remote_ipv6));
4728                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4729                        sizeof(from->local_ipv6));
4730                 info->key.label = cpu_to_be32(from->tunnel_label) &
4731                                   IPV6_FLOWLABEL_MASK;
4732         } else {
4733                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4734                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4735                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4736         }
4737
4738         return 0;
4739 }
4740
4741 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4742         .func           = bpf_skb_set_tunnel_key,
4743         .gpl_only       = false,
4744         .ret_type       = RET_INTEGER,
4745         .arg1_type      = ARG_PTR_TO_CTX,
4746         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4747         .arg3_type      = ARG_CONST_SIZE,
4748         .arg4_type      = ARG_ANYTHING,
4749 };
4750
4751 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4752            const u8 *, from, u32, size)
4753 {
4754         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4755         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4756
4757         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4758                 return -EINVAL;
4759         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4760                 return -ENOMEM;
4761
4762         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4763
4764         return 0;
4765 }
4766
4767 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4768         .func           = bpf_skb_set_tunnel_opt,
4769         .gpl_only       = false,
4770         .ret_type       = RET_INTEGER,
4771         .arg1_type      = ARG_PTR_TO_CTX,
4772         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4773         .arg3_type      = ARG_CONST_SIZE,
4774 };
4775
4776 static const struct bpf_func_proto *
4777 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4778 {
4779         if (!md_dst) {
4780                 struct metadata_dst __percpu *tmp;
4781
4782                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4783                                                 METADATA_IP_TUNNEL,
4784                                                 GFP_KERNEL);
4785                 if (!tmp)
4786                         return NULL;
4787                 if (cmpxchg(&md_dst, NULL, tmp))
4788                         metadata_dst_free_percpu(tmp);
4789         }
4790
4791         switch (which) {
4792         case BPF_FUNC_skb_set_tunnel_key:
4793                 return &bpf_skb_set_tunnel_key_proto;
4794         case BPF_FUNC_skb_set_tunnel_opt:
4795                 return &bpf_skb_set_tunnel_opt_proto;
4796         default:
4797                 return NULL;
4798         }
4799 }
4800
4801 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4802            u32, idx)
4803 {
4804         struct bpf_array *array = container_of(map, struct bpf_array, map);
4805         struct cgroup *cgrp;
4806         struct sock *sk;
4807
4808         sk = skb_to_full_sk(skb);
4809         if (!sk || !sk_fullsock(sk))
4810                 return -ENOENT;
4811         if (unlikely(idx >= array->map.max_entries))
4812                 return -E2BIG;
4813
4814         cgrp = READ_ONCE(array->ptrs[idx]);
4815         if (unlikely(!cgrp))
4816                 return -EAGAIN;
4817
4818         return sk_under_cgroup_hierarchy(sk, cgrp);
4819 }
4820
4821 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4822         .func           = bpf_skb_under_cgroup,
4823         .gpl_only       = false,
4824         .ret_type       = RET_INTEGER,
4825         .arg1_type      = ARG_PTR_TO_CTX,
4826         .arg2_type      = ARG_CONST_MAP_PTR,
4827         .arg3_type      = ARG_ANYTHING,
4828 };
4829
4830 #ifdef CONFIG_SOCK_CGROUP_DATA
4831 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4832 {
4833         struct cgroup *cgrp;
4834
4835         sk = sk_to_full_sk(sk);
4836         if (!sk || !sk_fullsock(sk))
4837                 return 0;
4838
4839         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4840         return cgroup_id(cgrp);
4841 }
4842
4843 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4844 {
4845         return __bpf_sk_cgroup_id(skb->sk);
4846 }
4847
4848 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4849         .func           = bpf_skb_cgroup_id,
4850         .gpl_only       = false,
4851         .ret_type       = RET_INTEGER,
4852         .arg1_type      = ARG_PTR_TO_CTX,
4853 };
4854
4855 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4856                                               int ancestor_level)
4857 {
4858         struct cgroup *ancestor;
4859         struct cgroup *cgrp;
4860
4861         sk = sk_to_full_sk(sk);
4862         if (!sk || !sk_fullsock(sk))
4863                 return 0;
4864
4865         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4866         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4867         if (!ancestor)
4868                 return 0;
4869
4870         return cgroup_id(ancestor);
4871 }
4872
4873 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4874            ancestor_level)
4875 {
4876         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4877 }
4878
4879 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4880         .func           = bpf_skb_ancestor_cgroup_id,
4881         .gpl_only       = false,
4882         .ret_type       = RET_INTEGER,
4883         .arg1_type      = ARG_PTR_TO_CTX,
4884         .arg2_type      = ARG_ANYTHING,
4885 };
4886
4887 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4888 {
4889         return __bpf_sk_cgroup_id(sk);
4890 }
4891
4892 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4893         .func           = bpf_sk_cgroup_id,
4894         .gpl_only       = false,
4895         .ret_type       = RET_INTEGER,
4896         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4897 };
4898
4899 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4900 {
4901         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4902 }
4903
4904 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4905         .func           = bpf_sk_ancestor_cgroup_id,
4906         .gpl_only       = false,
4907         .ret_type       = RET_INTEGER,
4908         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4909         .arg2_type      = ARG_ANYTHING,
4910 };
4911 #endif
4912
4913 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4914                                   unsigned long off, unsigned long len)
4915 {
4916         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4917
4918         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4919         return 0;
4920 }
4921
4922 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4923            u64, flags, void *, meta, u64, meta_size)
4924 {
4925         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4926
4927         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4928                 return -EINVAL;
4929
4930         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4931                 return -EFAULT;
4932
4933         return bpf_event_output(map, flags, meta, meta_size, xdp,
4934                                 xdp_size, bpf_xdp_copy);
4935 }
4936
4937 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4938         .func           = bpf_xdp_event_output,
4939         .gpl_only       = true,
4940         .ret_type       = RET_INTEGER,
4941         .arg1_type      = ARG_PTR_TO_CTX,
4942         .arg2_type      = ARG_CONST_MAP_PTR,
4943         .arg3_type      = ARG_ANYTHING,
4944         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4945         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4946 };
4947
4948 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4949
4950 const struct bpf_func_proto bpf_xdp_output_proto = {
4951         .func           = bpf_xdp_event_output,
4952         .gpl_only       = true,
4953         .ret_type       = RET_INTEGER,
4954         .arg1_type      = ARG_PTR_TO_BTF_ID,
4955         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4956         .arg2_type      = ARG_CONST_MAP_PTR,
4957         .arg3_type      = ARG_ANYTHING,
4958         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4959         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4960 };
4961
4962 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4963 {
4964         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4965 }
4966
4967 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4968         .func           = bpf_get_socket_cookie,
4969         .gpl_only       = false,
4970         .ret_type       = RET_INTEGER,
4971         .arg1_type      = ARG_PTR_TO_CTX,
4972 };
4973
4974 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4975 {
4976         return __sock_gen_cookie(ctx->sk);
4977 }
4978
4979 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4980         .func           = bpf_get_socket_cookie_sock_addr,
4981         .gpl_only       = false,
4982         .ret_type       = RET_INTEGER,
4983         .arg1_type      = ARG_PTR_TO_CTX,
4984 };
4985
4986 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4987 {
4988         return __sock_gen_cookie(ctx);
4989 }
4990
4991 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4992         .func           = bpf_get_socket_cookie_sock,
4993         .gpl_only       = false,
4994         .ret_type       = RET_INTEGER,
4995         .arg1_type      = ARG_PTR_TO_CTX,
4996 };
4997
4998 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4999 {
5000         return sk ? sock_gen_cookie(sk) : 0;
5001 }
5002
5003 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
5004         .func           = bpf_get_socket_ptr_cookie,
5005         .gpl_only       = false,
5006         .ret_type       = RET_INTEGER,
5007         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL,
5008 };
5009
5010 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5011 {
5012         return __sock_gen_cookie(ctx->sk);
5013 }
5014
5015 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
5016         .func           = bpf_get_socket_cookie_sock_ops,
5017         .gpl_only       = false,
5018         .ret_type       = RET_INTEGER,
5019         .arg1_type      = ARG_PTR_TO_CTX,
5020 };
5021
5022 static u64 __bpf_get_netns_cookie(struct sock *sk)
5023 {
5024         const struct net *net = sk ? sock_net(sk) : &init_net;
5025
5026         return net->net_cookie;
5027 }
5028
5029 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5030 {
5031         return __bpf_get_netns_cookie(ctx);
5032 }
5033
5034 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5035         .func           = bpf_get_netns_cookie_sock,
5036         .gpl_only       = false,
5037         .ret_type       = RET_INTEGER,
5038         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5039 };
5040
5041 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5042 {
5043         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5044 }
5045
5046 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5047         .func           = bpf_get_netns_cookie_sock_addr,
5048         .gpl_only       = false,
5049         .ret_type       = RET_INTEGER,
5050         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5051 };
5052
5053 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5054 {
5055         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5056 }
5057
5058 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5059         .func           = bpf_get_netns_cookie_sock_ops,
5060         .gpl_only       = false,
5061         .ret_type       = RET_INTEGER,
5062         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5063 };
5064
5065 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5066 {
5067         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5068 }
5069
5070 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5071         .func           = bpf_get_netns_cookie_sk_msg,
5072         .gpl_only       = false,
5073         .ret_type       = RET_INTEGER,
5074         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5075 };
5076
5077 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5078 {
5079         struct sock *sk = sk_to_full_sk(skb->sk);
5080         kuid_t kuid;
5081
5082         if (!sk || !sk_fullsock(sk))
5083                 return overflowuid;
5084         kuid = sock_net_uid(sock_net(sk), sk);
5085         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5086 }
5087
5088 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5089         .func           = bpf_get_socket_uid,
5090         .gpl_only       = false,
5091         .ret_type       = RET_INTEGER,
5092         .arg1_type      = ARG_PTR_TO_CTX,
5093 };
5094
5095 static int sol_socket_sockopt(struct sock *sk, int optname,
5096                               char *optval, int *optlen,
5097                               bool getopt)
5098 {
5099         switch (optname) {
5100         case SO_REUSEADDR:
5101         case SO_SNDBUF:
5102         case SO_RCVBUF:
5103         case SO_KEEPALIVE:
5104         case SO_PRIORITY:
5105         case SO_REUSEPORT:
5106         case SO_RCVLOWAT:
5107         case SO_MARK:
5108         case SO_MAX_PACING_RATE:
5109         case SO_BINDTOIFINDEX:
5110         case SO_TXREHASH:
5111                 if (*optlen != sizeof(int))
5112                         return -EINVAL;
5113                 break;
5114         case SO_BINDTODEVICE:
5115                 break;
5116         default:
5117                 return -EINVAL;
5118         }
5119
5120         if (getopt) {
5121                 if (optname == SO_BINDTODEVICE)
5122                         return -EINVAL;
5123                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5124                                      KERNEL_SOCKPTR(optval),
5125                                      KERNEL_SOCKPTR(optlen));
5126         }
5127
5128         return sk_setsockopt(sk, SOL_SOCKET, optname,
5129                              KERNEL_SOCKPTR(optval), *optlen);
5130 }
5131
5132 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5133                                   char *optval, int optlen)
5134 {
5135         struct tcp_sock *tp = tcp_sk(sk);
5136         unsigned long timeout;
5137         int val;
5138
5139         if (optlen != sizeof(int))
5140                 return -EINVAL;
5141
5142         val = *(int *)optval;
5143
5144         /* Only some options are supported */
5145         switch (optname) {
5146         case TCP_BPF_IW:
5147                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5148                         return -EINVAL;
5149                 tcp_snd_cwnd_set(tp, val);
5150                 break;
5151         case TCP_BPF_SNDCWND_CLAMP:
5152                 if (val <= 0)
5153                         return -EINVAL;
5154                 tp->snd_cwnd_clamp = val;
5155                 tp->snd_ssthresh = val;
5156                 break;
5157         case TCP_BPF_DELACK_MAX:
5158                 timeout = usecs_to_jiffies(val);
5159                 if (timeout > TCP_DELACK_MAX ||
5160                     timeout < TCP_TIMEOUT_MIN)
5161                         return -EINVAL;
5162                 inet_csk(sk)->icsk_delack_max = timeout;
5163                 break;
5164         case TCP_BPF_RTO_MIN:
5165                 timeout = usecs_to_jiffies(val);
5166                 if (timeout > TCP_RTO_MIN ||
5167                     timeout < TCP_TIMEOUT_MIN)
5168                         return -EINVAL;
5169                 inet_csk(sk)->icsk_rto_min = timeout;
5170                 break;
5171         default:
5172                 return -EINVAL;
5173         }
5174
5175         return 0;
5176 }
5177
5178 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5179                                       int *optlen, bool getopt)
5180 {
5181         struct tcp_sock *tp;
5182         int ret;
5183
5184         if (*optlen < 2)
5185                 return -EINVAL;
5186
5187         if (getopt) {
5188                 if (!inet_csk(sk)->icsk_ca_ops)
5189                         return -EINVAL;
5190                 /* BPF expects NULL-terminated tcp-cc string */
5191                 optval[--(*optlen)] = '\0';
5192                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5193                                          KERNEL_SOCKPTR(optval),
5194                                          KERNEL_SOCKPTR(optlen));
5195         }
5196
5197         /* "cdg" is the only cc that alloc a ptr
5198          * in inet_csk_ca area.  The bpf-tcp-cc may
5199          * overwrite this ptr after switching to cdg.
5200          */
5201         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5202                 return -ENOTSUPP;
5203
5204         /* It stops this looping
5205          *
5206          * .init => bpf_setsockopt(tcp_cc) => .init =>
5207          * bpf_setsockopt(tcp_cc)" => .init => ....
5208          *
5209          * The second bpf_setsockopt(tcp_cc) is not allowed
5210          * in order to break the loop when both .init
5211          * are the same bpf prog.
5212          *
5213          * This applies even the second bpf_setsockopt(tcp_cc)
5214          * does not cause a loop.  This limits only the first
5215          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5216          * pick a fallback cc (eg. peer does not support ECN)
5217          * and the second '.init' cannot fallback to
5218          * another.
5219          */
5220         tp = tcp_sk(sk);
5221         if (tp->bpf_chg_cc_inprogress)
5222                 return -EBUSY;
5223
5224         tp->bpf_chg_cc_inprogress = 1;
5225         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5226                                 KERNEL_SOCKPTR(optval), *optlen);
5227         tp->bpf_chg_cc_inprogress = 0;
5228         return ret;
5229 }
5230
5231 static int sol_tcp_sockopt(struct sock *sk, int optname,
5232                            char *optval, int *optlen,
5233                            bool getopt)
5234 {
5235         if (sk->sk_protocol != IPPROTO_TCP)
5236                 return -EINVAL;
5237
5238         switch (optname) {
5239         case TCP_NODELAY:
5240         case TCP_MAXSEG:
5241         case TCP_KEEPIDLE:
5242         case TCP_KEEPINTVL:
5243         case TCP_KEEPCNT:
5244         case TCP_SYNCNT:
5245         case TCP_WINDOW_CLAMP:
5246         case TCP_THIN_LINEAR_TIMEOUTS:
5247         case TCP_USER_TIMEOUT:
5248         case TCP_NOTSENT_LOWAT:
5249         case TCP_SAVE_SYN:
5250                 if (*optlen != sizeof(int))
5251                         return -EINVAL;
5252                 break;
5253         case TCP_CONGESTION:
5254                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5255         case TCP_SAVED_SYN:
5256                 if (*optlen < 1)
5257                         return -EINVAL;
5258                 break;
5259         default:
5260                 if (getopt)
5261                         return -EINVAL;
5262                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5263         }
5264
5265         if (getopt) {
5266                 if (optname == TCP_SAVED_SYN) {
5267                         struct tcp_sock *tp = tcp_sk(sk);
5268
5269                         if (!tp->saved_syn ||
5270                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5271                                 return -EINVAL;
5272                         memcpy(optval, tp->saved_syn->data, *optlen);
5273                         /* It cannot free tp->saved_syn here because it
5274                          * does not know if the user space still needs it.
5275                          */
5276                         return 0;
5277                 }
5278
5279                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5280                                          KERNEL_SOCKPTR(optval),
5281                                          KERNEL_SOCKPTR(optlen));
5282         }
5283
5284         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5285                                  KERNEL_SOCKPTR(optval), *optlen);
5286 }
5287
5288 static int sol_ip_sockopt(struct sock *sk, int optname,
5289                           char *optval, int *optlen,
5290                           bool getopt)
5291 {
5292         if (sk->sk_family != AF_INET)
5293                 return -EINVAL;
5294
5295         switch (optname) {
5296         case IP_TOS:
5297                 if (*optlen != sizeof(int))
5298                         return -EINVAL;
5299                 break;
5300         default:
5301                 return -EINVAL;
5302         }
5303
5304         if (getopt)
5305                 return do_ip_getsockopt(sk, SOL_IP, optname,
5306                                         KERNEL_SOCKPTR(optval),
5307                                         KERNEL_SOCKPTR(optlen));
5308
5309         return do_ip_setsockopt(sk, SOL_IP, optname,
5310                                 KERNEL_SOCKPTR(optval), *optlen);
5311 }
5312
5313 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5314                             char *optval, int *optlen,
5315                             bool getopt)
5316 {
5317         if (sk->sk_family != AF_INET6)
5318                 return -EINVAL;
5319
5320         switch (optname) {
5321         case IPV6_TCLASS:
5322         case IPV6_AUTOFLOWLABEL:
5323                 if (*optlen != sizeof(int))
5324                         return -EINVAL;
5325                 break;
5326         default:
5327                 return -EINVAL;
5328         }
5329
5330         if (getopt)
5331                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5332                                                       KERNEL_SOCKPTR(optval),
5333                                                       KERNEL_SOCKPTR(optlen));
5334
5335         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5336                                               KERNEL_SOCKPTR(optval), *optlen);
5337 }
5338
5339 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5340                             char *optval, int optlen)
5341 {
5342         if (!sk_fullsock(sk))
5343                 return -EINVAL;
5344
5345         if (level == SOL_SOCKET)
5346                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5347         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5348                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5349         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5350                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5351         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5352                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5353
5354         return -EINVAL;
5355 }
5356
5357 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5358                            char *optval, int optlen)
5359 {
5360         if (sk_fullsock(sk))
5361                 sock_owned_by_me(sk);
5362         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5363 }
5364
5365 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5366                             char *optval, int optlen)
5367 {
5368         int err, saved_optlen = optlen;
5369
5370         if (!sk_fullsock(sk)) {
5371                 err = -EINVAL;
5372                 goto done;
5373         }
5374
5375         if (level == SOL_SOCKET)
5376                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5377         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5378                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5379         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5380                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5381         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5382                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5383         else
5384                 err = -EINVAL;
5385
5386 done:
5387         if (err)
5388                 optlen = 0;
5389         if (optlen < saved_optlen)
5390                 memset(optval + optlen, 0, saved_optlen - optlen);
5391         return err;
5392 }
5393
5394 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5395                            char *optval, int optlen)
5396 {
5397         if (sk_fullsock(sk))
5398                 sock_owned_by_me(sk);
5399         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5400 }
5401
5402 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5403            int, optname, char *, optval, int, optlen)
5404 {
5405         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5406 }
5407
5408 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5409         .func           = bpf_sk_setsockopt,
5410         .gpl_only       = false,
5411         .ret_type       = RET_INTEGER,
5412         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5413         .arg2_type      = ARG_ANYTHING,
5414         .arg3_type      = ARG_ANYTHING,
5415         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5416         .arg5_type      = ARG_CONST_SIZE,
5417 };
5418
5419 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5420            int, optname, char *, optval, int, optlen)
5421 {
5422         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5423 }
5424
5425 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5426         .func           = bpf_sk_getsockopt,
5427         .gpl_only       = false,
5428         .ret_type       = RET_INTEGER,
5429         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5430         .arg2_type      = ARG_ANYTHING,
5431         .arg3_type      = ARG_ANYTHING,
5432         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5433         .arg5_type      = ARG_CONST_SIZE,
5434 };
5435
5436 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5437            int, optname, char *, optval, int, optlen)
5438 {
5439         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5440 }
5441
5442 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5443         .func           = bpf_unlocked_sk_setsockopt,
5444         .gpl_only       = false,
5445         .ret_type       = RET_INTEGER,
5446         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5447         .arg2_type      = ARG_ANYTHING,
5448         .arg3_type      = ARG_ANYTHING,
5449         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5450         .arg5_type      = ARG_CONST_SIZE,
5451 };
5452
5453 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5454            int, optname, char *, optval, int, optlen)
5455 {
5456         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5457 }
5458
5459 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5460         .func           = bpf_unlocked_sk_getsockopt,
5461         .gpl_only       = false,
5462         .ret_type       = RET_INTEGER,
5463         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5464         .arg2_type      = ARG_ANYTHING,
5465         .arg3_type      = ARG_ANYTHING,
5466         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5467         .arg5_type      = ARG_CONST_SIZE,
5468 };
5469
5470 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5471            int, level, int, optname, char *, optval, int, optlen)
5472 {
5473         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5474 }
5475
5476 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5477         .func           = bpf_sock_addr_setsockopt,
5478         .gpl_only       = false,
5479         .ret_type       = RET_INTEGER,
5480         .arg1_type      = ARG_PTR_TO_CTX,
5481         .arg2_type      = ARG_ANYTHING,
5482         .arg3_type      = ARG_ANYTHING,
5483         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5484         .arg5_type      = ARG_CONST_SIZE,
5485 };
5486
5487 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5488            int, level, int, optname, char *, optval, int, optlen)
5489 {
5490         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5491 }
5492
5493 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5494         .func           = bpf_sock_addr_getsockopt,
5495         .gpl_only       = false,
5496         .ret_type       = RET_INTEGER,
5497         .arg1_type      = ARG_PTR_TO_CTX,
5498         .arg2_type      = ARG_ANYTHING,
5499         .arg3_type      = ARG_ANYTHING,
5500         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5501         .arg5_type      = ARG_CONST_SIZE,
5502 };
5503
5504 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5505            int, level, int, optname, char *, optval, int, optlen)
5506 {
5507         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5508 }
5509
5510 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5511         .func           = bpf_sock_ops_setsockopt,
5512         .gpl_only       = false,
5513         .ret_type       = RET_INTEGER,
5514         .arg1_type      = ARG_PTR_TO_CTX,
5515         .arg2_type      = ARG_ANYTHING,
5516         .arg3_type      = ARG_ANYTHING,
5517         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5518         .arg5_type      = ARG_CONST_SIZE,
5519 };
5520
5521 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5522                                 int optname, const u8 **start)
5523 {
5524         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5525         const u8 *hdr_start;
5526         int ret;
5527
5528         if (syn_skb) {
5529                 /* sk is a request_sock here */
5530
5531                 if (optname == TCP_BPF_SYN) {
5532                         hdr_start = syn_skb->data;
5533                         ret = tcp_hdrlen(syn_skb);
5534                 } else if (optname == TCP_BPF_SYN_IP) {
5535                         hdr_start = skb_network_header(syn_skb);
5536                         ret = skb_network_header_len(syn_skb) +
5537                                 tcp_hdrlen(syn_skb);
5538                 } else {
5539                         /* optname == TCP_BPF_SYN_MAC */
5540                         hdr_start = skb_mac_header(syn_skb);
5541                         ret = skb_mac_header_len(syn_skb) +
5542                                 skb_network_header_len(syn_skb) +
5543                                 tcp_hdrlen(syn_skb);
5544                 }
5545         } else {
5546                 struct sock *sk = bpf_sock->sk;
5547                 struct saved_syn *saved_syn;
5548
5549                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5550                         /* synack retransmit. bpf_sock->syn_skb will
5551                          * not be available.  It has to resort to
5552                          * saved_syn (if it is saved).
5553                          */
5554                         saved_syn = inet_reqsk(sk)->saved_syn;
5555                 else
5556                         saved_syn = tcp_sk(sk)->saved_syn;
5557
5558                 if (!saved_syn)
5559                         return -ENOENT;
5560
5561                 if (optname == TCP_BPF_SYN) {
5562                         hdr_start = saved_syn->data +
5563                                 saved_syn->mac_hdrlen +
5564                                 saved_syn->network_hdrlen;
5565                         ret = saved_syn->tcp_hdrlen;
5566                 } else if (optname == TCP_BPF_SYN_IP) {
5567                         hdr_start = saved_syn->data +
5568                                 saved_syn->mac_hdrlen;
5569                         ret = saved_syn->network_hdrlen +
5570                                 saved_syn->tcp_hdrlen;
5571                 } else {
5572                         /* optname == TCP_BPF_SYN_MAC */
5573
5574                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5575                         if (!saved_syn->mac_hdrlen)
5576                                 return -ENOENT;
5577
5578                         hdr_start = saved_syn->data;
5579                         ret = saved_syn->mac_hdrlen +
5580                                 saved_syn->network_hdrlen +
5581                                 saved_syn->tcp_hdrlen;
5582                 }
5583         }
5584
5585         *start = hdr_start;
5586         return ret;
5587 }
5588
5589 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5590            int, level, int, optname, char *, optval, int, optlen)
5591 {
5592         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5593             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5594                 int ret, copy_len = 0;
5595                 const u8 *start;
5596
5597                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5598                 if (ret > 0) {
5599                         copy_len = ret;
5600                         if (optlen < copy_len) {
5601                                 copy_len = optlen;
5602                                 ret = -ENOSPC;
5603                         }
5604
5605                         memcpy(optval, start, copy_len);
5606                 }
5607
5608                 /* Zero out unused buffer at the end */
5609                 memset(optval + copy_len, 0, optlen - copy_len);
5610
5611                 return ret;
5612         }
5613
5614         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5615 }
5616
5617 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5618         .func           = bpf_sock_ops_getsockopt,
5619         .gpl_only       = false,
5620         .ret_type       = RET_INTEGER,
5621         .arg1_type      = ARG_PTR_TO_CTX,
5622         .arg2_type      = ARG_ANYTHING,
5623         .arg3_type      = ARG_ANYTHING,
5624         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5625         .arg5_type      = ARG_CONST_SIZE,
5626 };
5627
5628 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5629            int, argval)
5630 {
5631         struct sock *sk = bpf_sock->sk;
5632         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5633
5634         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5635                 return -EINVAL;
5636
5637         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5638
5639         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5640 }
5641
5642 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5643         .func           = bpf_sock_ops_cb_flags_set,
5644         .gpl_only       = false,
5645         .ret_type       = RET_INTEGER,
5646         .arg1_type      = ARG_PTR_TO_CTX,
5647         .arg2_type      = ARG_ANYTHING,
5648 };
5649
5650 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5651 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5652
5653 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5654            int, addr_len)
5655 {
5656 #ifdef CONFIG_INET
5657         struct sock *sk = ctx->sk;
5658         u32 flags = BIND_FROM_BPF;
5659         int err;
5660
5661         err = -EINVAL;
5662         if (addr_len < offsetofend(struct sockaddr, sa_family))
5663                 return err;
5664         if (addr->sa_family == AF_INET) {
5665                 if (addr_len < sizeof(struct sockaddr_in))
5666                         return err;
5667                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5668                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5669                 return __inet_bind(sk, addr, addr_len, flags);
5670 #if IS_ENABLED(CONFIG_IPV6)
5671         } else if (addr->sa_family == AF_INET6) {
5672                 if (addr_len < SIN6_LEN_RFC2133)
5673                         return err;
5674                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5675                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5676                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5677                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5678                  */
5679                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5680 #endif /* CONFIG_IPV6 */
5681         }
5682 #endif /* CONFIG_INET */
5683
5684         return -EAFNOSUPPORT;
5685 }
5686
5687 static const struct bpf_func_proto bpf_bind_proto = {
5688         .func           = bpf_bind,
5689         .gpl_only       = false,
5690         .ret_type       = RET_INTEGER,
5691         .arg1_type      = ARG_PTR_TO_CTX,
5692         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5693         .arg3_type      = ARG_CONST_SIZE,
5694 };
5695
5696 #ifdef CONFIG_XFRM
5697
5698 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5699     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5700
5701 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5702 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5703
5704 #endif
5705
5706 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5707            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5708 {
5709         const struct sec_path *sp = skb_sec_path(skb);
5710         const struct xfrm_state *x;
5711
5712         if (!sp || unlikely(index >= sp->len || flags))
5713                 goto err_clear;
5714
5715         x = sp->xvec[index];
5716
5717         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5718                 goto err_clear;
5719
5720         to->reqid = x->props.reqid;
5721         to->spi = x->id.spi;
5722         to->family = x->props.family;
5723         to->ext = 0;
5724
5725         if (to->family == AF_INET6) {
5726                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5727                        sizeof(to->remote_ipv6));
5728         } else {
5729                 to->remote_ipv4 = x->props.saddr.a4;
5730                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5731         }
5732
5733         return 0;
5734 err_clear:
5735         memset(to, 0, size);
5736         return -EINVAL;
5737 }
5738
5739 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5740         .func           = bpf_skb_get_xfrm_state,
5741         .gpl_only       = false,
5742         .ret_type       = RET_INTEGER,
5743         .arg1_type      = ARG_PTR_TO_CTX,
5744         .arg2_type      = ARG_ANYTHING,
5745         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5746         .arg4_type      = ARG_CONST_SIZE,
5747         .arg5_type      = ARG_ANYTHING,
5748 };
5749 #endif
5750
5751 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5752 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu)
5753 {
5754         params->h_vlan_TCI = 0;
5755         params->h_vlan_proto = 0;
5756         if (mtu)
5757                 params->mtu_result = mtu; /* union with tot_len */
5758
5759         return 0;
5760 }
5761 #endif
5762
5763 #if IS_ENABLED(CONFIG_INET)
5764 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5765                                u32 flags, bool check_mtu)
5766 {
5767         struct fib_nh_common *nhc;
5768         struct in_device *in_dev;
5769         struct neighbour *neigh;
5770         struct net_device *dev;
5771         struct fib_result res;
5772         struct flowi4 fl4;
5773         u32 mtu = 0;
5774         int err;
5775
5776         dev = dev_get_by_index_rcu(net, params->ifindex);
5777         if (unlikely(!dev))
5778                 return -ENODEV;
5779
5780         /* verify forwarding is enabled on this interface */
5781         in_dev = __in_dev_get_rcu(dev);
5782         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5783                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5784
5785         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5786                 fl4.flowi4_iif = 1;
5787                 fl4.flowi4_oif = params->ifindex;
5788         } else {
5789                 fl4.flowi4_iif = params->ifindex;
5790                 fl4.flowi4_oif = 0;
5791         }
5792         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5793         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5794         fl4.flowi4_flags = 0;
5795
5796         fl4.flowi4_proto = params->l4_protocol;
5797         fl4.daddr = params->ipv4_dst;
5798         fl4.saddr = params->ipv4_src;
5799         fl4.fl4_sport = params->sport;
5800         fl4.fl4_dport = params->dport;
5801         fl4.flowi4_multipath_hash = 0;
5802
5803         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5804                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5805                 struct fib_table *tb;
5806
5807                 if (flags & BPF_FIB_LOOKUP_TBID) {
5808                         tbid = params->tbid;
5809                         /* zero out for vlan output */
5810                         params->tbid = 0;
5811                 }
5812
5813                 tb = fib_get_table(net, tbid);
5814                 if (unlikely(!tb))
5815                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5816
5817                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5818         } else {
5819                 fl4.flowi4_mark = 0;
5820                 fl4.flowi4_secid = 0;
5821                 fl4.flowi4_tun_key.tun_id = 0;
5822                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5823
5824                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5825         }
5826
5827         if (err) {
5828                 /* map fib lookup errors to RTN_ type */
5829                 if (err == -EINVAL)
5830                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5831                 if (err == -EHOSTUNREACH)
5832                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5833                 if (err == -EACCES)
5834                         return BPF_FIB_LKUP_RET_PROHIBIT;
5835
5836                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5837         }
5838
5839         if (res.type != RTN_UNICAST)
5840                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5841
5842         if (fib_info_num_path(res.fi) > 1)
5843                 fib_select_path(net, &res, &fl4, NULL);
5844
5845         if (check_mtu) {
5846                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5847                 if (params->tot_len > mtu) {
5848                         params->mtu_result = mtu; /* union with tot_len */
5849                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5850                 }
5851         }
5852
5853         nhc = res.nhc;
5854
5855         /* do not handle lwt encaps right now */
5856         if (nhc->nhc_lwtstate)
5857                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5858
5859         dev = nhc->nhc_dev;
5860
5861         params->rt_metric = res.fi->fib_priority;
5862         params->ifindex = dev->ifindex;
5863
5864         /* xdp and cls_bpf programs are run in RCU-bh so
5865          * rcu_read_lock_bh is not needed here
5866          */
5867         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5868                 if (nhc->nhc_gw_family)
5869                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5870         } else {
5871                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5872
5873                 params->family = AF_INET6;
5874                 *dst = nhc->nhc_gw.ipv6;
5875         }
5876
5877         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
5878                 goto set_fwd_params;
5879
5880         if (likely(nhc->nhc_gw_family != AF_INET6))
5881                 neigh = __ipv4_neigh_lookup_noref(dev,
5882                                                   (__force u32)params->ipv4_dst);
5883         else
5884                 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst);
5885
5886         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
5887                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5888         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5889         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5890
5891 set_fwd_params:
5892         return bpf_fib_set_fwd_params(params, mtu);
5893 }
5894 #endif
5895
5896 #if IS_ENABLED(CONFIG_IPV6)
5897 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5898                                u32 flags, bool check_mtu)
5899 {
5900         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5901         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5902         struct fib6_result res = {};
5903         struct neighbour *neigh;
5904         struct net_device *dev;
5905         struct inet6_dev *idev;
5906         struct flowi6 fl6;
5907         int strict = 0;
5908         int oif, err;
5909         u32 mtu = 0;
5910
5911         /* link local addresses are never forwarded */
5912         if (rt6_need_strict(dst) || rt6_need_strict(src))
5913                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5914
5915         dev = dev_get_by_index_rcu(net, params->ifindex);
5916         if (unlikely(!dev))
5917                 return -ENODEV;
5918
5919         idev = __in6_dev_get_safely(dev);
5920         if (unlikely(!idev || !idev->cnf.forwarding))
5921                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5922
5923         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5924                 fl6.flowi6_iif = 1;
5925                 oif = fl6.flowi6_oif = params->ifindex;
5926         } else {
5927                 oif = fl6.flowi6_iif = params->ifindex;
5928                 fl6.flowi6_oif = 0;
5929                 strict = RT6_LOOKUP_F_HAS_SADDR;
5930         }
5931         fl6.flowlabel = params->flowinfo;
5932         fl6.flowi6_scope = 0;
5933         fl6.flowi6_flags = 0;
5934         fl6.mp_hash = 0;
5935
5936         fl6.flowi6_proto = params->l4_protocol;
5937         fl6.daddr = *dst;
5938         fl6.saddr = *src;
5939         fl6.fl6_sport = params->sport;
5940         fl6.fl6_dport = params->dport;
5941
5942         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5943                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5944                 struct fib6_table *tb;
5945
5946                 if (flags & BPF_FIB_LOOKUP_TBID) {
5947                         tbid = params->tbid;
5948                         /* zero out for vlan output */
5949                         params->tbid = 0;
5950                 }
5951
5952                 tb = ipv6_stub->fib6_get_table(net, tbid);
5953                 if (unlikely(!tb))
5954                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5955
5956                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5957                                                    strict);
5958         } else {
5959                 fl6.flowi6_mark = 0;
5960                 fl6.flowi6_secid = 0;
5961                 fl6.flowi6_tun_key.tun_id = 0;
5962                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5963
5964                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5965         }
5966
5967         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5968                      res.f6i == net->ipv6.fib6_null_entry))
5969                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5970
5971         switch (res.fib6_type) {
5972         /* only unicast is forwarded */
5973         case RTN_UNICAST:
5974                 break;
5975         case RTN_BLACKHOLE:
5976                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5977         case RTN_UNREACHABLE:
5978                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5979         case RTN_PROHIBIT:
5980                 return BPF_FIB_LKUP_RET_PROHIBIT;
5981         default:
5982                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5983         }
5984
5985         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5986                                     fl6.flowi6_oif != 0, NULL, strict);
5987
5988         if (check_mtu) {
5989                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5990                 if (params->tot_len > mtu) {
5991                         params->mtu_result = mtu; /* union with tot_len */
5992                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5993                 }
5994         }
5995
5996         if (res.nh->fib_nh_lws)
5997                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5998
5999         if (res.nh->fib_nh_gw_family)
6000                 *dst = res.nh->fib_nh_gw6;
6001
6002         dev = res.nh->fib_nh_dev;
6003         params->rt_metric = res.f6i->fib6_metric;
6004         params->ifindex = dev->ifindex;
6005
6006         if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH)
6007                 goto set_fwd_params;
6008
6009         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
6010          * not needed here.
6011          */
6012         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
6013         if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID))
6014                 return BPF_FIB_LKUP_RET_NO_NEIGH;
6015         memcpy(params->dmac, neigh->ha, ETH_ALEN);
6016         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
6017
6018 set_fwd_params:
6019         return bpf_fib_set_fwd_params(params, mtu);
6020 }
6021 #endif
6022
6023 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \
6024                              BPF_FIB_LOOKUP_SKIP_NEIGH | BPF_FIB_LOOKUP_TBID)
6025
6026 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
6027            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6028 {
6029         if (plen < sizeof(*params))
6030                 return -EINVAL;
6031
6032         if (flags & ~BPF_FIB_LOOKUP_MASK)
6033                 return -EINVAL;
6034
6035         switch (params->family) {
6036 #if IS_ENABLED(CONFIG_INET)
6037         case AF_INET:
6038                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
6039                                            flags, true);
6040 #endif
6041 #if IS_ENABLED(CONFIG_IPV6)
6042         case AF_INET6:
6043                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
6044                                            flags, true);
6045 #endif
6046         }
6047         return -EAFNOSUPPORT;
6048 }
6049
6050 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
6051         .func           = bpf_xdp_fib_lookup,
6052         .gpl_only       = true,
6053         .ret_type       = RET_INTEGER,
6054         .arg1_type      = ARG_PTR_TO_CTX,
6055         .arg2_type      = ARG_PTR_TO_MEM,
6056         .arg3_type      = ARG_CONST_SIZE,
6057         .arg4_type      = ARG_ANYTHING,
6058 };
6059
6060 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6061            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6062 {
6063         struct net *net = dev_net(skb->dev);
6064         int rc = -EAFNOSUPPORT;
6065         bool check_mtu = false;
6066
6067         if (plen < sizeof(*params))
6068                 return -EINVAL;
6069
6070         if (flags & ~BPF_FIB_LOOKUP_MASK)
6071                 return -EINVAL;
6072
6073         if (params->tot_len)
6074                 check_mtu = true;
6075
6076         switch (params->family) {
6077 #if IS_ENABLED(CONFIG_INET)
6078         case AF_INET:
6079                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6080                 break;
6081 #endif
6082 #if IS_ENABLED(CONFIG_IPV6)
6083         case AF_INET6:
6084                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6085                 break;
6086 #endif
6087         }
6088
6089         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6090                 struct net_device *dev;
6091
6092                 /* When tot_len isn't provided by user, check skb
6093                  * against MTU of FIB lookup resulting net_device
6094                  */
6095                 dev = dev_get_by_index_rcu(net, params->ifindex);
6096                 if (!is_skb_forwardable(dev, skb))
6097                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6098
6099                 params->mtu_result = dev->mtu; /* union with tot_len */
6100         }
6101
6102         return rc;
6103 }
6104
6105 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6106         .func           = bpf_skb_fib_lookup,
6107         .gpl_only       = true,
6108         .ret_type       = RET_INTEGER,
6109         .arg1_type      = ARG_PTR_TO_CTX,
6110         .arg2_type      = ARG_PTR_TO_MEM,
6111         .arg3_type      = ARG_CONST_SIZE,
6112         .arg4_type      = ARG_ANYTHING,
6113 };
6114
6115 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6116                                             u32 ifindex)
6117 {
6118         struct net *netns = dev_net(dev_curr);
6119
6120         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6121         if (ifindex == 0)
6122                 return dev_curr;
6123
6124         return dev_get_by_index_rcu(netns, ifindex);
6125 }
6126
6127 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6128            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6129 {
6130         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6131         struct net_device *dev = skb->dev;
6132         int skb_len, dev_len;
6133         int mtu;
6134
6135         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6136                 return -EINVAL;
6137
6138         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6139                 return -EINVAL;
6140
6141         dev = __dev_via_ifindex(dev, ifindex);
6142         if (unlikely(!dev))
6143                 return -ENODEV;
6144
6145         mtu = READ_ONCE(dev->mtu);
6146
6147         dev_len = mtu + dev->hard_header_len;
6148
6149         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6150         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6151
6152         skb_len += len_diff; /* minus result pass check */
6153         if (skb_len <= dev_len) {
6154                 ret = BPF_MTU_CHK_RET_SUCCESS;
6155                 goto out;
6156         }
6157         /* At this point, skb->len exceed MTU, but as it include length of all
6158          * segments, it can still be below MTU.  The SKB can possibly get
6159          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6160          * must choose if segs are to be MTU checked.
6161          */
6162         if (skb_is_gso(skb)) {
6163                 ret = BPF_MTU_CHK_RET_SUCCESS;
6164
6165                 if (flags & BPF_MTU_CHK_SEGS &&
6166                     !skb_gso_validate_network_len(skb, mtu))
6167                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6168         }
6169 out:
6170         /* BPF verifier guarantees valid pointer */
6171         *mtu_len = mtu;
6172
6173         return ret;
6174 }
6175
6176 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6177            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6178 {
6179         struct net_device *dev = xdp->rxq->dev;
6180         int xdp_len = xdp->data_end - xdp->data;
6181         int ret = BPF_MTU_CHK_RET_SUCCESS;
6182         int mtu, dev_len;
6183
6184         /* XDP variant doesn't support multi-buffer segment check (yet) */
6185         if (unlikely(flags))
6186                 return -EINVAL;
6187
6188         dev = __dev_via_ifindex(dev, ifindex);
6189         if (unlikely(!dev))
6190                 return -ENODEV;
6191
6192         mtu = READ_ONCE(dev->mtu);
6193
6194         /* Add L2-header as dev MTU is L3 size */
6195         dev_len = mtu + dev->hard_header_len;
6196
6197         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6198         if (*mtu_len)
6199                 xdp_len = *mtu_len + dev->hard_header_len;
6200
6201         xdp_len += len_diff; /* minus result pass check */
6202         if (xdp_len > dev_len)
6203                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6204
6205         /* BPF verifier guarantees valid pointer */
6206         *mtu_len = mtu;
6207
6208         return ret;
6209 }
6210
6211 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6212         .func           = bpf_skb_check_mtu,
6213         .gpl_only       = true,
6214         .ret_type       = RET_INTEGER,
6215         .arg1_type      = ARG_PTR_TO_CTX,
6216         .arg2_type      = ARG_ANYTHING,
6217         .arg3_type      = ARG_PTR_TO_INT,
6218         .arg4_type      = ARG_ANYTHING,
6219         .arg5_type      = ARG_ANYTHING,
6220 };
6221
6222 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6223         .func           = bpf_xdp_check_mtu,
6224         .gpl_only       = true,
6225         .ret_type       = RET_INTEGER,
6226         .arg1_type      = ARG_PTR_TO_CTX,
6227         .arg2_type      = ARG_ANYTHING,
6228         .arg3_type      = ARG_PTR_TO_INT,
6229         .arg4_type      = ARG_ANYTHING,
6230         .arg5_type      = ARG_ANYTHING,
6231 };
6232
6233 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6234 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6235 {
6236         int err;
6237         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6238
6239         if (!seg6_validate_srh(srh, len, false))
6240                 return -EINVAL;
6241
6242         switch (type) {
6243         case BPF_LWT_ENCAP_SEG6_INLINE:
6244                 if (skb->protocol != htons(ETH_P_IPV6))
6245                         return -EBADMSG;
6246
6247                 err = seg6_do_srh_inline(skb, srh);
6248                 break;
6249         case BPF_LWT_ENCAP_SEG6:
6250                 skb_reset_inner_headers(skb);
6251                 skb->encapsulation = 1;
6252                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6253                 break;
6254         default:
6255                 return -EINVAL;
6256         }
6257
6258         bpf_compute_data_pointers(skb);
6259         if (err)
6260                 return err;
6261
6262         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6263
6264         return seg6_lookup_nexthop(skb, NULL, 0);
6265 }
6266 #endif /* CONFIG_IPV6_SEG6_BPF */
6267
6268 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6269 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6270                              bool ingress)
6271 {
6272         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6273 }
6274 #endif
6275
6276 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6277            u32, len)
6278 {
6279         switch (type) {
6280 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6281         case BPF_LWT_ENCAP_SEG6:
6282         case BPF_LWT_ENCAP_SEG6_INLINE:
6283                 return bpf_push_seg6_encap(skb, type, hdr, len);
6284 #endif
6285 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6286         case BPF_LWT_ENCAP_IP:
6287                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6288 #endif
6289         default:
6290                 return -EINVAL;
6291         }
6292 }
6293
6294 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6295            void *, hdr, u32, len)
6296 {
6297         switch (type) {
6298 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6299         case BPF_LWT_ENCAP_IP:
6300                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6301 #endif
6302         default:
6303                 return -EINVAL;
6304         }
6305 }
6306
6307 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6308         .func           = bpf_lwt_in_push_encap,
6309         .gpl_only       = false,
6310         .ret_type       = RET_INTEGER,
6311         .arg1_type      = ARG_PTR_TO_CTX,
6312         .arg2_type      = ARG_ANYTHING,
6313         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6314         .arg4_type      = ARG_CONST_SIZE
6315 };
6316
6317 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6318         .func           = bpf_lwt_xmit_push_encap,
6319         .gpl_only       = false,
6320         .ret_type       = RET_INTEGER,
6321         .arg1_type      = ARG_PTR_TO_CTX,
6322         .arg2_type      = ARG_ANYTHING,
6323         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6324         .arg4_type      = ARG_CONST_SIZE
6325 };
6326
6327 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6328 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6329            const void *, from, u32, len)
6330 {
6331         struct seg6_bpf_srh_state *srh_state =
6332                 this_cpu_ptr(&seg6_bpf_srh_states);
6333         struct ipv6_sr_hdr *srh = srh_state->srh;
6334         void *srh_tlvs, *srh_end, *ptr;
6335         int srhoff = 0;
6336
6337         if (srh == NULL)
6338                 return -EINVAL;
6339
6340         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6341         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6342
6343         ptr = skb->data + offset;
6344         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6345                 srh_state->valid = false;
6346         else if (ptr < (void *)&srh->flags ||
6347                  ptr + len > (void *)&srh->segments)
6348                 return -EFAULT;
6349
6350         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6351                 return -EFAULT;
6352         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6353                 return -EINVAL;
6354         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6355
6356         memcpy(skb->data + offset, from, len);
6357         return 0;
6358 }
6359
6360 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6361         .func           = bpf_lwt_seg6_store_bytes,
6362         .gpl_only       = false,
6363         .ret_type       = RET_INTEGER,
6364         .arg1_type      = ARG_PTR_TO_CTX,
6365         .arg2_type      = ARG_ANYTHING,
6366         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6367         .arg4_type      = ARG_CONST_SIZE
6368 };
6369
6370 static void bpf_update_srh_state(struct sk_buff *skb)
6371 {
6372         struct seg6_bpf_srh_state *srh_state =
6373                 this_cpu_ptr(&seg6_bpf_srh_states);
6374         int srhoff = 0;
6375
6376         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6377                 srh_state->srh = NULL;
6378         } else {
6379                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6380                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6381                 srh_state->valid = true;
6382         }
6383 }
6384
6385 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6386            u32, action, void *, param, u32, param_len)
6387 {
6388         struct seg6_bpf_srh_state *srh_state =
6389                 this_cpu_ptr(&seg6_bpf_srh_states);
6390         int hdroff = 0;
6391         int err;
6392
6393         switch (action) {
6394         case SEG6_LOCAL_ACTION_END_X:
6395                 if (!seg6_bpf_has_valid_srh(skb))
6396                         return -EBADMSG;
6397                 if (param_len != sizeof(struct in6_addr))
6398                         return -EINVAL;
6399                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6400         case SEG6_LOCAL_ACTION_END_T:
6401                 if (!seg6_bpf_has_valid_srh(skb))
6402                         return -EBADMSG;
6403                 if (param_len != sizeof(int))
6404                         return -EINVAL;
6405                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6406         case SEG6_LOCAL_ACTION_END_DT6:
6407                 if (!seg6_bpf_has_valid_srh(skb))
6408                         return -EBADMSG;
6409                 if (param_len != sizeof(int))
6410                         return -EINVAL;
6411
6412                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6413                         return -EBADMSG;
6414                 if (!pskb_pull(skb, hdroff))
6415                         return -EBADMSG;
6416
6417                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6418                 skb_reset_network_header(skb);
6419                 skb_reset_transport_header(skb);
6420                 skb->encapsulation = 0;
6421
6422                 bpf_compute_data_pointers(skb);
6423                 bpf_update_srh_state(skb);
6424                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6425         case SEG6_LOCAL_ACTION_END_B6:
6426                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6427                         return -EBADMSG;
6428                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6429                                           param, param_len);
6430                 if (!err)
6431                         bpf_update_srh_state(skb);
6432
6433                 return err;
6434         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6435                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6436                         return -EBADMSG;
6437                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6438                                           param, param_len);
6439                 if (!err)
6440                         bpf_update_srh_state(skb);
6441
6442                 return err;
6443         default:
6444                 return -EINVAL;
6445         }
6446 }
6447
6448 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6449         .func           = bpf_lwt_seg6_action,
6450         .gpl_only       = false,
6451         .ret_type       = RET_INTEGER,
6452         .arg1_type      = ARG_PTR_TO_CTX,
6453         .arg2_type      = ARG_ANYTHING,
6454         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6455         .arg4_type      = ARG_CONST_SIZE
6456 };
6457
6458 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6459            s32, len)
6460 {
6461         struct seg6_bpf_srh_state *srh_state =
6462                 this_cpu_ptr(&seg6_bpf_srh_states);
6463         struct ipv6_sr_hdr *srh = srh_state->srh;
6464         void *srh_end, *srh_tlvs, *ptr;
6465         struct ipv6hdr *hdr;
6466         int srhoff = 0;
6467         int ret;
6468
6469         if (unlikely(srh == NULL))
6470                 return -EINVAL;
6471
6472         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6473                         ((srh->first_segment + 1) << 4));
6474         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6475                         srh_state->hdrlen);
6476         ptr = skb->data + offset;
6477
6478         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6479                 return -EFAULT;
6480         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6481                 return -EFAULT;
6482
6483         if (len > 0) {
6484                 ret = skb_cow_head(skb, len);
6485                 if (unlikely(ret < 0))
6486                         return ret;
6487
6488                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6489         } else {
6490                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6491         }
6492
6493         bpf_compute_data_pointers(skb);
6494         if (unlikely(ret < 0))
6495                 return ret;
6496
6497         hdr = (struct ipv6hdr *)skb->data;
6498         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6499
6500         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6501                 return -EINVAL;
6502         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6503         srh_state->hdrlen += len;
6504         srh_state->valid = false;
6505         return 0;
6506 }
6507
6508 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6509         .func           = bpf_lwt_seg6_adjust_srh,
6510         .gpl_only       = false,
6511         .ret_type       = RET_INTEGER,
6512         .arg1_type      = ARG_PTR_TO_CTX,
6513         .arg2_type      = ARG_ANYTHING,
6514         .arg3_type      = ARG_ANYTHING,
6515 };
6516 #endif /* CONFIG_IPV6_SEG6_BPF */
6517
6518 #ifdef CONFIG_INET
6519 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6520                               int dif, int sdif, u8 family, u8 proto)
6521 {
6522         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6523         bool refcounted = false;
6524         struct sock *sk = NULL;
6525
6526         if (family == AF_INET) {
6527                 __be32 src4 = tuple->ipv4.saddr;
6528                 __be32 dst4 = tuple->ipv4.daddr;
6529
6530                 if (proto == IPPROTO_TCP)
6531                         sk = __inet_lookup(net, hinfo, NULL, 0,
6532                                            src4, tuple->ipv4.sport,
6533                                            dst4, tuple->ipv4.dport,
6534                                            dif, sdif, &refcounted);
6535                 else
6536                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6537                                                dst4, tuple->ipv4.dport,
6538                                                dif, sdif, net->ipv4.udp_table, NULL);
6539 #if IS_ENABLED(CONFIG_IPV6)
6540         } else {
6541                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6542                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6543
6544                 if (proto == IPPROTO_TCP)
6545                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6546                                             src6, tuple->ipv6.sport,
6547                                             dst6, ntohs(tuple->ipv6.dport),
6548                                             dif, sdif, &refcounted);
6549                 else if (likely(ipv6_bpf_stub))
6550                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6551                                                             src6, tuple->ipv6.sport,
6552                                                             dst6, tuple->ipv6.dport,
6553                                                             dif, sdif,
6554                                                             net->ipv4.udp_table, NULL);
6555 #endif
6556         }
6557
6558         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6559                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6560                 sk = NULL;
6561         }
6562         return sk;
6563 }
6564
6565 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6566  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6567  */
6568 static struct sock *
6569 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6570                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6571                  u64 flags, int sdif)
6572 {
6573         struct sock *sk = NULL;
6574         struct net *net;
6575         u8 family;
6576
6577         if (len == sizeof(tuple->ipv4))
6578                 family = AF_INET;
6579         else if (len == sizeof(tuple->ipv6))
6580                 family = AF_INET6;
6581         else
6582                 return NULL;
6583
6584         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6585                 goto out;
6586
6587         if (sdif < 0) {
6588                 if (family == AF_INET)
6589                         sdif = inet_sdif(skb);
6590                 else
6591                         sdif = inet6_sdif(skb);
6592         }
6593
6594         if ((s32)netns_id < 0) {
6595                 net = caller_net;
6596                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6597         } else {
6598                 net = get_net_ns_by_id(caller_net, netns_id);
6599                 if (unlikely(!net))
6600                         goto out;
6601                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6602                 put_net(net);
6603         }
6604
6605 out:
6606         return sk;
6607 }
6608
6609 static struct sock *
6610 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6611                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6612                 u64 flags, int sdif)
6613 {
6614         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6615                                            ifindex, proto, netns_id, flags,
6616                                            sdif);
6617
6618         if (sk) {
6619                 struct sock *sk2 = sk_to_full_sk(sk);
6620
6621                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6622                  * sock refcnt is decremented to prevent a request_sock leak.
6623                  */
6624                 if (!sk_fullsock(sk2))
6625                         sk2 = NULL;
6626                 if (sk2 != sk) {
6627                         sock_gen_put(sk);
6628                         /* Ensure there is no need to bump sk2 refcnt */
6629                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6630                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6631                                 return NULL;
6632                         }
6633                         sk = sk2;
6634                 }
6635         }
6636
6637         return sk;
6638 }
6639
6640 static struct sock *
6641 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6642                u8 proto, u64 netns_id, u64 flags)
6643 {
6644         struct net *caller_net;
6645         int ifindex;
6646
6647         if (skb->dev) {
6648                 caller_net = dev_net(skb->dev);
6649                 ifindex = skb->dev->ifindex;
6650         } else {
6651                 caller_net = sock_net(skb->sk);
6652                 ifindex = 0;
6653         }
6654
6655         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6656                                 netns_id, flags, -1);
6657 }
6658
6659 static struct sock *
6660 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6661               u8 proto, u64 netns_id, u64 flags)
6662 {
6663         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6664                                          flags);
6665
6666         if (sk) {
6667                 struct sock *sk2 = sk_to_full_sk(sk);
6668
6669                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6670                  * sock refcnt is decremented to prevent a request_sock leak.
6671                  */
6672                 if (!sk_fullsock(sk2))
6673                         sk2 = NULL;
6674                 if (sk2 != sk) {
6675                         sock_gen_put(sk);
6676                         /* Ensure there is no need to bump sk2 refcnt */
6677                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6678                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6679                                 return NULL;
6680                         }
6681                         sk = sk2;
6682                 }
6683         }
6684
6685         return sk;
6686 }
6687
6688 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6689            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6690 {
6691         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6692                                              netns_id, flags);
6693 }
6694
6695 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6696         .func           = bpf_skc_lookup_tcp,
6697         .gpl_only       = false,
6698         .pkt_access     = true,
6699         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6700         .arg1_type      = ARG_PTR_TO_CTX,
6701         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6702         .arg3_type      = ARG_CONST_SIZE,
6703         .arg4_type      = ARG_ANYTHING,
6704         .arg5_type      = ARG_ANYTHING,
6705 };
6706
6707 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6708            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6709 {
6710         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6711                                             netns_id, flags);
6712 }
6713
6714 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6715         .func           = bpf_sk_lookup_tcp,
6716         .gpl_only       = false,
6717         .pkt_access     = true,
6718         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6719         .arg1_type      = ARG_PTR_TO_CTX,
6720         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6721         .arg3_type      = ARG_CONST_SIZE,
6722         .arg4_type      = ARG_ANYTHING,
6723         .arg5_type      = ARG_ANYTHING,
6724 };
6725
6726 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6727            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6728 {
6729         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6730                                             netns_id, flags);
6731 }
6732
6733 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6734         .func           = bpf_sk_lookup_udp,
6735         .gpl_only       = false,
6736         .pkt_access     = true,
6737         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6738         .arg1_type      = ARG_PTR_TO_CTX,
6739         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6740         .arg3_type      = ARG_CONST_SIZE,
6741         .arg4_type      = ARG_ANYTHING,
6742         .arg5_type      = ARG_ANYTHING,
6743 };
6744
6745 BPF_CALL_5(bpf_tc_skc_lookup_tcp, struct sk_buff *, skb,
6746            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6747 {
6748         struct net_device *dev = skb->dev;
6749         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6750         struct net *caller_net = dev_net(dev);
6751
6752         return (unsigned long)__bpf_skc_lookup(skb, tuple, len, caller_net,
6753                                                ifindex, IPPROTO_TCP, netns_id,
6754                                                flags, sdif);
6755 }
6756
6757 static const struct bpf_func_proto bpf_tc_skc_lookup_tcp_proto = {
6758         .func           = bpf_tc_skc_lookup_tcp,
6759         .gpl_only       = false,
6760         .pkt_access     = true,
6761         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6762         .arg1_type      = ARG_PTR_TO_CTX,
6763         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6764         .arg3_type      = ARG_CONST_SIZE,
6765         .arg4_type      = ARG_ANYTHING,
6766         .arg5_type      = ARG_ANYTHING,
6767 };
6768
6769 BPF_CALL_5(bpf_tc_sk_lookup_tcp, struct sk_buff *, skb,
6770            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6771 {
6772         struct net_device *dev = skb->dev;
6773         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6774         struct net *caller_net = dev_net(dev);
6775
6776         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6777                                               ifindex, IPPROTO_TCP, netns_id,
6778                                               flags, sdif);
6779 }
6780
6781 static const struct bpf_func_proto bpf_tc_sk_lookup_tcp_proto = {
6782         .func           = bpf_tc_sk_lookup_tcp,
6783         .gpl_only       = false,
6784         .pkt_access     = true,
6785         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6786         .arg1_type      = ARG_PTR_TO_CTX,
6787         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6788         .arg3_type      = ARG_CONST_SIZE,
6789         .arg4_type      = ARG_ANYTHING,
6790         .arg5_type      = ARG_ANYTHING,
6791 };
6792
6793 BPF_CALL_5(bpf_tc_sk_lookup_udp, struct sk_buff *, skb,
6794            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6795 {
6796         struct net_device *dev = skb->dev;
6797         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6798         struct net *caller_net = dev_net(dev);
6799
6800         return (unsigned long)__bpf_sk_lookup(skb, tuple, len, caller_net,
6801                                               ifindex, IPPROTO_UDP, netns_id,
6802                                               flags, sdif);
6803 }
6804
6805 static const struct bpf_func_proto bpf_tc_sk_lookup_udp_proto = {
6806         .func           = bpf_tc_sk_lookup_udp,
6807         .gpl_only       = false,
6808         .pkt_access     = true,
6809         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6810         .arg1_type      = ARG_PTR_TO_CTX,
6811         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6812         .arg3_type      = ARG_CONST_SIZE,
6813         .arg4_type      = ARG_ANYTHING,
6814         .arg5_type      = ARG_ANYTHING,
6815 };
6816
6817 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6818 {
6819         if (sk && sk_is_refcounted(sk))
6820                 sock_gen_put(sk);
6821         return 0;
6822 }
6823
6824 static const struct bpf_func_proto bpf_sk_release_proto = {
6825         .func           = bpf_sk_release,
6826         .gpl_only       = false,
6827         .ret_type       = RET_INTEGER,
6828         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6829 };
6830
6831 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6832            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6833 {
6834         struct net_device *dev = ctx->rxq->dev;
6835         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6836         struct net *caller_net = dev_net(dev);
6837
6838         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6839                                               ifindex, IPPROTO_UDP, netns_id,
6840                                               flags, sdif);
6841 }
6842
6843 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6844         .func           = bpf_xdp_sk_lookup_udp,
6845         .gpl_only       = false,
6846         .pkt_access     = true,
6847         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6848         .arg1_type      = ARG_PTR_TO_CTX,
6849         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6850         .arg3_type      = ARG_CONST_SIZE,
6851         .arg4_type      = ARG_ANYTHING,
6852         .arg5_type      = ARG_ANYTHING,
6853 };
6854
6855 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6856            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6857 {
6858         struct net_device *dev = ctx->rxq->dev;
6859         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6860         struct net *caller_net = dev_net(dev);
6861
6862         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6863                                                ifindex, IPPROTO_TCP, netns_id,
6864                                                flags, sdif);
6865 }
6866
6867 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6868         .func           = bpf_xdp_skc_lookup_tcp,
6869         .gpl_only       = false,
6870         .pkt_access     = true,
6871         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6872         .arg1_type      = ARG_PTR_TO_CTX,
6873         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6874         .arg3_type      = ARG_CONST_SIZE,
6875         .arg4_type      = ARG_ANYTHING,
6876         .arg5_type      = ARG_ANYTHING,
6877 };
6878
6879 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6880            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6881 {
6882         struct net_device *dev = ctx->rxq->dev;
6883         int ifindex = dev->ifindex, sdif = dev_sdif(dev);
6884         struct net *caller_net = dev_net(dev);
6885
6886         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6887                                               ifindex, IPPROTO_TCP, netns_id,
6888                                               flags, sdif);
6889 }
6890
6891 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6892         .func           = bpf_xdp_sk_lookup_tcp,
6893         .gpl_only       = false,
6894         .pkt_access     = true,
6895         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6896         .arg1_type      = ARG_PTR_TO_CTX,
6897         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6898         .arg3_type      = ARG_CONST_SIZE,
6899         .arg4_type      = ARG_ANYTHING,
6900         .arg5_type      = ARG_ANYTHING,
6901 };
6902
6903 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6904            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6905 {
6906         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6907                                                sock_net(ctx->sk), 0,
6908                                                IPPROTO_TCP, netns_id, flags,
6909                                                -1);
6910 }
6911
6912 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6913         .func           = bpf_sock_addr_skc_lookup_tcp,
6914         .gpl_only       = false,
6915         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6916         .arg1_type      = ARG_PTR_TO_CTX,
6917         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6918         .arg3_type      = ARG_CONST_SIZE,
6919         .arg4_type      = ARG_ANYTHING,
6920         .arg5_type      = ARG_ANYTHING,
6921 };
6922
6923 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6924            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6925 {
6926         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6927                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6928                                               netns_id, flags, -1);
6929 }
6930
6931 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6932         .func           = bpf_sock_addr_sk_lookup_tcp,
6933         .gpl_only       = false,
6934         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6935         .arg1_type      = ARG_PTR_TO_CTX,
6936         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6937         .arg3_type      = ARG_CONST_SIZE,
6938         .arg4_type      = ARG_ANYTHING,
6939         .arg5_type      = ARG_ANYTHING,
6940 };
6941
6942 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6943            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6944 {
6945         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6946                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6947                                               netns_id, flags, -1);
6948 }
6949
6950 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6951         .func           = bpf_sock_addr_sk_lookup_udp,
6952         .gpl_only       = false,
6953         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6954         .arg1_type      = ARG_PTR_TO_CTX,
6955         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6956         .arg3_type      = ARG_CONST_SIZE,
6957         .arg4_type      = ARG_ANYTHING,
6958         .arg5_type      = ARG_ANYTHING,
6959 };
6960
6961 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6962                                   struct bpf_insn_access_aux *info)
6963 {
6964         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6965                                           icsk_retransmits))
6966                 return false;
6967
6968         if (off % size != 0)
6969                 return false;
6970
6971         switch (off) {
6972         case offsetof(struct bpf_tcp_sock, bytes_received):
6973         case offsetof(struct bpf_tcp_sock, bytes_acked):
6974                 return size == sizeof(__u64);
6975         default:
6976                 return size == sizeof(__u32);
6977         }
6978 }
6979
6980 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6981                                     const struct bpf_insn *si,
6982                                     struct bpf_insn *insn_buf,
6983                                     struct bpf_prog *prog, u32 *target_size)
6984 {
6985         struct bpf_insn *insn = insn_buf;
6986
6987 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6988         do {                                                            \
6989                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6990                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6991                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6992                                       si->dst_reg, si->src_reg,         \
6993                                       offsetof(struct tcp_sock, FIELD)); \
6994         } while (0)
6995
6996 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6997         do {                                                            \
6998                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6999                                           FIELD) >                      \
7000                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
7001                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
7002                                         struct inet_connection_sock,    \
7003                                         FIELD),                         \
7004                                       si->dst_reg, si->src_reg,         \
7005                                       offsetof(                         \
7006                                         struct inet_connection_sock,    \
7007                                         FIELD));                        \
7008         } while (0)
7009
7010         BTF_TYPE_EMIT(struct bpf_tcp_sock);
7011
7012         switch (si->off) {
7013         case offsetof(struct bpf_tcp_sock, rtt_min):
7014                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
7015                              sizeof(struct minmax));
7016                 BUILD_BUG_ON(sizeof(struct minmax) <
7017                              sizeof(struct minmax_sample));
7018
7019                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
7020                                       offsetof(struct tcp_sock, rtt_min) +
7021                                       offsetof(struct minmax_sample, v));
7022                 break;
7023         case offsetof(struct bpf_tcp_sock, snd_cwnd):
7024                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
7025                 break;
7026         case offsetof(struct bpf_tcp_sock, srtt_us):
7027                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
7028                 break;
7029         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
7030                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
7031                 break;
7032         case offsetof(struct bpf_tcp_sock, rcv_nxt):
7033                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
7034                 break;
7035         case offsetof(struct bpf_tcp_sock, snd_nxt):
7036                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
7037                 break;
7038         case offsetof(struct bpf_tcp_sock, snd_una):
7039                 BPF_TCP_SOCK_GET_COMMON(snd_una);
7040                 break;
7041         case offsetof(struct bpf_tcp_sock, mss_cache):
7042                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
7043                 break;
7044         case offsetof(struct bpf_tcp_sock, ecn_flags):
7045                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
7046                 break;
7047         case offsetof(struct bpf_tcp_sock, rate_delivered):
7048                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
7049                 break;
7050         case offsetof(struct bpf_tcp_sock, rate_interval_us):
7051                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
7052                 break;
7053         case offsetof(struct bpf_tcp_sock, packets_out):
7054                 BPF_TCP_SOCK_GET_COMMON(packets_out);
7055                 break;
7056         case offsetof(struct bpf_tcp_sock, retrans_out):
7057                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
7058                 break;
7059         case offsetof(struct bpf_tcp_sock, total_retrans):
7060                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
7061                 break;
7062         case offsetof(struct bpf_tcp_sock, segs_in):
7063                 BPF_TCP_SOCK_GET_COMMON(segs_in);
7064                 break;
7065         case offsetof(struct bpf_tcp_sock, data_segs_in):
7066                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
7067                 break;
7068         case offsetof(struct bpf_tcp_sock, segs_out):
7069                 BPF_TCP_SOCK_GET_COMMON(segs_out);
7070                 break;
7071         case offsetof(struct bpf_tcp_sock, data_segs_out):
7072                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
7073                 break;
7074         case offsetof(struct bpf_tcp_sock, lost_out):
7075                 BPF_TCP_SOCK_GET_COMMON(lost_out);
7076                 break;
7077         case offsetof(struct bpf_tcp_sock, sacked_out):
7078                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
7079                 break;
7080         case offsetof(struct bpf_tcp_sock, bytes_received):
7081                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
7082                 break;
7083         case offsetof(struct bpf_tcp_sock, bytes_acked):
7084                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
7085                 break;
7086         case offsetof(struct bpf_tcp_sock, dsack_dups):
7087                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
7088                 break;
7089         case offsetof(struct bpf_tcp_sock, delivered):
7090                 BPF_TCP_SOCK_GET_COMMON(delivered);
7091                 break;
7092         case offsetof(struct bpf_tcp_sock, delivered_ce):
7093                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
7094                 break;
7095         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
7096                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
7097                 break;
7098         }
7099
7100         return insn - insn_buf;
7101 }
7102
7103 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
7104 {
7105         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
7106                 return (unsigned long)sk;
7107
7108         return (unsigned long)NULL;
7109 }
7110
7111 const struct bpf_func_proto bpf_tcp_sock_proto = {
7112         .func           = bpf_tcp_sock,
7113         .gpl_only       = false,
7114         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
7115         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7116 };
7117
7118 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
7119 {
7120         sk = sk_to_full_sk(sk);
7121
7122         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
7123                 return (unsigned long)sk;
7124
7125         return (unsigned long)NULL;
7126 }
7127
7128 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
7129         .func           = bpf_get_listener_sock,
7130         .gpl_only       = false,
7131         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7132         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7133 };
7134
7135 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7136 {
7137         unsigned int iphdr_len;
7138
7139         switch (skb_protocol(skb, true)) {
7140         case cpu_to_be16(ETH_P_IP):
7141                 iphdr_len = sizeof(struct iphdr);
7142                 break;
7143         case cpu_to_be16(ETH_P_IPV6):
7144                 iphdr_len = sizeof(struct ipv6hdr);
7145                 break;
7146         default:
7147                 return 0;
7148         }
7149
7150         if (skb_headlen(skb) < iphdr_len)
7151                 return 0;
7152
7153         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7154                 return 0;
7155
7156         return INET_ECN_set_ce(skb);
7157 }
7158
7159 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7160                                   struct bpf_insn_access_aux *info)
7161 {
7162         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7163                 return false;
7164
7165         if (off % size != 0)
7166                 return false;
7167
7168         switch (off) {
7169         default:
7170                 return size == sizeof(__u32);
7171         }
7172 }
7173
7174 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7175                                     const struct bpf_insn *si,
7176                                     struct bpf_insn *insn_buf,
7177                                     struct bpf_prog *prog, u32 *target_size)
7178 {
7179         struct bpf_insn *insn = insn_buf;
7180
7181 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7182         do {                                                            \
7183                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7184                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7185                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7186                                       si->dst_reg, si->src_reg,         \
7187                                       offsetof(struct xdp_sock, FIELD)); \
7188         } while (0)
7189
7190         switch (si->off) {
7191         case offsetof(struct bpf_xdp_sock, queue_id):
7192                 BPF_XDP_SOCK_GET(queue_id);
7193                 break;
7194         }
7195
7196         return insn - insn_buf;
7197 }
7198
7199 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7200         .func           = bpf_skb_ecn_set_ce,
7201         .gpl_only       = false,
7202         .ret_type       = RET_INTEGER,
7203         .arg1_type      = ARG_PTR_TO_CTX,
7204 };
7205
7206 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7207            struct tcphdr *, th, u32, th_len)
7208 {
7209 #ifdef CONFIG_SYN_COOKIES
7210         u32 cookie;
7211         int ret;
7212
7213         if (unlikely(!sk || th_len < sizeof(*th)))
7214                 return -EINVAL;
7215
7216         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7217         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7218                 return -EINVAL;
7219
7220         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7221                 return -EINVAL;
7222
7223         if (!th->ack || th->rst || th->syn)
7224                 return -ENOENT;
7225
7226         if (unlikely(iph_len < sizeof(struct iphdr)))
7227                 return -EINVAL;
7228
7229         if (tcp_synq_no_recent_overflow(sk))
7230                 return -ENOENT;
7231
7232         cookie = ntohl(th->ack_seq) - 1;
7233
7234         /* Both struct iphdr and struct ipv6hdr have the version field at the
7235          * same offset so we can cast to the shorter header (struct iphdr).
7236          */
7237         switch (((struct iphdr *)iph)->version) {
7238         case 4:
7239                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7240                         return -EINVAL;
7241
7242                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7243                 break;
7244
7245 #if IS_BUILTIN(CONFIG_IPV6)
7246         case 6:
7247                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7248                         return -EINVAL;
7249
7250                 if (sk->sk_family != AF_INET6)
7251                         return -EINVAL;
7252
7253                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7254                 break;
7255 #endif /* CONFIG_IPV6 */
7256
7257         default:
7258                 return -EPROTONOSUPPORT;
7259         }
7260
7261         if (ret > 0)
7262                 return 0;
7263
7264         return -ENOENT;
7265 #else
7266         return -ENOTSUPP;
7267 #endif
7268 }
7269
7270 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7271         .func           = bpf_tcp_check_syncookie,
7272         .gpl_only       = true,
7273         .pkt_access     = true,
7274         .ret_type       = RET_INTEGER,
7275         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7276         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7277         .arg3_type      = ARG_CONST_SIZE,
7278         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7279         .arg5_type      = ARG_CONST_SIZE,
7280 };
7281
7282 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7283            struct tcphdr *, th, u32, th_len)
7284 {
7285 #ifdef CONFIG_SYN_COOKIES
7286         u32 cookie;
7287         u16 mss;
7288
7289         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7290                 return -EINVAL;
7291
7292         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7293                 return -EINVAL;
7294
7295         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7296                 return -ENOENT;
7297
7298         if (!th->syn || th->ack || th->fin || th->rst)
7299                 return -EINVAL;
7300
7301         if (unlikely(iph_len < sizeof(struct iphdr)))
7302                 return -EINVAL;
7303
7304         /* Both struct iphdr and struct ipv6hdr have the version field at the
7305          * same offset so we can cast to the shorter header (struct iphdr).
7306          */
7307         switch (((struct iphdr *)iph)->version) {
7308         case 4:
7309                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7310                         return -EINVAL;
7311
7312                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7313                 break;
7314
7315 #if IS_BUILTIN(CONFIG_IPV6)
7316         case 6:
7317                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7318                         return -EINVAL;
7319
7320                 if (sk->sk_family != AF_INET6)
7321                         return -EINVAL;
7322
7323                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7324                 break;
7325 #endif /* CONFIG_IPV6 */
7326
7327         default:
7328                 return -EPROTONOSUPPORT;
7329         }
7330         if (mss == 0)
7331                 return -ENOENT;
7332
7333         return cookie | ((u64)mss << 32);
7334 #else
7335         return -EOPNOTSUPP;
7336 #endif /* CONFIG_SYN_COOKIES */
7337 }
7338
7339 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7340         .func           = bpf_tcp_gen_syncookie,
7341         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7342         .pkt_access     = true,
7343         .ret_type       = RET_INTEGER,
7344         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7345         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7346         .arg3_type      = ARG_CONST_SIZE,
7347         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7348         .arg5_type      = ARG_CONST_SIZE,
7349 };
7350
7351 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7352 {
7353         if (!sk || flags != 0)
7354                 return -EINVAL;
7355         if (!skb_at_tc_ingress(skb))
7356                 return -EOPNOTSUPP;
7357         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7358                 return -ENETUNREACH;
7359         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7360                 return -ESOCKTNOSUPPORT;
7361         if (sk_is_refcounted(sk) &&
7362             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7363                 return -ENOENT;
7364
7365         skb_orphan(skb);
7366         skb->sk = sk;
7367         skb->destructor = sock_pfree;
7368
7369         return 0;
7370 }
7371
7372 static const struct bpf_func_proto bpf_sk_assign_proto = {
7373         .func           = bpf_sk_assign,
7374         .gpl_only       = false,
7375         .ret_type       = RET_INTEGER,
7376         .arg1_type      = ARG_PTR_TO_CTX,
7377         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7378         .arg3_type      = ARG_ANYTHING,
7379 };
7380
7381 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7382                                     u8 search_kind, const u8 *magic,
7383                                     u8 magic_len, bool *eol)
7384 {
7385         u8 kind, kind_len;
7386
7387         *eol = false;
7388
7389         while (op < opend) {
7390                 kind = op[0];
7391
7392                 if (kind == TCPOPT_EOL) {
7393                         *eol = true;
7394                         return ERR_PTR(-ENOMSG);
7395                 } else if (kind == TCPOPT_NOP) {
7396                         op++;
7397                         continue;
7398                 }
7399
7400                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7401                         /* Something is wrong in the received header.
7402                          * Follow the TCP stack's tcp_parse_options()
7403                          * and just bail here.
7404                          */
7405                         return ERR_PTR(-EFAULT);
7406
7407                 kind_len = op[1];
7408                 if (search_kind == kind) {
7409                         if (!magic_len)
7410                                 return op;
7411
7412                         if (magic_len > kind_len - 2)
7413                                 return ERR_PTR(-ENOMSG);
7414
7415                         if (!memcmp(&op[2], magic, magic_len))
7416                                 return op;
7417                 }
7418
7419                 op += kind_len;
7420         }
7421
7422         return ERR_PTR(-ENOMSG);
7423 }
7424
7425 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7426            void *, search_res, u32, len, u64, flags)
7427 {
7428         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7429         const u8 *op, *opend, *magic, *search = search_res;
7430         u8 search_kind, search_len, copy_len, magic_len;
7431         int ret;
7432
7433         /* 2 byte is the minimal option len except TCPOPT_NOP and
7434          * TCPOPT_EOL which are useless for the bpf prog to learn
7435          * and this helper disallow loading them also.
7436          */
7437         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7438                 return -EINVAL;
7439
7440         search_kind = search[0];
7441         search_len = search[1];
7442
7443         if (search_len > len || search_kind == TCPOPT_NOP ||
7444             search_kind == TCPOPT_EOL)
7445                 return -EINVAL;
7446
7447         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7448                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7449                 if (search_len != 4 && search_len != 6)
7450                         return -EINVAL;
7451                 magic = &search[2];
7452                 magic_len = search_len - 2;
7453         } else {
7454                 if (search_len)
7455                         return -EINVAL;
7456                 magic = NULL;
7457                 magic_len = 0;
7458         }
7459
7460         if (load_syn) {
7461                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7462                 if (ret < 0)
7463                         return ret;
7464
7465                 opend = op + ret;
7466                 op += sizeof(struct tcphdr);
7467         } else {
7468                 if (!bpf_sock->skb ||
7469                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7470                         /* This bpf_sock->op cannot call this helper */
7471                         return -EPERM;
7472
7473                 opend = bpf_sock->skb_data_end;
7474                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7475         }
7476
7477         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7478                                 &eol);
7479         if (IS_ERR(op))
7480                 return PTR_ERR(op);
7481
7482         copy_len = op[1];
7483         ret = copy_len;
7484         if (copy_len > len) {
7485                 ret = -ENOSPC;
7486                 copy_len = len;
7487         }
7488
7489         memcpy(search_res, op, copy_len);
7490         return ret;
7491 }
7492
7493 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7494         .func           = bpf_sock_ops_load_hdr_opt,
7495         .gpl_only       = false,
7496         .ret_type       = RET_INTEGER,
7497         .arg1_type      = ARG_PTR_TO_CTX,
7498         .arg2_type      = ARG_PTR_TO_MEM,
7499         .arg3_type      = ARG_CONST_SIZE,
7500         .arg4_type      = ARG_ANYTHING,
7501 };
7502
7503 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7504            const void *, from, u32, len, u64, flags)
7505 {
7506         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7507         const u8 *op, *new_op, *magic = NULL;
7508         struct sk_buff *skb;
7509         bool eol;
7510
7511         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7512                 return -EPERM;
7513
7514         if (len < 2 || flags)
7515                 return -EINVAL;
7516
7517         new_op = from;
7518         new_kind = new_op[0];
7519         new_kind_len = new_op[1];
7520
7521         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7522             new_kind == TCPOPT_EOL)
7523                 return -EINVAL;
7524
7525         if (new_kind_len > bpf_sock->remaining_opt_len)
7526                 return -ENOSPC;
7527
7528         /* 253 is another experimental kind */
7529         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7530                 if (new_kind_len < 4)
7531                         return -EINVAL;
7532                 /* Match for the 2 byte magic also.
7533                  * RFC 6994: the magic could be 2 or 4 bytes.
7534                  * Hence, matching by 2 byte only is on the
7535                  * conservative side but it is the right
7536                  * thing to do for the 'search-for-duplication'
7537                  * purpose.
7538                  */
7539                 magic = &new_op[2];
7540                 magic_len = 2;
7541         }
7542
7543         /* Check for duplication */
7544         skb = bpf_sock->skb;
7545         op = skb->data + sizeof(struct tcphdr);
7546         opend = bpf_sock->skb_data_end;
7547
7548         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7549                                 &eol);
7550         if (!IS_ERR(op))
7551                 return -EEXIST;
7552
7553         if (PTR_ERR(op) != -ENOMSG)
7554                 return PTR_ERR(op);
7555
7556         if (eol)
7557                 /* The option has been ended.  Treat it as no more
7558                  * header option can be written.
7559                  */
7560                 return -ENOSPC;
7561
7562         /* No duplication found.  Store the header option. */
7563         memcpy(opend, from, new_kind_len);
7564
7565         bpf_sock->remaining_opt_len -= new_kind_len;
7566         bpf_sock->skb_data_end += new_kind_len;
7567
7568         return 0;
7569 }
7570
7571 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7572         .func           = bpf_sock_ops_store_hdr_opt,
7573         .gpl_only       = false,
7574         .ret_type       = RET_INTEGER,
7575         .arg1_type      = ARG_PTR_TO_CTX,
7576         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7577         .arg3_type      = ARG_CONST_SIZE,
7578         .arg4_type      = ARG_ANYTHING,
7579 };
7580
7581 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7582            u32, len, u64, flags)
7583 {
7584         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7585                 return -EPERM;
7586
7587         if (flags || len < 2)
7588                 return -EINVAL;
7589
7590         if (len > bpf_sock->remaining_opt_len)
7591                 return -ENOSPC;
7592
7593         bpf_sock->remaining_opt_len -= len;
7594
7595         return 0;
7596 }
7597
7598 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7599         .func           = bpf_sock_ops_reserve_hdr_opt,
7600         .gpl_only       = false,
7601         .ret_type       = RET_INTEGER,
7602         .arg1_type      = ARG_PTR_TO_CTX,
7603         .arg2_type      = ARG_ANYTHING,
7604         .arg3_type      = ARG_ANYTHING,
7605 };
7606
7607 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7608            u64, tstamp, u32, tstamp_type)
7609 {
7610         /* skb_clear_delivery_time() is done for inet protocol */
7611         if (skb->protocol != htons(ETH_P_IP) &&
7612             skb->protocol != htons(ETH_P_IPV6))
7613                 return -EOPNOTSUPP;
7614
7615         switch (tstamp_type) {
7616         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7617                 if (!tstamp)
7618                         return -EINVAL;
7619                 skb->tstamp = tstamp;
7620                 skb->mono_delivery_time = 1;
7621                 break;
7622         case BPF_SKB_TSTAMP_UNSPEC:
7623                 if (tstamp)
7624                         return -EINVAL;
7625                 skb->tstamp = 0;
7626                 skb->mono_delivery_time = 0;
7627                 break;
7628         default:
7629                 return -EINVAL;
7630         }
7631
7632         return 0;
7633 }
7634
7635 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7636         .func           = bpf_skb_set_tstamp,
7637         .gpl_only       = false,
7638         .ret_type       = RET_INTEGER,
7639         .arg1_type      = ARG_PTR_TO_CTX,
7640         .arg2_type      = ARG_ANYTHING,
7641         .arg3_type      = ARG_ANYTHING,
7642 };
7643
7644 #ifdef CONFIG_SYN_COOKIES
7645 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7646            struct tcphdr *, th, u32, th_len)
7647 {
7648         u32 cookie;
7649         u16 mss;
7650
7651         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7652                 return -EINVAL;
7653
7654         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7655         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7656
7657         return cookie | ((u64)mss << 32);
7658 }
7659
7660 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7661         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7662         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7663         .pkt_access     = true,
7664         .ret_type       = RET_INTEGER,
7665         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7666         .arg1_size      = sizeof(struct iphdr),
7667         .arg2_type      = ARG_PTR_TO_MEM,
7668         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7669 };
7670
7671 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7672            struct tcphdr *, th, u32, th_len)
7673 {
7674 #if IS_BUILTIN(CONFIG_IPV6)
7675         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7676                 sizeof(struct ipv6hdr);
7677         u32 cookie;
7678         u16 mss;
7679
7680         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7681                 return -EINVAL;
7682
7683         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7684         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7685
7686         return cookie | ((u64)mss << 32);
7687 #else
7688         return -EPROTONOSUPPORT;
7689 #endif
7690 }
7691
7692 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7693         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7694         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7695         .pkt_access     = true,
7696         .ret_type       = RET_INTEGER,
7697         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7698         .arg1_size      = sizeof(struct ipv6hdr),
7699         .arg2_type      = ARG_PTR_TO_MEM,
7700         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7701 };
7702
7703 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7704            struct tcphdr *, th)
7705 {
7706         u32 cookie = ntohl(th->ack_seq) - 1;
7707
7708         if (__cookie_v4_check(iph, th, cookie) > 0)
7709                 return 0;
7710
7711         return -EACCES;
7712 }
7713
7714 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7715         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7716         .gpl_only       = true, /* __cookie_v4_check is GPL */
7717         .pkt_access     = true,
7718         .ret_type       = RET_INTEGER,
7719         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7720         .arg1_size      = sizeof(struct iphdr),
7721         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7722         .arg2_size      = sizeof(struct tcphdr),
7723 };
7724
7725 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7726            struct tcphdr *, th)
7727 {
7728 #if IS_BUILTIN(CONFIG_IPV6)
7729         u32 cookie = ntohl(th->ack_seq) - 1;
7730
7731         if (__cookie_v6_check(iph, th, cookie) > 0)
7732                 return 0;
7733
7734         return -EACCES;
7735 #else
7736         return -EPROTONOSUPPORT;
7737 #endif
7738 }
7739
7740 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7741         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7742         .gpl_only       = true, /* __cookie_v6_check is GPL */
7743         .pkt_access     = true,
7744         .ret_type       = RET_INTEGER,
7745         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7746         .arg1_size      = sizeof(struct ipv6hdr),
7747         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7748         .arg2_size      = sizeof(struct tcphdr),
7749 };
7750 #endif /* CONFIG_SYN_COOKIES */
7751
7752 #endif /* CONFIG_INET */
7753
7754 bool bpf_helper_changes_pkt_data(void *func)
7755 {
7756         if (func == bpf_skb_vlan_push ||
7757             func == bpf_skb_vlan_pop ||
7758             func == bpf_skb_store_bytes ||
7759             func == bpf_skb_change_proto ||
7760             func == bpf_skb_change_head ||
7761             func == sk_skb_change_head ||
7762             func == bpf_skb_change_tail ||
7763             func == sk_skb_change_tail ||
7764             func == bpf_skb_adjust_room ||
7765             func == sk_skb_adjust_room ||
7766             func == bpf_skb_pull_data ||
7767             func == sk_skb_pull_data ||
7768             func == bpf_clone_redirect ||
7769             func == bpf_l3_csum_replace ||
7770             func == bpf_l4_csum_replace ||
7771             func == bpf_xdp_adjust_head ||
7772             func == bpf_xdp_adjust_meta ||
7773             func == bpf_msg_pull_data ||
7774             func == bpf_msg_push_data ||
7775             func == bpf_msg_pop_data ||
7776             func == bpf_xdp_adjust_tail ||
7777 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7778             func == bpf_lwt_seg6_store_bytes ||
7779             func == bpf_lwt_seg6_adjust_srh ||
7780             func == bpf_lwt_seg6_action ||
7781 #endif
7782 #ifdef CONFIG_INET
7783             func == bpf_sock_ops_store_hdr_opt ||
7784 #endif
7785             func == bpf_lwt_in_push_encap ||
7786             func == bpf_lwt_xmit_push_encap)
7787                 return true;
7788
7789         return false;
7790 }
7791
7792 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7793 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7794
7795 static const struct bpf_func_proto *
7796 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7797 {
7798         const struct bpf_func_proto *func_proto;
7799
7800         func_proto = cgroup_common_func_proto(func_id, prog);
7801         if (func_proto)
7802                 return func_proto;
7803
7804         func_proto = cgroup_current_func_proto(func_id, prog);
7805         if (func_proto)
7806                 return func_proto;
7807
7808         switch (func_id) {
7809         case BPF_FUNC_get_socket_cookie:
7810                 return &bpf_get_socket_cookie_sock_proto;
7811         case BPF_FUNC_get_netns_cookie:
7812                 return &bpf_get_netns_cookie_sock_proto;
7813         case BPF_FUNC_perf_event_output:
7814                 return &bpf_event_output_data_proto;
7815         case BPF_FUNC_sk_storage_get:
7816                 return &bpf_sk_storage_get_cg_sock_proto;
7817         case BPF_FUNC_ktime_get_coarse_ns:
7818                 return &bpf_ktime_get_coarse_ns_proto;
7819         default:
7820                 return bpf_base_func_proto(func_id);
7821         }
7822 }
7823
7824 static const struct bpf_func_proto *
7825 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7826 {
7827         const struct bpf_func_proto *func_proto;
7828
7829         func_proto = cgroup_common_func_proto(func_id, prog);
7830         if (func_proto)
7831                 return func_proto;
7832
7833         func_proto = cgroup_current_func_proto(func_id, prog);
7834         if (func_proto)
7835                 return func_proto;
7836
7837         switch (func_id) {
7838         case BPF_FUNC_bind:
7839                 switch (prog->expected_attach_type) {
7840                 case BPF_CGROUP_INET4_CONNECT:
7841                 case BPF_CGROUP_INET6_CONNECT:
7842                         return &bpf_bind_proto;
7843                 default:
7844                         return NULL;
7845                 }
7846         case BPF_FUNC_get_socket_cookie:
7847                 return &bpf_get_socket_cookie_sock_addr_proto;
7848         case BPF_FUNC_get_netns_cookie:
7849                 return &bpf_get_netns_cookie_sock_addr_proto;
7850         case BPF_FUNC_perf_event_output:
7851                 return &bpf_event_output_data_proto;
7852 #ifdef CONFIG_INET
7853         case BPF_FUNC_sk_lookup_tcp:
7854                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7855         case BPF_FUNC_sk_lookup_udp:
7856                 return &bpf_sock_addr_sk_lookup_udp_proto;
7857         case BPF_FUNC_sk_release:
7858                 return &bpf_sk_release_proto;
7859         case BPF_FUNC_skc_lookup_tcp:
7860                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7861 #endif /* CONFIG_INET */
7862         case BPF_FUNC_sk_storage_get:
7863                 return &bpf_sk_storage_get_proto;
7864         case BPF_FUNC_sk_storage_delete:
7865                 return &bpf_sk_storage_delete_proto;
7866         case BPF_FUNC_setsockopt:
7867                 switch (prog->expected_attach_type) {
7868                 case BPF_CGROUP_INET4_BIND:
7869                 case BPF_CGROUP_INET6_BIND:
7870                 case BPF_CGROUP_INET4_CONNECT:
7871                 case BPF_CGROUP_INET6_CONNECT:
7872                 case BPF_CGROUP_UDP4_RECVMSG:
7873                 case BPF_CGROUP_UDP6_RECVMSG:
7874                 case BPF_CGROUP_UDP4_SENDMSG:
7875                 case BPF_CGROUP_UDP6_SENDMSG:
7876                 case BPF_CGROUP_INET4_GETPEERNAME:
7877                 case BPF_CGROUP_INET6_GETPEERNAME:
7878                 case BPF_CGROUP_INET4_GETSOCKNAME:
7879                 case BPF_CGROUP_INET6_GETSOCKNAME:
7880                         return &bpf_sock_addr_setsockopt_proto;
7881                 default:
7882                         return NULL;
7883                 }
7884         case BPF_FUNC_getsockopt:
7885                 switch (prog->expected_attach_type) {
7886                 case BPF_CGROUP_INET4_BIND:
7887                 case BPF_CGROUP_INET6_BIND:
7888                 case BPF_CGROUP_INET4_CONNECT:
7889                 case BPF_CGROUP_INET6_CONNECT:
7890                 case BPF_CGROUP_UDP4_RECVMSG:
7891                 case BPF_CGROUP_UDP6_RECVMSG:
7892                 case BPF_CGROUP_UDP4_SENDMSG:
7893                 case BPF_CGROUP_UDP6_SENDMSG:
7894                 case BPF_CGROUP_INET4_GETPEERNAME:
7895                 case BPF_CGROUP_INET6_GETPEERNAME:
7896                 case BPF_CGROUP_INET4_GETSOCKNAME:
7897                 case BPF_CGROUP_INET6_GETSOCKNAME:
7898                         return &bpf_sock_addr_getsockopt_proto;
7899                 default:
7900                         return NULL;
7901                 }
7902         default:
7903                 return bpf_sk_base_func_proto(func_id);
7904         }
7905 }
7906
7907 static const struct bpf_func_proto *
7908 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7909 {
7910         switch (func_id) {
7911         case BPF_FUNC_skb_load_bytes:
7912                 return &bpf_skb_load_bytes_proto;
7913         case BPF_FUNC_skb_load_bytes_relative:
7914                 return &bpf_skb_load_bytes_relative_proto;
7915         case BPF_FUNC_get_socket_cookie:
7916                 return &bpf_get_socket_cookie_proto;
7917         case BPF_FUNC_get_socket_uid:
7918                 return &bpf_get_socket_uid_proto;
7919         case BPF_FUNC_perf_event_output:
7920                 return &bpf_skb_event_output_proto;
7921         default:
7922                 return bpf_sk_base_func_proto(func_id);
7923         }
7924 }
7925
7926 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7927 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7928
7929 static const struct bpf_func_proto *
7930 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7931 {
7932         const struct bpf_func_proto *func_proto;
7933
7934         func_proto = cgroup_common_func_proto(func_id, prog);
7935         if (func_proto)
7936                 return func_proto;
7937
7938         switch (func_id) {
7939         case BPF_FUNC_sk_fullsock:
7940                 return &bpf_sk_fullsock_proto;
7941         case BPF_FUNC_sk_storage_get:
7942                 return &bpf_sk_storage_get_proto;
7943         case BPF_FUNC_sk_storage_delete:
7944                 return &bpf_sk_storage_delete_proto;
7945         case BPF_FUNC_perf_event_output:
7946                 return &bpf_skb_event_output_proto;
7947 #ifdef CONFIG_SOCK_CGROUP_DATA
7948         case BPF_FUNC_skb_cgroup_id:
7949                 return &bpf_skb_cgroup_id_proto;
7950         case BPF_FUNC_skb_ancestor_cgroup_id:
7951                 return &bpf_skb_ancestor_cgroup_id_proto;
7952         case BPF_FUNC_sk_cgroup_id:
7953                 return &bpf_sk_cgroup_id_proto;
7954         case BPF_FUNC_sk_ancestor_cgroup_id:
7955                 return &bpf_sk_ancestor_cgroup_id_proto;
7956 #endif
7957 #ifdef CONFIG_INET
7958         case BPF_FUNC_sk_lookup_tcp:
7959                 return &bpf_sk_lookup_tcp_proto;
7960         case BPF_FUNC_sk_lookup_udp:
7961                 return &bpf_sk_lookup_udp_proto;
7962         case BPF_FUNC_sk_release:
7963                 return &bpf_sk_release_proto;
7964         case BPF_FUNC_skc_lookup_tcp:
7965                 return &bpf_skc_lookup_tcp_proto;
7966         case BPF_FUNC_tcp_sock:
7967                 return &bpf_tcp_sock_proto;
7968         case BPF_FUNC_get_listener_sock:
7969                 return &bpf_get_listener_sock_proto;
7970         case BPF_FUNC_skb_ecn_set_ce:
7971                 return &bpf_skb_ecn_set_ce_proto;
7972 #endif
7973         default:
7974                 return sk_filter_func_proto(func_id, prog);
7975         }
7976 }
7977
7978 static const struct bpf_func_proto *
7979 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7980 {
7981         switch (func_id) {
7982         case BPF_FUNC_skb_store_bytes:
7983                 return &bpf_skb_store_bytes_proto;
7984         case BPF_FUNC_skb_load_bytes:
7985                 return &bpf_skb_load_bytes_proto;
7986         case BPF_FUNC_skb_load_bytes_relative:
7987                 return &bpf_skb_load_bytes_relative_proto;
7988         case BPF_FUNC_skb_pull_data:
7989                 return &bpf_skb_pull_data_proto;
7990         case BPF_FUNC_csum_diff:
7991                 return &bpf_csum_diff_proto;
7992         case BPF_FUNC_csum_update:
7993                 return &bpf_csum_update_proto;
7994         case BPF_FUNC_csum_level:
7995                 return &bpf_csum_level_proto;
7996         case BPF_FUNC_l3_csum_replace:
7997                 return &bpf_l3_csum_replace_proto;
7998         case BPF_FUNC_l4_csum_replace:
7999                 return &bpf_l4_csum_replace_proto;
8000         case BPF_FUNC_clone_redirect:
8001                 return &bpf_clone_redirect_proto;
8002         case BPF_FUNC_get_cgroup_classid:
8003                 return &bpf_get_cgroup_classid_proto;
8004         case BPF_FUNC_skb_vlan_push:
8005                 return &bpf_skb_vlan_push_proto;
8006         case BPF_FUNC_skb_vlan_pop:
8007                 return &bpf_skb_vlan_pop_proto;
8008         case BPF_FUNC_skb_change_proto:
8009                 return &bpf_skb_change_proto_proto;
8010         case BPF_FUNC_skb_change_type:
8011                 return &bpf_skb_change_type_proto;
8012         case BPF_FUNC_skb_adjust_room:
8013                 return &bpf_skb_adjust_room_proto;
8014         case BPF_FUNC_skb_change_tail:
8015                 return &bpf_skb_change_tail_proto;
8016         case BPF_FUNC_skb_change_head:
8017                 return &bpf_skb_change_head_proto;
8018         case BPF_FUNC_skb_get_tunnel_key:
8019                 return &bpf_skb_get_tunnel_key_proto;
8020         case BPF_FUNC_skb_set_tunnel_key:
8021                 return bpf_get_skb_set_tunnel_proto(func_id);
8022         case BPF_FUNC_skb_get_tunnel_opt:
8023                 return &bpf_skb_get_tunnel_opt_proto;
8024         case BPF_FUNC_skb_set_tunnel_opt:
8025                 return bpf_get_skb_set_tunnel_proto(func_id);
8026         case BPF_FUNC_redirect:
8027                 return &bpf_redirect_proto;
8028         case BPF_FUNC_redirect_neigh:
8029                 return &bpf_redirect_neigh_proto;
8030         case BPF_FUNC_redirect_peer:
8031                 return &bpf_redirect_peer_proto;
8032         case BPF_FUNC_get_route_realm:
8033                 return &bpf_get_route_realm_proto;
8034         case BPF_FUNC_get_hash_recalc:
8035                 return &bpf_get_hash_recalc_proto;
8036         case BPF_FUNC_set_hash_invalid:
8037                 return &bpf_set_hash_invalid_proto;
8038         case BPF_FUNC_set_hash:
8039                 return &bpf_set_hash_proto;
8040         case BPF_FUNC_perf_event_output:
8041                 return &bpf_skb_event_output_proto;
8042         case BPF_FUNC_get_smp_processor_id:
8043                 return &bpf_get_smp_processor_id_proto;
8044         case BPF_FUNC_skb_under_cgroup:
8045                 return &bpf_skb_under_cgroup_proto;
8046         case BPF_FUNC_get_socket_cookie:
8047                 return &bpf_get_socket_cookie_proto;
8048         case BPF_FUNC_get_socket_uid:
8049                 return &bpf_get_socket_uid_proto;
8050         case BPF_FUNC_fib_lookup:
8051                 return &bpf_skb_fib_lookup_proto;
8052         case BPF_FUNC_check_mtu:
8053                 return &bpf_skb_check_mtu_proto;
8054         case BPF_FUNC_sk_fullsock:
8055                 return &bpf_sk_fullsock_proto;
8056         case BPF_FUNC_sk_storage_get:
8057                 return &bpf_sk_storage_get_proto;
8058         case BPF_FUNC_sk_storage_delete:
8059                 return &bpf_sk_storage_delete_proto;
8060 #ifdef CONFIG_XFRM
8061         case BPF_FUNC_skb_get_xfrm_state:
8062                 return &bpf_skb_get_xfrm_state_proto;
8063 #endif
8064 #ifdef CONFIG_CGROUP_NET_CLASSID
8065         case BPF_FUNC_skb_cgroup_classid:
8066                 return &bpf_skb_cgroup_classid_proto;
8067 #endif
8068 #ifdef CONFIG_SOCK_CGROUP_DATA
8069         case BPF_FUNC_skb_cgroup_id:
8070                 return &bpf_skb_cgroup_id_proto;
8071         case BPF_FUNC_skb_ancestor_cgroup_id:
8072                 return &bpf_skb_ancestor_cgroup_id_proto;
8073 #endif
8074 #ifdef CONFIG_INET
8075         case BPF_FUNC_sk_lookup_tcp:
8076                 return &bpf_tc_sk_lookup_tcp_proto;
8077         case BPF_FUNC_sk_lookup_udp:
8078                 return &bpf_tc_sk_lookup_udp_proto;
8079         case BPF_FUNC_sk_release:
8080                 return &bpf_sk_release_proto;
8081         case BPF_FUNC_tcp_sock:
8082                 return &bpf_tcp_sock_proto;
8083         case BPF_FUNC_get_listener_sock:
8084                 return &bpf_get_listener_sock_proto;
8085         case BPF_FUNC_skc_lookup_tcp:
8086                 return &bpf_tc_skc_lookup_tcp_proto;
8087         case BPF_FUNC_tcp_check_syncookie:
8088                 return &bpf_tcp_check_syncookie_proto;
8089         case BPF_FUNC_skb_ecn_set_ce:
8090                 return &bpf_skb_ecn_set_ce_proto;
8091         case BPF_FUNC_tcp_gen_syncookie:
8092                 return &bpf_tcp_gen_syncookie_proto;
8093         case BPF_FUNC_sk_assign:
8094                 return &bpf_sk_assign_proto;
8095         case BPF_FUNC_skb_set_tstamp:
8096                 return &bpf_skb_set_tstamp_proto;
8097 #ifdef CONFIG_SYN_COOKIES
8098         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8099                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8100         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8101                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8102         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8103                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8104         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8105                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8106 #endif
8107 #endif
8108         default:
8109                 return bpf_sk_base_func_proto(func_id);
8110         }
8111 }
8112
8113 static const struct bpf_func_proto *
8114 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8115 {
8116         switch (func_id) {
8117         case BPF_FUNC_perf_event_output:
8118                 return &bpf_xdp_event_output_proto;
8119         case BPF_FUNC_get_smp_processor_id:
8120                 return &bpf_get_smp_processor_id_proto;
8121         case BPF_FUNC_csum_diff:
8122                 return &bpf_csum_diff_proto;
8123         case BPF_FUNC_xdp_adjust_head:
8124                 return &bpf_xdp_adjust_head_proto;
8125         case BPF_FUNC_xdp_adjust_meta:
8126                 return &bpf_xdp_adjust_meta_proto;
8127         case BPF_FUNC_redirect:
8128                 return &bpf_xdp_redirect_proto;
8129         case BPF_FUNC_redirect_map:
8130                 return &bpf_xdp_redirect_map_proto;
8131         case BPF_FUNC_xdp_adjust_tail:
8132                 return &bpf_xdp_adjust_tail_proto;
8133         case BPF_FUNC_xdp_get_buff_len:
8134                 return &bpf_xdp_get_buff_len_proto;
8135         case BPF_FUNC_xdp_load_bytes:
8136                 return &bpf_xdp_load_bytes_proto;
8137         case BPF_FUNC_xdp_store_bytes:
8138                 return &bpf_xdp_store_bytes_proto;
8139         case BPF_FUNC_fib_lookup:
8140                 return &bpf_xdp_fib_lookup_proto;
8141         case BPF_FUNC_check_mtu:
8142                 return &bpf_xdp_check_mtu_proto;
8143 #ifdef CONFIG_INET
8144         case BPF_FUNC_sk_lookup_udp:
8145                 return &bpf_xdp_sk_lookup_udp_proto;
8146         case BPF_FUNC_sk_lookup_tcp:
8147                 return &bpf_xdp_sk_lookup_tcp_proto;
8148         case BPF_FUNC_sk_release:
8149                 return &bpf_sk_release_proto;
8150         case BPF_FUNC_skc_lookup_tcp:
8151                 return &bpf_xdp_skc_lookup_tcp_proto;
8152         case BPF_FUNC_tcp_check_syncookie:
8153                 return &bpf_tcp_check_syncookie_proto;
8154         case BPF_FUNC_tcp_gen_syncookie:
8155                 return &bpf_tcp_gen_syncookie_proto;
8156 #ifdef CONFIG_SYN_COOKIES
8157         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8158                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8159         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8160                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8161         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8162                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8163         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8164                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8165 #endif
8166 #endif
8167         default:
8168                 return bpf_sk_base_func_proto(func_id);
8169         }
8170
8171 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8172         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8173          * kfuncs are defined in two different modules, and we want to be able
8174          * to use them interchangably with the same BTF type ID. Because modules
8175          * can't de-duplicate BTF IDs between each other, we need the type to be
8176          * referenced in the vmlinux BTF or the verifier will get confused about
8177          * the different types. So we add this dummy type reference which will
8178          * be included in vmlinux BTF, allowing both modules to refer to the
8179          * same type ID.
8180          */
8181         BTF_TYPE_EMIT(struct nf_conn___init);
8182 #endif
8183 }
8184
8185 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8186 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8187
8188 static const struct bpf_func_proto *
8189 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8190 {
8191         const struct bpf_func_proto *func_proto;
8192
8193         func_proto = cgroup_common_func_proto(func_id, prog);
8194         if (func_proto)
8195                 return func_proto;
8196
8197         switch (func_id) {
8198         case BPF_FUNC_setsockopt:
8199                 return &bpf_sock_ops_setsockopt_proto;
8200         case BPF_FUNC_getsockopt:
8201                 return &bpf_sock_ops_getsockopt_proto;
8202         case BPF_FUNC_sock_ops_cb_flags_set:
8203                 return &bpf_sock_ops_cb_flags_set_proto;
8204         case BPF_FUNC_sock_map_update:
8205                 return &bpf_sock_map_update_proto;
8206         case BPF_FUNC_sock_hash_update:
8207                 return &bpf_sock_hash_update_proto;
8208         case BPF_FUNC_get_socket_cookie:
8209                 return &bpf_get_socket_cookie_sock_ops_proto;
8210         case BPF_FUNC_perf_event_output:
8211                 return &bpf_event_output_data_proto;
8212         case BPF_FUNC_sk_storage_get:
8213                 return &bpf_sk_storage_get_proto;
8214         case BPF_FUNC_sk_storage_delete:
8215                 return &bpf_sk_storage_delete_proto;
8216         case BPF_FUNC_get_netns_cookie:
8217                 return &bpf_get_netns_cookie_sock_ops_proto;
8218 #ifdef CONFIG_INET
8219         case BPF_FUNC_load_hdr_opt:
8220                 return &bpf_sock_ops_load_hdr_opt_proto;
8221         case BPF_FUNC_store_hdr_opt:
8222                 return &bpf_sock_ops_store_hdr_opt_proto;
8223         case BPF_FUNC_reserve_hdr_opt:
8224                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8225         case BPF_FUNC_tcp_sock:
8226                 return &bpf_tcp_sock_proto;
8227 #endif /* CONFIG_INET */
8228         default:
8229                 return bpf_sk_base_func_proto(func_id);
8230         }
8231 }
8232
8233 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8234 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8235
8236 static const struct bpf_func_proto *
8237 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8238 {
8239         switch (func_id) {
8240         case BPF_FUNC_msg_redirect_map:
8241                 return &bpf_msg_redirect_map_proto;
8242         case BPF_FUNC_msg_redirect_hash:
8243                 return &bpf_msg_redirect_hash_proto;
8244         case BPF_FUNC_msg_apply_bytes:
8245                 return &bpf_msg_apply_bytes_proto;
8246         case BPF_FUNC_msg_cork_bytes:
8247                 return &bpf_msg_cork_bytes_proto;
8248         case BPF_FUNC_msg_pull_data:
8249                 return &bpf_msg_pull_data_proto;
8250         case BPF_FUNC_msg_push_data:
8251                 return &bpf_msg_push_data_proto;
8252         case BPF_FUNC_msg_pop_data:
8253                 return &bpf_msg_pop_data_proto;
8254         case BPF_FUNC_perf_event_output:
8255                 return &bpf_event_output_data_proto;
8256         case BPF_FUNC_get_current_uid_gid:
8257                 return &bpf_get_current_uid_gid_proto;
8258         case BPF_FUNC_get_current_pid_tgid:
8259                 return &bpf_get_current_pid_tgid_proto;
8260         case BPF_FUNC_sk_storage_get:
8261                 return &bpf_sk_storage_get_proto;
8262         case BPF_FUNC_sk_storage_delete:
8263                 return &bpf_sk_storage_delete_proto;
8264         case BPF_FUNC_get_netns_cookie:
8265                 return &bpf_get_netns_cookie_sk_msg_proto;
8266 #ifdef CONFIG_CGROUP_NET_CLASSID
8267         case BPF_FUNC_get_cgroup_classid:
8268                 return &bpf_get_cgroup_classid_curr_proto;
8269 #endif
8270         default:
8271                 return bpf_sk_base_func_proto(func_id);
8272         }
8273 }
8274
8275 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8276 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8277
8278 static const struct bpf_func_proto *
8279 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8280 {
8281         switch (func_id) {
8282         case BPF_FUNC_skb_store_bytes:
8283                 return &bpf_skb_store_bytes_proto;
8284         case BPF_FUNC_skb_load_bytes:
8285                 return &bpf_skb_load_bytes_proto;
8286         case BPF_FUNC_skb_pull_data:
8287                 return &sk_skb_pull_data_proto;
8288         case BPF_FUNC_skb_change_tail:
8289                 return &sk_skb_change_tail_proto;
8290         case BPF_FUNC_skb_change_head:
8291                 return &sk_skb_change_head_proto;
8292         case BPF_FUNC_skb_adjust_room:
8293                 return &sk_skb_adjust_room_proto;
8294         case BPF_FUNC_get_socket_cookie:
8295                 return &bpf_get_socket_cookie_proto;
8296         case BPF_FUNC_get_socket_uid:
8297                 return &bpf_get_socket_uid_proto;
8298         case BPF_FUNC_sk_redirect_map:
8299                 return &bpf_sk_redirect_map_proto;
8300         case BPF_FUNC_sk_redirect_hash:
8301                 return &bpf_sk_redirect_hash_proto;
8302         case BPF_FUNC_perf_event_output:
8303                 return &bpf_skb_event_output_proto;
8304 #ifdef CONFIG_INET
8305         case BPF_FUNC_sk_lookup_tcp:
8306                 return &bpf_sk_lookup_tcp_proto;
8307         case BPF_FUNC_sk_lookup_udp:
8308                 return &bpf_sk_lookup_udp_proto;
8309         case BPF_FUNC_sk_release:
8310                 return &bpf_sk_release_proto;
8311         case BPF_FUNC_skc_lookup_tcp:
8312                 return &bpf_skc_lookup_tcp_proto;
8313 #endif
8314         default:
8315                 return bpf_sk_base_func_proto(func_id);
8316         }
8317 }
8318
8319 static const struct bpf_func_proto *
8320 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8321 {
8322         switch (func_id) {
8323         case BPF_FUNC_skb_load_bytes:
8324                 return &bpf_flow_dissector_load_bytes_proto;
8325         default:
8326                 return bpf_sk_base_func_proto(func_id);
8327         }
8328 }
8329
8330 static const struct bpf_func_proto *
8331 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8332 {
8333         switch (func_id) {
8334         case BPF_FUNC_skb_load_bytes:
8335                 return &bpf_skb_load_bytes_proto;
8336         case BPF_FUNC_skb_pull_data:
8337                 return &bpf_skb_pull_data_proto;
8338         case BPF_FUNC_csum_diff:
8339                 return &bpf_csum_diff_proto;
8340         case BPF_FUNC_get_cgroup_classid:
8341                 return &bpf_get_cgroup_classid_proto;
8342         case BPF_FUNC_get_route_realm:
8343                 return &bpf_get_route_realm_proto;
8344         case BPF_FUNC_get_hash_recalc:
8345                 return &bpf_get_hash_recalc_proto;
8346         case BPF_FUNC_perf_event_output:
8347                 return &bpf_skb_event_output_proto;
8348         case BPF_FUNC_get_smp_processor_id:
8349                 return &bpf_get_smp_processor_id_proto;
8350         case BPF_FUNC_skb_under_cgroup:
8351                 return &bpf_skb_under_cgroup_proto;
8352         default:
8353                 return bpf_sk_base_func_proto(func_id);
8354         }
8355 }
8356
8357 static const struct bpf_func_proto *
8358 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8359 {
8360         switch (func_id) {
8361         case BPF_FUNC_lwt_push_encap:
8362                 return &bpf_lwt_in_push_encap_proto;
8363         default:
8364                 return lwt_out_func_proto(func_id, prog);
8365         }
8366 }
8367
8368 static const struct bpf_func_proto *
8369 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8370 {
8371         switch (func_id) {
8372         case BPF_FUNC_skb_get_tunnel_key:
8373                 return &bpf_skb_get_tunnel_key_proto;
8374         case BPF_FUNC_skb_set_tunnel_key:
8375                 return bpf_get_skb_set_tunnel_proto(func_id);
8376         case BPF_FUNC_skb_get_tunnel_opt:
8377                 return &bpf_skb_get_tunnel_opt_proto;
8378         case BPF_FUNC_skb_set_tunnel_opt:
8379                 return bpf_get_skb_set_tunnel_proto(func_id);
8380         case BPF_FUNC_redirect:
8381                 return &bpf_redirect_proto;
8382         case BPF_FUNC_clone_redirect:
8383                 return &bpf_clone_redirect_proto;
8384         case BPF_FUNC_skb_change_tail:
8385                 return &bpf_skb_change_tail_proto;
8386         case BPF_FUNC_skb_change_head:
8387                 return &bpf_skb_change_head_proto;
8388         case BPF_FUNC_skb_store_bytes:
8389                 return &bpf_skb_store_bytes_proto;
8390         case BPF_FUNC_csum_update:
8391                 return &bpf_csum_update_proto;
8392         case BPF_FUNC_csum_level:
8393                 return &bpf_csum_level_proto;
8394         case BPF_FUNC_l3_csum_replace:
8395                 return &bpf_l3_csum_replace_proto;
8396         case BPF_FUNC_l4_csum_replace:
8397                 return &bpf_l4_csum_replace_proto;
8398         case BPF_FUNC_set_hash_invalid:
8399                 return &bpf_set_hash_invalid_proto;
8400         case BPF_FUNC_lwt_push_encap:
8401                 return &bpf_lwt_xmit_push_encap_proto;
8402         default:
8403                 return lwt_out_func_proto(func_id, prog);
8404         }
8405 }
8406
8407 static const struct bpf_func_proto *
8408 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8409 {
8410         switch (func_id) {
8411 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8412         case BPF_FUNC_lwt_seg6_store_bytes:
8413                 return &bpf_lwt_seg6_store_bytes_proto;
8414         case BPF_FUNC_lwt_seg6_action:
8415                 return &bpf_lwt_seg6_action_proto;
8416         case BPF_FUNC_lwt_seg6_adjust_srh:
8417                 return &bpf_lwt_seg6_adjust_srh_proto;
8418 #endif
8419         default:
8420                 return lwt_out_func_proto(func_id, prog);
8421         }
8422 }
8423
8424 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8425                                     const struct bpf_prog *prog,
8426                                     struct bpf_insn_access_aux *info)
8427 {
8428         const int size_default = sizeof(__u32);
8429
8430         if (off < 0 || off >= sizeof(struct __sk_buff))
8431                 return false;
8432
8433         /* The verifier guarantees that size > 0. */
8434         if (off % size != 0)
8435                 return false;
8436
8437         switch (off) {
8438         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8439                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8440                         return false;
8441                 break;
8442         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8443         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8444         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8445         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8446         case bpf_ctx_range(struct __sk_buff, data):
8447         case bpf_ctx_range(struct __sk_buff, data_meta):
8448         case bpf_ctx_range(struct __sk_buff, data_end):
8449                 if (size != size_default)
8450                         return false;
8451                 break;
8452         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8453                 return false;
8454         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8455                 if (type == BPF_WRITE || size != sizeof(__u64))
8456                         return false;
8457                 break;
8458         case bpf_ctx_range(struct __sk_buff, tstamp):
8459                 if (size != sizeof(__u64))
8460                         return false;
8461                 break;
8462         case offsetof(struct __sk_buff, sk):
8463                 if (type == BPF_WRITE || size != sizeof(__u64))
8464                         return false;
8465                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8466                 break;
8467         case offsetof(struct __sk_buff, tstamp_type):
8468                 return false;
8469         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8470                 /* Explicitly prohibit access to padding in __sk_buff. */
8471                 return false;
8472         default:
8473                 /* Only narrow read access allowed for now. */
8474                 if (type == BPF_WRITE) {
8475                         if (size != size_default)
8476                                 return false;
8477                 } else {
8478                         bpf_ctx_record_field_size(info, size_default);
8479                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8480                                 return false;
8481                 }
8482         }
8483
8484         return true;
8485 }
8486
8487 static bool sk_filter_is_valid_access(int off, int size,
8488                                       enum bpf_access_type type,
8489                                       const struct bpf_prog *prog,
8490                                       struct bpf_insn_access_aux *info)
8491 {
8492         switch (off) {
8493         case bpf_ctx_range(struct __sk_buff, tc_classid):
8494         case bpf_ctx_range(struct __sk_buff, data):
8495         case bpf_ctx_range(struct __sk_buff, data_meta):
8496         case bpf_ctx_range(struct __sk_buff, data_end):
8497         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8498         case bpf_ctx_range(struct __sk_buff, tstamp):
8499         case bpf_ctx_range(struct __sk_buff, wire_len):
8500         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8501                 return false;
8502         }
8503
8504         if (type == BPF_WRITE) {
8505                 switch (off) {
8506                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8507                         break;
8508                 default:
8509                         return false;
8510                 }
8511         }
8512
8513         return bpf_skb_is_valid_access(off, size, type, prog, info);
8514 }
8515
8516 static bool cg_skb_is_valid_access(int off, int size,
8517                                    enum bpf_access_type type,
8518                                    const struct bpf_prog *prog,
8519                                    struct bpf_insn_access_aux *info)
8520 {
8521         switch (off) {
8522         case bpf_ctx_range(struct __sk_buff, tc_classid):
8523         case bpf_ctx_range(struct __sk_buff, data_meta):
8524         case bpf_ctx_range(struct __sk_buff, wire_len):
8525                 return false;
8526         case bpf_ctx_range(struct __sk_buff, data):
8527         case bpf_ctx_range(struct __sk_buff, data_end):
8528                 if (!bpf_capable())
8529                         return false;
8530                 break;
8531         }
8532
8533         if (type == BPF_WRITE) {
8534                 switch (off) {
8535                 case bpf_ctx_range(struct __sk_buff, mark):
8536                 case bpf_ctx_range(struct __sk_buff, priority):
8537                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8538                         break;
8539                 case bpf_ctx_range(struct __sk_buff, tstamp):
8540                         if (!bpf_capable())
8541                                 return false;
8542                         break;
8543                 default:
8544                         return false;
8545                 }
8546         }
8547
8548         switch (off) {
8549         case bpf_ctx_range(struct __sk_buff, data):
8550                 info->reg_type = PTR_TO_PACKET;
8551                 break;
8552         case bpf_ctx_range(struct __sk_buff, data_end):
8553                 info->reg_type = PTR_TO_PACKET_END;
8554                 break;
8555         }
8556
8557         return bpf_skb_is_valid_access(off, size, type, prog, info);
8558 }
8559
8560 static bool lwt_is_valid_access(int off, int size,
8561                                 enum bpf_access_type type,
8562                                 const struct bpf_prog *prog,
8563                                 struct bpf_insn_access_aux *info)
8564 {
8565         switch (off) {
8566         case bpf_ctx_range(struct __sk_buff, tc_classid):
8567         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8568         case bpf_ctx_range(struct __sk_buff, data_meta):
8569         case bpf_ctx_range(struct __sk_buff, tstamp):
8570         case bpf_ctx_range(struct __sk_buff, wire_len):
8571         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8572                 return false;
8573         }
8574
8575         if (type == BPF_WRITE) {
8576                 switch (off) {
8577                 case bpf_ctx_range(struct __sk_buff, mark):
8578                 case bpf_ctx_range(struct __sk_buff, priority):
8579                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8580                         break;
8581                 default:
8582                         return false;
8583                 }
8584         }
8585
8586         switch (off) {
8587         case bpf_ctx_range(struct __sk_buff, data):
8588                 info->reg_type = PTR_TO_PACKET;
8589                 break;
8590         case bpf_ctx_range(struct __sk_buff, data_end):
8591                 info->reg_type = PTR_TO_PACKET_END;
8592                 break;
8593         }
8594
8595         return bpf_skb_is_valid_access(off, size, type, prog, info);
8596 }
8597
8598 /* Attach type specific accesses */
8599 static bool __sock_filter_check_attach_type(int off,
8600                                             enum bpf_access_type access_type,
8601                                             enum bpf_attach_type attach_type)
8602 {
8603         switch (off) {
8604         case offsetof(struct bpf_sock, bound_dev_if):
8605         case offsetof(struct bpf_sock, mark):
8606         case offsetof(struct bpf_sock, priority):
8607                 switch (attach_type) {
8608                 case BPF_CGROUP_INET_SOCK_CREATE:
8609                 case BPF_CGROUP_INET_SOCK_RELEASE:
8610                         goto full_access;
8611                 default:
8612                         return false;
8613                 }
8614         case bpf_ctx_range(struct bpf_sock, src_ip4):
8615                 switch (attach_type) {
8616                 case BPF_CGROUP_INET4_POST_BIND:
8617                         goto read_only;
8618                 default:
8619                         return false;
8620                 }
8621         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8622                 switch (attach_type) {
8623                 case BPF_CGROUP_INET6_POST_BIND:
8624                         goto read_only;
8625                 default:
8626                         return false;
8627                 }
8628         case bpf_ctx_range(struct bpf_sock, src_port):
8629                 switch (attach_type) {
8630                 case BPF_CGROUP_INET4_POST_BIND:
8631                 case BPF_CGROUP_INET6_POST_BIND:
8632                         goto read_only;
8633                 default:
8634                         return false;
8635                 }
8636         }
8637 read_only:
8638         return access_type == BPF_READ;
8639 full_access:
8640         return true;
8641 }
8642
8643 bool bpf_sock_common_is_valid_access(int off, int size,
8644                                      enum bpf_access_type type,
8645                                      struct bpf_insn_access_aux *info)
8646 {
8647         switch (off) {
8648         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8649                 return false;
8650         default:
8651                 return bpf_sock_is_valid_access(off, size, type, info);
8652         }
8653 }
8654
8655 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8656                               struct bpf_insn_access_aux *info)
8657 {
8658         const int size_default = sizeof(__u32);
8659         int field_size;
8660
8661         if (off < 0 || off >= sizeof(struct bpf_sock))
8662                 return false;
8663         if (off % size != 0)
8664                 return false;
8665
8666         switch (off) {
8667         case offsetof(struct bpf_sock, state):
8668         case offsetof(struct bpf_sock, family):
8669         case offsetof(struct bpf_sock, type):
8670         case offsetof(struct bpf_sock, protocol):
8671         case offsetof(struct bpf_sock, src_port):
8672         case offsetof(struct bpf_sock, rx_queue_mapping):
8673         case bpf_ctx_range(struct bpf_sock, src_ip4):
8674         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8675         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8676         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8677                 bpf_ctx_record_field_size(info, size_default);
8678                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8679         case bpf_ctx_range(struct bpf_sock, dst_port):
8680                 field_size = size == size_default ?
8681                         size_default : sizeof_field(struct bpf_sock, dst_port);
8682                 bpf_ctx_record_field_size(info, field_size);
8683                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8684         case offsetofend(struct bpf_sock, dst_port) ...
8685              offsetof(struct bpf_sock, dst_ip4) - 1:
8686                 return false;
8687         }
8688
8689         return size == size_default;
8690 }
8691
8692 static bool sock_filter_is_valid_access(int off, int size,
8693                                         enum bpf_access_type type,
8694                                         const struct bpf_prog *prog,
8695                                         struct bpf_insn_access_aux *info)
8696 {
8697         if (!bpf_sock_is_valid_access(off, size, type, info))
8698                 return false;
8699         return __sock_filter_check_attach_type(off, type,
8700                                                prog->expected_attach_type);
8701 }
8702
8703 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8704                              const struct bpf_prog *prog)
8705 {
8706         /* Neither direct read nor direct write requires any preliminary
8707          * action.
8708          */
8709         return 0;
8710 }
8711
8712 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8713                                 const struct bpf_prog *prog, int drop_verdict)
8714 {
8715         struct bpf_insn *insn = insn_buf;
8716
8717         if (!direct_write)
8718                 return 0;
8719
8720         /* if (!skb->cloned)
8721          *       goto start;
8722          *
8723          * (Fast-path, otherwise approximation that we might be
8724          *  a clone, do the rest in helper.)
8725          */
8726         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8727         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8728         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8729
8730         /* ret = bpf_skb_pull_data(skb, 0); */
8731         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8732         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8733         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8734                                BPF_FUNC_skb_pull_data);
8735         /* if (!ret)
8736          *      goto restore;
8737          * return TC_ACT_SHOT;
8738          */
8739         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8740         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8741         *insn++ = BPF_EXIT_INSN();
8742
8743         /* restore: */
8744         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8745         /* start: */
8746         *insn++ = prog->insnsi[0];
8747
8748         return insn - insn_buf;
8749 }
8750
8751 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8752                           struct bpf_insn *insn_buf)
8753 {
8754         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8755         struct bpf_insn *insn = insn_buf;
8756
8757         if (!indirect) {
8758                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8759         } else {
8760                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8761                 if (orig->imm)
8762                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8763         }
8764         /* We're guaranteed here that CTX is in R6. */
8765         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8766
8767         switch (BPF_SIZE(orig->code)) {
8768         case BPF_B:
8769                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8770                 break;
8771         case BPF_H:
8772                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8773                 break;
8774         case BPF_W:
8775                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8776                 break;
8777         }
8778
8779         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8780         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8781         *insn++ = BPF_EXIT_INSN();
8782
8783         return insn - insn_buf;
8784 }
8785
8786 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8787                                const struct bpf_prog *prog)
8788 {
8789         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8790 }
8791
8792 static bool tc_cls_act_is_valid_access(int off, int size,
8793                                        enum bpf_access_type type,
8794                                        const struct bpf_prog *prog,
8795                                        struct bpf_insn_access_aux *info)
8796 {
8797         if (type == BPF_WRITE) {
8798                 switch (off) {
8799                 case bpf_ctx_range(struct __sk_buff, mark):
8800                 case bpf_ctx_range(struct __sk_buff, tc_index):
8801                 case bpf_ctx_range(struct __sk_buff, priority):
8802                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8803                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8804                 case bpf_ctx_range(struct __sk_buff, tstamp):
8805                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8806                         break;
8807                 default:
8808                         return false;
8809                 }
8810         }
8811
8812         switch (off) {
8813         case bpf_ctx_range(struct __sk_buff, data):
8814                 info->reg_type = PTR_TO_PACKET;
8815                 break;
8816         case bpf_ctx_range(struct __sk_buff, data_meta):
8817                 info->reg_type = PTR_TO_PACKET_META;
8818                 break;
8819         case bpf_ctx_range(struct __sk_buff, data_end):
8820                 info->reg_type = PTR_TO_PACKET_END;
8821                 break;
8822         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8823                 return false;
8824         case offsetof(struct __sk_buff, tstamp_type):
8825                 /* The convert_ctx_access() on reading and writing
8826                  * __sk_buff->tstamp depends on whether the bpf prog
8827                  * has used __sk_buff->tstamp_type or not.
8828                  * Thus, we need to set prog->tstamp_type_access
8829                  * earlier during is_valid_access() here.
8830                  */
8831                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8832                 return size == sizeof(__u8);
8833         }
8834
8835         return bpf_skb_is_valid_access(off, size, type, prog, info);
8836 }
8837
8838 DEFINE_MUTEX(nf_conn_btf_access_lock);
8839 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8840
8841 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8842                               const struct bpf_reg_state *reg,
8843                               int off, int size);
8844 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8845
8846 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8847                                         const struct bpf_reg_state *reg,
8848                                         int off, int size)
8849 {
8850         int ret = -EACCES;
8851
8852         mutex_lock(&nf_conn_btf_access_lock);
8853         if (nfct_btf_struct_access)
8854                 ret = nfct_btf_struct_access(log, reg, off, size);
8855         mutex_unlock(&nf_conn_btf_access_lock);
8856
8857         return ret;
8858 }
8859
8860 static bool __is_valid_xdp_access(int off, int size)
8861 {
8862         if (off < 0 || off >= sizeof(struct xdp_md))
8863                 return false;
8864         if (off % size != 0)
8865                 return false;
8866         if (size != sizeof(__u32))
8867                 return false;
8868
8869         return true;
8870 }
8871
8872 static bool xdp_is_valid_access(int off, int size,
8873                                 enum bpf_access_type type,
8874                                 const struct bpf_prog *prog,
8875                                 struct bpf_insn_access_aux *info)
8876 {
8877         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8878                 switch (off) {
8879                 case offsetof(struct xdp_md, egress_ifindex):
8880                         return false;
8881                 }
8882         }
8883
8884         if (type == BPF_WRITE) {
8885                 if (bpf_prog_is_offloaded(prog->aux)) {
8886                         switch (off) {
8887                         case offsetof(struct xdp_md, rx_queue_index):
8888                                 return __is_valid_xdp_access(off, size);
8889                         }
8890                 }
8891                 return false;
8892         }
8893
8894         switch (off) {
8895         case offsetof(struct xdp_md, data):
8896                 info->reg_type = PTR_TO_PACKET;
8897                 break;
8898         case offsetof(struct xdp_md, data_meta):
8899                 info->reg_type = PTR_TO_PACKET_META;
8900                 break;
8901         case offsetof(struct xdp_md, data_end):
8902                 info->reg_type = PTR_TO_PACKET_END;
8903                 break;
8904         }
8905
8906         return __is_valid_xdp_access(off, size);
8907 }
8908
8909 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8910 {
8911         const u32 act_max = XDP_REDIRECT;
8912
8913         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8914                      act > act_max ? "Illegal" : "Driver unsupported",
8915                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8916 }
8917 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8918
8919 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8920                                  const struct bpf_reg_state *reg,
8921                                  int off, int size)
8922 {
8923         int ret = -EACCES;
8924
8925         mutex_lock(&nf_conn_btf_access_lock);
8926         if (nfct_btf_struct_access)
8927                 ret = nfct_btf_struct_access(log, reg, off, size);
8928         mutex_unlock(&nf_conn_btf_access_lock);
8929
8930         return ret;
8931 }
8932
8933 static bool sock_addr_is_valid_access(int off, int size,
8934                                       enum bpf_access_type type,
8935                                       const struct bpf_prog *prog,
8936                                       struct bpf_insn_access_aux *info)
8937 {
8938         const int size_default = sizeof(__u32);
8939
8940         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8941                 return false;
8942         if (off % size != 0)
8943                 return false;
8944
8945         /* Disallow access to IPv6 fields from IPv4 contex and vise
8946          * versa.
8947          */
8948         switch (off) {
8949         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8950                 switch (prog->expected_attach_type) {
8951                 case BPF_CGROUP_INET4_BIND:
8952                 case BPF_CGROUP_INET4_CONNECT:
8953                 case BPF_CGROUP_INET4_GETPEERNAME:
8954                 case BPF_CGROUP_INET4_GETSOCKNAME:
8955                 case BPF_CGROUP_UDP4_SENDMSG:
8956                 case BPF_CGROUP_UDP4_RECVMSG:
8957                         break;
8958                 default:
8959                         return false;
8960                 }
8961                 break;
8962         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8963                 switch (prog->expected_attach_type) {
8964                 case BPF_CGROUP_INET6_BIND:
8965                 case BPF_CGROUP_INET6_CONNECT:
8966                 case BPF_CGROUP_INET6_GETPEERNAME:
8967                 case BPF_CGROUP_INET6_GETSOCKNAME:
8968                 case BPF_CGROUP_UDP6_SENDMSG:
8969                 case BPF_CGROUP_UDP6_RECVMSG:
8970                         break;
8971                 default:
8972                         return false;
8973                 }
8974                 break;
8975         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8976                 switch (prog->expected_attach_type) {
8977                 case BPF_CGROUP_UDP4_SENDMSG:
8978                         break;
8979                 default:
8980                         return false;
8981                 }
8982                 break;
8983         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8984                                 msg_src_ip6[3]):
8985                 switch (prog->expected_attach_type) {
8986                 case BPF_CGROUP_UDP6_SENDMSG:
8987                         break;
8988                 default:
8989                         return false;
8990                 }
8991                 break;
8992         }
8993
8994         switch (off) {
8995         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8996         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8997         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8998         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8999                                 msg_src_ip6[3]):
9000         case bpf_ctx_range(struct bpf_sock_addr, user_port):
9001                 if (type == BPF_READ) {
9002                         bpf_ctx_record_field_size(info, size_default);
9003
9004                         if (bpf_ctx_wide_access_ok(off, size,
9005                                                    struct bpf_sock_addr,
9006                                                    user_ip6))
9007                                 return true;
9008
9009                         if (bpf_ctx_wide_access_ok(off, size,
9010                                                    struct bpf_sock_addr,
9011                                                    msg_src_ip6))
9012                                 return true;
9013
9014                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
9015                                 return false;
9016                 } else {
9017                         if (bpf_ctx_wide_access_ok(off, size,
9018                                                    struct bpf_sock_addr,
9019                                                    user_ip6))
9020                                 return true;
9021
9022                         if (bpf_ctx_wide_access_ok(off, size,
9023                                                    struct bpf_sock_addr,
9024                                                    msg_src_ip6))
9025                                 return true;
9026
9027                         if (size != size_default)
9028                                 return false;
9029                 }
9030                 break;
9031         case offsetof(struct bpf_sock_addr, sk):
9032                 if (type != BPF_READ)
9033                         return false;
9034                 if (size != sizeof(__u64))
9035                         return false;
9036                 info->reg_type = PTR_TO_SOCKET;
9037                 break;
9038         default:
9039                 if (type == BPF_READ) {
9040                         if (size != size_default)
9041                                 return false;
9042                 } else {
9043                         return false;
9044                 }
9045         }
9046
9047         return true;
9048 }
9049
9050 static bool sock_ops_is_valid_access(int off, int size,
9051                                      enum bpf_access_type type,
9052                                      const struct bpf_prog *prog,
9053                                      struct bpf_insn_access_aux *info)
9054 {
9055         const int size_default = sizeof(__u32);
9056
9057         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
9058                 return false;
9059
9060         /* The verifier guarantees that size > 0. */
9061         if (off % size != 0)
9062                 return false;
9063
9064         if (type == BPF_WRITE) {
9065                 switch (off) {
9066                 case offsetof(struct bpf_sock_ops, reply):
9067                 case offsetof(struct bpf_sock_ops, sk_txhash):
9068                         if (size != size_default)
9069                                 return false;
9070                         break;
9071                 default:
9072                         return false;
9073                 }
9074         } else {
9075                 switch (off) {
9076                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
9077                                         bytes_acked):
9078                         if (size != sizeof(__u64))
9079                                 return false;
9080                         break;
9081                 case offsetof(struct bpf_sock_ops, sk):
9082                         if (size != sizeof(__u64))
9083                                 return false;
9084                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
9085                         break;
9086                 case offsetof(struct bpf_sock_ops, skb_data):
9087                         if (size != sizeof(__u64))
9088                                 return false;
9089                         info->reg_type = PTR_TO_PACKET;
9090                         break;
9091                 case offsetof(struct bpf_sock_ops, skb_data_end):
9092                         if (size != sizeof(__u64))
9093                                 return false;
9094                         info->reg_type = PTR_TO_PACKET_END;
9095                         break;
9096                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9097                         bpf_ctx_record_field_size(info, size_default);
9098                         return bpf_ctx_narrow_access_ok(off, size,
9099                                                         size_default);
9100                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
9101                         if (size != sizeof(__u64))
9102                                 return false;
9103                         break;
9104                 default:
9105                         if (size != size_default)
9106                                 return false;
9107                         break;
9108                 }
9109         }
9110
9111         return true;
9112 }
9113
9114 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
9115                            const struct bpf_prog *prog)
9116 {
9117         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9118 }
9119
9120 static bool sk_skb_is_valid_access(int off, int size,
9121                                    enum bpf_access_type type,
9122                                    const struct bpf_prog *prog,
9123                                    struct bpf_insn_access_aux *info)
9124 {
9125         switch (off) {
9126         case bpf_ctx_range(struct __sk_buff, tc_classid):
9127         case bpf_ctx_range(struct __sk_buff, data_meta):
9128         case bpf_ctx_range(struct __sk_buff, tstamp):
9129         case bpf_ctx_range(struct __sk_buff, wire_len):
9130         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9131                 return false;
9132         }
9133
9134         if (type == BPF_WRITE) {
9135                 switch (off) {
9136                 case bpf_ctx_range(struct __sk_buff, tc_index):
9137                 case bpf_ctx_range(struct __sk_buff, priority):
9138                         break;
9139                 default:
9140                         return false;
9141                 }
9142         }
9143
9144         switch (off) {
9145         case bpf_ctx_range(struct __sk_buff, mark):
9146                 return false;
9147         case bpf_ctx_range(struct __sk_buff, data):
9148                 info->reg_type = PTR_TO_PACKET;
9149                 break;
9150         case bpf_ctx_range(struct __sk_buff, data_end):
9151                 info->reg_type = PTR_TO_PACKET_END;
9152                 break;
9153         }
9154
9155         return bpf_skb_is_valid_access(off, size, type, prog, info);
9156 }
9157
9158 static bool sk_msg_is_valid_access(int off, int size,
9159                                    enum bpf_access_type type,
9160                                    const struct bpf_prog *prog,
9161                                    struct bpf_insn_access_aux *info)
9162 {
9163         if (type == BPF_WRITE)
9164                 return false;
9165
9166         if (off % size != 0)
9167                 return false;
9168
9169         switch (off) {
9170         case offsetof(struct sk_msg_md, data):
9171                 info->reg_type = PTR_TO_PACKET;
9172                 if (size != sizeof(__u64))
9173                         return false;
9174                 break;
9175         case offsetof(struct sk_msg_md, data_end):
9176                 info->reg_type = PTR_TO_PACKET_END;
9177                 if (size != sizeof(__u64))
9178                         return false;
9179                 break;
9180         case offsetof(struct sk_msg_md, sk):
9181                 if (size != sizeof(__u64))
9182                         return false;
9183                 info->reg_type = PTR_TO_SOCKET;
9184                 break;
9185         case bpf_ctx_range(struct sk_msg_md, family):
9186         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9187         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9188         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9189         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9190         case bpf_ctx_range(struct sk_msg_md, remote_port):
9191         case bpf_ctx_range(struct sk_msg_md, local_port):
9192         case bpf_ctx_range(struct sk_msg_md, size):
9193                 if (size != sizeof(__u32))
9194                         return false;
9195                 break;
9196         default:
9197                 return false;
9198         }
9199         return true;
9200 }
9201
9202 static bool flow_dissector_is_valid_access(int off, int size,
9203                                            enum bpf_access_type type,
9204                                            const struct bpf_prog *prog,
9205                                            struct bpf_insn_access_aux *info)
9206 {
9207         const int size_default = sizeof(__u32);
9208
9209         if (off < 0 || off >= sizeof(struct __sk_buff))
9210                 return false;
9211
9212         if (type == BPF_WRITE)
9213                 return false;
9214
9215         switch (off) {
9216         case bpf_ctx_range(struct __sk_buff, data):
9217                 if (size != size_default)
9218                         return false;
9219                 info->reg_type = PTR_TO_PACKET;
9220                 return true;
9221         case bpf_ctx_range(struct __sk_buff, data_end):
9222                 if (size != size_default)
9223                         return false;
9224                 info->reg_type = PTR_TO_PACKET_END;
9225                 return true;
9226         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9227                 if (size != sizeof(__u64))
9228                         return false;
9229                 info->reg_type = PTR_TO_FLOW_KEYS;
9230                 return true;
9231         default:
9232                 return false;
9233         }
9234 }
9235
9236 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9237                                              const struct bpf_insn *si,
9238                                              struct bpf_insn *insn_buf,
9239                                              struct bpf_prog *prog,
9240                                              u32 *target_size)
9241
9242 {
9243         struct bpf_insn *insn = insn_buf;
9244
9245         switch (si->off) {
9246         case offsetof(struct __sk_buff, data):
9247                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9248                                       si->dst_reg, si->src_reg,
9249                                       offsetof(struct bpf_flow_dissector, data));
9250                 break;
9251
9252         case offsetof(struct __sk_buff, data_end):
9253                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9254                                       si->dst_reg, si->src_reg,
9255                                       offsetof(struct bpf_flow_dissector, data_end));
9256                 break;
9257
9258         case offsetof(struct __sk_buff, flow_keys):
9259                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9260                                       si->dst_reg, si->src_reg,
9261                                       offsetof(struct bpf_flow_dissector, flow_keys));
9262                 break;
9263         }
9264
9265         return insn - insn_buf;
9266 }
9267
9268 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9269                                                      struct bpf_insn *insn)
9270 {
9271         __u8 value_reg = si->dst_reg;
9272         __u8 skb_reg = si->src_reg;
9273         /* AX is needed because src_reg and dst_reg could be the same */
9274         __u8 tmp_reg = BPF_REG_AX;
9275
9276         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9277                               SKB_BF_MONO_TC_OFFSET);
9278         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9279                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9280         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9281         *insn++ = BPF_JMP_A(1);
9282         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9283
9284         return insn;
9285 }
9286
9287 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9288                                                   struct bpf_insn *insn)
9289 {
9290         /* si->dst_reg = skb_shinfo(SKB); */
9291 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9292         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9293                               BPF_REG_AX, skb_reg,
9294                               offsetof(struct sk_buff, end));
9295         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9296                               dst_reg, skb_reg,
9297                               offsetof(struct sk_buff, head));
9298         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9299 #else
9300         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9301                               dst_reg, skb_reg,
9302                               offsetof(struct sk_buff, end));
9303 #endif
9304
9305         return insn;
9306 }
9307
9308 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9309                                                 const struct bpf_insn *si,
9310                                                 struct bpf_insn *insn)
9311 {
9312         __u8 value_reg = si->dst_reg;
9313         __u8 skb_reg = si->src_reg;
9314
9315 #ifdef CONFIG_NET_CLS_ACT
9316         /* If the tstamp_type is read,
9317          * the bpf prog is aware the tstamp could have delivery time.
9318          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9319          */
9320         if (!prog->tstamp_type_access) {
9321                 /* AX is needed because src_reg and dst_reg could be the same */
9322                 __u8 tmp_reg = BPF_REG_AX;
9323
9324                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9325                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9326                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9327                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9328                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9329                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9330                  * read 0 as the (rcv) timestamp.
9331                  */
9332                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9333                 *insn++ = BPF_JMP_A(1);
9334         }
9335 #endif
9336
9337         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9338                               offsetof(struct sk_buff, tstamp));
9339         return insn;
9340 }
9341
9342 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9343                                                  const struct bpf_insn *si,
9344                                                  struct bpf_insn *insn)
9345 {
9346         __u8 value_reg = si->src_reg;
9347         __u8 skb_reg = si->dst_reg;
9348
9349 #ifdef CONFIG_NET_CLS_ACT
9350         /* If the tstamp_type is read,
9351          * the bpf prog is aware the tstamp could have delivery time.
9352          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9353          * Otherwise, writing at ingress will have to clear the
9354          * mono_delivery_time bit also.
9355          */
9356         if (!prog->tstamp_type_access) {
9357                 __u8 tmp_reg = BPF_REG_AX;
9358
9359                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET);
9360                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9361                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9362                 /* goto <store> */
9363                 *insn++ = BPF_JMP_A(2);
9364                 /* <clear>: mono_delivery_time */
9365                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9366                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET);
9367         }
9368 #endif
9369
9370         /* <store>: skb->tstamp = tstamp */
9371         *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM,
9372                                skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm);
9373         return insn;
9374 }
9375
9376 #define BPF_EMIT_STORE(size, si, off)                                   \
9377         BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM,          \
9378                      (si)->dst_reg, (si)->src_reg, (off), (si)->imm)
9379
9380 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9381                                   const struct bpf_insn *si,
9382                                   struct bpf_insn *insn_buf,
9383                                   struct bpf_prog *prog, u32 *target_size)
9384 {
9385         struct bpf_insn *insn = insn_buf;
9386         int off;
9387
9388         switch (si->off) {
9389         case offsetof(struct __sk_buff, len):
9390                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9391                                       bpf_target_off(struct sk_buff, len, 4,
9392                                                      target_size));
9393                 break;
9394
9395         case offsetof(struct __sk_buff, protocol):
9396                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9397                                       bpf_target_off(struct sk_buff, protocol, 2,
9398                                                      target_size));
9399                 break;
9400
9401         case offsetof(struct __sk_buff, vlan_proto):
9402                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9403                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9404                                                      target_size));
9405                 break;
9406
9407         case offsetof(struct __sk_buff, priority):
9408                 if (type == BPF_WRITE)
9409                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9410                                                  bpf_target_off(struct sk_buff, priority, 4,
9411                                                                 target_size));
9412                 else
9413                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9414                                               bpf_target_off(struct sk_buff, priority, 4,
9415                                                              target_size));
9416                 break;
9417
9418         case offsetof(struct __sk_buff, ingress_ifindex):
9419                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9420                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9421                                                      target_size));
9422                 break;
9423
9424         case offsetof(struct __sk_buff, ifindex):
9425                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9426                                       si->dst_reg, si->src_reg,
9427                                       offsetof(struct sk_buff, dev));
9428                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9429                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9430                                       bpf_target_off(struct net_device, ifindex, 4,
9431                                                      target_size));
9432                 break;
9433
9434         case offsetof(struct __sk_buff, hash):
9435                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9436                                       bpf_target_off(struct sk_buff, hash, 4,
9437                                                      target_size));
9438                 break;
9439
9440         case offsetof(struct __sk_buff, mark):
9441                 if (type == BPF_WRITE)
9442                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9443                                                  bpf_target_off(struct sk_buff, mark, 4,
9444                                                                 target_size));
9445                 else
9446                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9447                                               bpf_target_off(struct sk_buff, mark, 4,
9448                                                              target_size));
9449                 break;
9450
9451         case offsetof(struct __sk_buff, pkt_type):
9452                 *target_size = 1;
9453                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9454                                       PKT_TYPE_OFFSET);
9455                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9456 #ifdef __BIG_ENDIAN_BITFIELD
9457                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9458 #endif
9459                 break;
9460
9461         case offsetof(struct __sk_buff, queue_mapping):
9462                 if (type == BPF_WRITE) {
9463                         u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size);
9464
9465                         if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) {
9466                                 *insn++ = BPF_JMP_A(0); /* noop */
9467                                 break;
9468                         }
9469
9470                         if (BPF_CLASS(si->code) == BPF_STX)
9471                                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9472                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9473                 } else {
9474                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9475                                               bpf_target_off(struct sk_buff,
9476                                                              queue_mapping,
9477                                                              2, target_size));
9478                 }
9479                 break;
9480
9481         case offsetof(struct __sk_buff, vlan_present):
9482                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9483                                       bpf_target_off(struct sk_buff,
9484                                                      vlan_all, 4, target_size));
9485                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9486                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9487                 break;
9488
9489         case offsetof(struct __sk_buff, vlan_tci):
9490                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9491                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9492                                                      target_size));
9493                 break;
9494
9495         case offsetof(struct __sk_buff, cb[0]) ...
9496              offsetofend(struct __sk_buff, cb[4]) - 1:
9497                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9498                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9499                               offsetof(struct qdisc_skb_cb, data)) %
9500                              sizeof(__u64));
9501
9502                 prog->cb_access = 1;
9503                 off  = si->off;
9504                 off -= offsetof(struct __sk_buff, cb[0]);
9505                 off += offsetof(struct sk_buff, cb);
9506                 off += offsetof(struct qdisc_skb_cb, data);
9507                 if (type == BPF_WRITE)
9508                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
9509                 else
9510                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9511                                               si->src_reg, off);
9512                 break;
9513
9514         case offsetof(struct __sk_buff, tc_classid):
9515                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9516
9517                 off  = si->off;
9518                 off -= offsetof(struct __sk_buff, tc_classid);
9519                 off += offsetof(struct sk_buff, cb);
9520                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9521                 *target_size = 2;
9522                 if (type == BPF_WRITE)
9523                         *insn++ = BPF_EMIT_STORE(BPF_H, si, off);
9524                 else
9525                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9526                                               si->src_reg, off);
9527                 break;
9528
9529         case offsetof(struct __sk_buff, data):
9530                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9531                                       si->dst_reg, si->src_reg,
9532                                       offsetof(struct sk_buff, data));
9533                 break;
9534
9535         case offsetof(struct __sk_buff, data_meta):
9536                 off  = si->off;
9537                 off -= offsetof(struct __sk_buff, data_meta);
9538                 off += offsetof(struct sk_buff, cb);
9539                 off += offsetof(struct bpf_skb_data_end, data_meta);
9540                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9541                                       si->src_reg, off);
9542                 break;
9543
9544         case offsetof(struct __sk_buff, data_end):
9545                 off  = si->off;
9546                 off -= offsetof(struct __sk_buff, data_end);
9547                 off += offsetof(struct sk_buff, cb);
9548                 off += offsetof(struct bpf_skb_data_end, data_end);
9549                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9550                                       si->src_reg, off);
9551                 break;
9552
9553         case offsetof(struct __sk_buff, tc_index):
9554 #ifdef CONFIG_NET_SCHED
9555                 if (type == BPF_WRITE)
9556                         *insn++ = BPF_EMIT_STORE(BPF_H, si,
9557                                                  bpf_target_off(struct sk_buff, tc_index, 2,
9558                                                                 target_size));
9559                 else
9560                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9561                                               bpf_target_off(struct sk_buff, tc_index, 2,
9562                                                              target_size));
9563 #else
9564                 *target_size = 2;
9565                 if (type == BPF_WRITE)
9566                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9567                 else
9568                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9569 #endif
9570                 break;
9571
9572         case offsetof(struct __sk_buff, napi_id):
9573 #if defined(CONFIG_NET_RX_BUSY_POLL)
9574                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9575                                       bpf_target_off(struct sk_buff, napi_id, 4,
9576                                                      target_size));
9577                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9578                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9579 #else
9580                 *target_size = 4;
9581                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9582 #endif
9583                 break;
9584         case offsetof(struct __sk_buff, family):
9585                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9586
9587                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9588                                       si->dst_reg, si->src_reg,
9589                                       offsetof(struct sk_buff, sk));
9590                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9591                                       bpf_target_off(struct sock_common,
9592                                                      skc_family,
9593                                                      2, target_size));
9594                 break;
9595         case offsetof(struct __sk_buff, remote_ip4):
9596                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9597
9598                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9599                                       si->dst_reg, si->src_reg,
9600                                       offsetof(struct sk_buff, sk));
9601                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9602                                       bpf_target_off(struct sock_common,
9603                                                      skc_daddr,
9604                                                      4, target_size));
9605                 break;
9606         case offsetof(struct __sk_buff, local_ip4):
9607                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9608                                           skc_rcv_saddr) != 4);
9609
9610                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9611                                       si->dst_reg, si->src_reg,
9612                                       offsetof(struct sk_buff, sk));
9613                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9614                                       bpf_target_off(struct sock_common,
9615                                                      skc_rcv_saddr,
9616                                                      4, target_size));
9617                 break;
9618         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9619              offsetof(struct __sk_buff, remote_ip6[3]):
9620 #if IS_ENABLED(CONFIG_IPV6)
9621                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9622                                           skc_v6_daddr.s6_addr32[0]) != 4);
9623
9624                 off = si->off;
9625                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9626
9627                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9628                                       si->dst_reg, si->src_reg,
9629                                       offsetof(struct sk_buff, sk));
9630                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9631                                       offsetof(struct sock_common,
9632                                                skc_v6_daddr.s6_addr32[0]) +
9633                                       off);
9634 #else
9635                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9636 #endif
9637                 break;
9638         case offsetof(struct __sk_buff, local_ip6[0]) ...
9639              offsetof(struct __sk_buff, local_ip6[3]):
9640 #if IS_ENABLED(CONFIG_IPV6)
9641                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9642                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9643
9644                 off = si->off;
9645                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9646
9647                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9648                                       si->dst_reg, si->src_reg,
9649                                       offsetof(struct sk_buff, sk));
9650                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9651                                       offsetof(struct sock_common,
9652                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9653                                       off);
9654 #else
9655                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9656 #endif
9657                 break;
9658
9659         case offsetof(struct __sk_buff, remote_port):
9660                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9661
9662                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9663                                       si->dst_reg, si->src_reg,
9664                                       offsetof(struct sk_buff, sk));
9665                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9666                                       bpf_target_off(struct sock_common,
9667                                                      skc_dport,
9668                                                      2, target_size));
9669 #ifndef __BIG_ENDIAN_BITFIELD
9670                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9671 #endif
9672                 break;
9673
9674         case offsetof(struct __sk_buff, local_port):
9675                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9676
9677                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9678                                       si->dst_reg, si->src_reg,
9679                                       offsetof(struct sk_buff, sk));
9680                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9681                                       bpf_target_off(struct sock_common,
9682                                                      skc_num, 2, target_size));
9683                 break;
9684
9685         case offsetof(struct __sk_buff, tstamp):
9686                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9687
9688                 if (type == BPF_WRITE)
9689                         insn = bpf_convert_tstamp_write(prog, si, insn);
9690                 else
9691                         insn = bpf_convert_tstamp_read(prog, si, insn);
9692                 break;
9693
9694         case offsetof(struct __sk_buff, tstamp_type):
9695                 insn = bpf_convert_tstamp_type_read(si, insn);
9696                 break;
9697
9698         case offsetof(struct __sk_buff, gso_segs):
9699                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9700                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9701                                       si->dst_reg, si->dst_reg,
9702                                       bpf_target_off(struct skb_shared_info,
9703                                                      gso_segs, 2,
9704                                                      target_size));
9705                 break;
9706         case offsetof(struct __sk_buff, gso_size):
9707                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9708                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9709                                       si->dst_reg, si->dst_reg,
9710                                       bpf_target_off(struct skb_shared_info,
9711                                                      gso_size, 2,
9712                                                      target_size));
9713                 break;
9714         case offsetof(struct __sk_buff, wire_len):
9715                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9716
9717                 off = si->off;
9718                 off -= offsetof(struct __sk_buff, wire_len);
9719                 off += offsetof(struct sk_buff, cb);
9720                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9721                 *target_size = 4;
9722                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9723                 break;
9724
9725         case offsetof(struct __sk_buff, sk):
9726                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9727                                       si->dst_reg, si->src_reg,
9728                                       offsetof(struct sk_buff, sk));
9729                 break;
9730         case offsetof(struct __sk_buff, hwtstamp):
9731                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9732                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9733
9734                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9735                 *insn++ = BPF_LDX_MEM(BPF_DW,
9736                                       si->dst_reg, si->dst_reg,
9737                                       bpf_target_off(struct skb_shared_info,
9738                                                      hwtstamps, 8,
9739                                                      target_size));
9740                 break;
9741         }
9742
9743         return insn - insn_buf;
9744 }
9745
9746 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9747                                 const struct bpf_insn *si,
9748                                 struct bpf_insn *insn_buf,
9749                                 struct bpf_prog *prog, u32 *target_size)
9750 {
9751         struct bpf_insn *insn = insn_buf;
9752         int off;
9753
9754         switch (si->off) {
9755         case offsetof(struct bpf_sock, bound_dev_if):
9756                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9757
9758                 if (type == BPF_WRITE)
9759                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9760                                                  offsetof(struct sock, sk_bound_dev_if));
9761                 else
9762                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9763                                       offsetof(struct sock, sk_bound_dev_if));
9764                 break;
9765
9766         case offsetof(struct bpf_sock, mark):
9767                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9768
9769                 if (type == BPF_WRITE)
9770                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9771                                                  offsetof(struct sock, sk_mark));
9772                 else
9773                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9774                                       offsetof(struct sock, sk_mark));
9775                 break;
9776
9777         case offsetof(struct bpf_sock, priority):
9778                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9779
9780                 if (type == BPF_WRITE)
9781                         *insn++ = BPF_EMIT_STORE(BPF_W, si,
9782                                                  offsetof(struct sock, sk_priority));
9783                 else
9784                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9785                                       offsetof(struct sock, sk_priority));
9786                 break;
9787
9788         case offsetof(struct bpf_sock, family):
9789                 *insn++ = BPF_LDX_MEM(
9790                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9791                         si->dst_reg, si->src_reg,
9792                         bpf_target_off(struct sock_common,
9793                                        skc_family,
9794                                        sizeof_field(struct sock_common,
9795                                                     skc_family),
9796                                        target_size));
9797                 break;
9798
9799         case offsetof(struct bpf_sock, type):
9800                 *insn++ = BPF_LDX_MEM(
9801                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9802                         si->dst_reg, si->src_reg,
9803                         bpf_target_off(struct sock, sk_type,
9804                                        sizeof_field(struct sock, sk_type),
9805                                        target_size));
9806                 break;
9807
9808         case offsetof(struct bpf_sock, protocol):
9809                 *insn++ = BPF_LDX_MEM(
9810                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9811                         si->dst_reg, si->src_reg,
9812                         bpf_target_off(struct sock, sk_protocol,
9813                                        sizeof_field(struct sock, sk_protocol),
9814                                        target_size));
9815                 break;
9816
9817         case offsetof(struct bpf_sock, src_ip4):
9818                 *insn++ = BPF_LDX_MEM(
9819                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9820                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9821                                        sizeof_field(struct sock_common,
9822                                                     skc_rcv_saddr),
9823                                        target_size));
9824                 break;
9825
9826         case offsetof(struct bpf_sock, dst_ip4):
9827                 *insn++ = BPF_LDX_MEM(
9828                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9829                         bpf_target_off(struct sock_common, skc_daddr,
9830                                        sizeof_field(struct sock_common,
9831                                                     skc_daddr),
9832                                        target_size));
9833                 break;
9834
9835         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9836 #if IS_ENABLED(CONFIG_IPV6)
9837                 off = si->off;
9838                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9839                 *insn++ = BPF_LDX_MEM(
9840                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9841                         bpf_target_off(
9842                                 struct sock_common,
9843                                 skc_v6_rcv_saddr.s6_addr32[0],
9844                                 sizeof_field(struct sock_common,
9845                                              skc_v6_rcv_saddr.s6_addr32[0]),
9846                                 target_size) + off);
9847 #else
9848                 (void)off;
9849                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9850 #endif
9851                 break;
9852
9853         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9854 #if IS_ENABLED(CONFIG_IPV6)
9855                 off = si->off;
9856                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9857                 *insn++ = BPF_LDX_MEM(
9858                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9859                         bpf_target_off(struct sock_common,
9860                                        skc_v6_daddr.s6_addr32[0],
9861                                        sizeof_field(struct sock_common,
9862                                                     skc_v6_daddr.s6_addr32[0]),
9863                                        target_size) + off);
9864 #else
9865                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9866                 *target_size = 4;
9867 #endif
9868                 break;
9869
9870         case offsetof(struct bpf_sock, src_port):
9871                 *insn++ = BPF_LDX_MEM(
9872                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9873                         si->dst_reg, si->src_reg,
9874                         bpf_target_off(struct sock_common, skc_num,
9875                                        sizeof_field(struct sock_common,
9876                                                     skc_num),
9877                                        target_size));
9878                 break;
9879
9880         case offsetof(struct bpf_sock, dst_port):
9881                 *insn++ = BPF_LDX_MEM(
9882                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9883                         si->dst_reg, si->src_reg,
9884                         bpf_target_off(struct sock_common, skc_dport,
9885                                        sizeof_field(struct sock_common,
9886                                                     skc_dport),
9887                                        target_size));
9888                 break;
9889
9890         case offsetof(struct bpf_sock, state):
9891                 *insn++ = BPF_LDX_MEM(
9892                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9893                         si->dst_reg, si->src_reg,
9894                         bpf_target_off(struct sock_common, skc_state,
9895                                        sizeof_field(struct sock_common,
9896                                                     skc_state),
9897                                        target_size));
9898                 break;
9899         case offsetof(struct bpf_sock, rx_queue_mapping):
9900 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9901                 *insn++ = BPF_LDX_MEM(
9902                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9903                         si->dst_reg, si->src_reg,
9904                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9905                                        sizeof_field(struct sock,
9906                                                     sk_rx_queue_mapping),
9907                                        target_size));
9908                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9909                                       1);
9910                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9911 #else
9912                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9913                 *target_size = 2;
9914 #endif
9915                 break;
9916         }
9917
9918         return insn - insn_buf;
9919 }
9920
9921 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9922                                          const struct bpf_insn *si,
9923                                          struct bpf_insn *insn_buf,
9924                                          struct bpf_prog *prog, u32 *target_size)
9925 {
9926         struct bpf_insn *insn = insn_buf;
9927
9928         switch (si->off) {
9929         case offsetof(struct __sk_buff, ifindex):
9930                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9931                                       si->dst_reg, si->src_reg,
9932                                       offsetof(struct sk_buff, dev));
9933                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9934                                       bpf_target_off(struct net_device, ifindex, 4,
9935                                                      target_size));
9936                 break;
9937         default:
9938                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9939                                               target_size);
9940         }
9941
9942         return insn - insn_buf;
9943 }
9944
9945 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9946                                   const struct bpf_insn *si,
9947                                   struct bpf_insn *insn_buf,
9948                                   struct bpf_prog *prog, u32 *target_size)
9949 {
9950         struct bpf_insn *insn = insn_buf;
9951
9952         switch (si->off) {
9953         case offsetof(struct xdp_md, data):
9954                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9955                                       si->dst_reg, si->src_reg,
9956                                       offsetof(struct xdp_buff, data));
9957                 break;
9958         case offsetof(struct xdp_md, data_meta):
9959                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9960                                       si->dst_reg, si->src_reg,
9961                                       offsetof(struct xdp_buff, data_meta));
9962                 break;
9963         case offsetof(struct xdp_md, data_end):
9964                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9965                                       si->dst_reg, si->src_reg,
9966                                       offsetof(struct xdp_buff, data_end));
9967                 break;
9968         case offsetof(struct xdp_md, ingress_ifindex):
9969                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9970                                       si->dst_reg, si->src_reg,
9971                                       offsetof(struct xdp_buff, rxq));
9972                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9973                                       si->dst_reg, si->dst_reg,
9974                                       offsetof(struct xdp_rxq_info, dev));
9975                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9976                                       offsetof(struct net_device, ifindex));
9977                 break;
9978         case offsetof(struct xdp_md, rx_queue_index):
9979                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9980                                       si->dst_reg, si->src_reg,
9981                                       offsetof(struct xdp_buff, rxq));
9982                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9983                                       offsetof(struct xdp_rxq_info,
9984                                                queue_index));
9985                 break;
9986         case offsetof(struct xdp_md, egress_ifindex):
9987                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9988                                       si->dst_reg, si->src_reg,
9989                                       offsetof(struct xdp_buff, txq));
9990                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9991                                       si->dst_reg, si->dst_reg,
9992                                       offsetof(struct xdp_txq_info, dev));
9993                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9994                                       offsetof(struct net_device, ifindex));
9995                 break;
9996         }
9997
9998         return insn - insn_buf;
9999 }
10000
10001 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
10002  * context Structure, F is Field in context structure that contains a pointer
10003  * to Nested Structure of type NS that has the field NF.
10004  *
10005  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
10006  * sure that SIZE is not greater than actual size of S.F.NF.
10007  *
10008  * If offset OFF is provided, the load happens from that offset relative to
10009  * offset of NF.
10010  */
10011 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
10012         do {                                                                   \
10013                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
10014                                       si->src_reg, offsetof(S, F));            \
10015                 *insn++ = BPF_LDX_MEM(                                         \
10016                         SIZE, si->dst_reg, si->dst_reg,                        \
10017                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10018                                        target_size)                            \
10019                                 + OFF);                                        \
10020         } while (0)
10021
10022 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
10023         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
10024                                              BPF_FIELD_SIZEOF(NS, NF), 0)
10025
10026 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
10027  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
10028  *
10029  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
10030  * "register" since two registers available in convert_ctx_access are not
10031  * enough: we can't override neither SRC, since it contains value to store, nor
10032  * DST since it contains pointer to context that may be used by later
10033  * instructions. But we need a temporary place to save pointer to nested
10034  * structure whose field we want to store to.
10035  */
10036 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
10037         do {                                                                   \
10038                 int tmp_reg = BPF_REG_9;                                       \
10039                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10040                         --tmp_reg;                                             \
10041                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
10042                         --tmp_reg;                                             \
10043                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
10044                                       offsetof(S, TF));                        \
10045                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
10046                                       si->dst_reg, offsetof(S, F));            \
10047                 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code),   \
10048                                        tmp_reg, si->src_reg,                   \
10049                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
10050                                        target_size)                            \
10051                                        + OFF,                                  \
10052                                        si->imm);                               \
10053                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
10054                                       offsetof(S, TF));                        \
10055         } while (0)
10056
10057 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
10058                                                       TF)                      \
10059         do {                                                                   \
10060                 if (type == BPF_WRITE) {                                       \
10061                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
10062                                                          OFF, TF);             \
10063                 } else {                                                       \
10064                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
10065                                 S, NS, F, NF, SIZE, OFF);  \
10066                 }                                                              \
10067         } while (0)
10068
10069 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
10070         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
10071                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
10072
10073 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
10074                                         const struct bpf_insn *si,
10075                                         struct bpf_insn *insn_buf,
10076                                         struct bpf_prog *prog, u32 *target_size)
10077 {
10078         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
10079         struct bpf_insn *insn = insn_buf;
10080
10081         switch (si->off) {
10082         case offsetof(struct bpf_sock_addr, user_family):
10083                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10084                                             struct sockaddr, uaddr, sa_family);
10085                 break;
10086
10087         case offsetof(struct bpf_sock_addr, user_ip4):
10088                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10089                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
10090                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
10091                 break;
10092
10093         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
10094                 off = si->off;
10095                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
10096                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10097                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10098                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
10099                         tmp_reg);
10100                 break;
10101
10102         case offsetof(struct bpf_sock_addr, user_port):
10103                 /* To get port we need to know sa_family first and then treat
10104                  * sockaddr as either sockaddr_in or sockaddr_in6.
10105                  * Though we can simplify since port field has same offset and
10106                  * size in both structures.
10107                  * Here we check this invariant and use just one of the
10108                  * structures if it's true.
10109                  */
10110                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
10111                              offsetof(struct sockaddr_in6, sin6_port));
10112                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
10113                              sizeof_field(struct sockaddr_in6, sin6_port));
10114                 /* Account for sin6_port being smaller than user_port. */
10115                 port_size = min(port_size, BPF_LDST_BYTES(si));
10116                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10117                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
10118                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
10119                 break;
10120
10121         case offsetof(struct bpf_sock_addr, family):
10122                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10123                                             struct sock, sk, sk_family);
10124                 break;
10125
10126         case offsetof(struct bpf_sock_addr, type):
10127                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10128                                             struct sock, sk, sk_type);
10129                 break;
10130
10131         case offsetof(struct bpf_sock_addr, protocol):
10132                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10133                                             struct sock, sk, sk_protocol);
10134                 break;
10135
10136         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10137                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10138                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10139                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10140                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10141                 break;
10142
10143         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10144                                 msg_src_ip6[3]):
10145                 off = si->off;
10146                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10147                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10148                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10149                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10150                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10151                 break;
10152         case offsetof(struct bpf_sock_addr, sk):
10153                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10154                                       si->dst_reg, si->src_reg,
10155                                       offsetof(struct bpf_sock_addr_kern, sk));
10156                 break;
10157         }
10158
10159         return insn - insn_buf;
10160 }
10161
10162 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10163                                        const struct bpf_insn *si,
10164                                        struct bpf_insn *insn_buf,
10165                                        struct bpf_prog *prog,
10166                                        u32 *target_size)
10167 {
10168         struct bpf_insn *insn = insn_buf;
10169         int off;
10170
10171 /* Helper macro for adding read access to tcp_sock or sock fields. */
10172 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10173         do {                                                                  \
10174                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10175                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10176                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10177                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10178                         reg--;                                                \
10179                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10180                         reg--;                                                \
10181                 if (si->dst_reg == si->src_reg) {                             \
10182                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10183                                           offsetof(struct bpf_sock_ops_kern,  \
10184                                           temp));                             \
10185                         fullsock_reg = reg;                                   \
10186                         jmp += 2;                                             \
10187                 }                                                             \
10188                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10189                                                 struct bpf_sock_ops_kern,     \
10190                                                 is_fullsock),                 \
10191                                       fullsock_reg, si->src_reg,              \
10192                                       offsetof(struct bpf_sock_ops_kern,      \
10193                                                is_fullsock));                 \
10194                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10195                 if (si->dst_reg == si->src_reg)                               \
10196                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10197                                       offsetof(struct bpf_sock_ops_kern,      \
10198                                       temp));                                 \
10199                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10200                                                 struct bpf_sock_ops_kern, sk),\
10201                                       si->dst_reg, si->src_reg,               \
10202                                       offsetof(struct bpf_sock_ops_kern, sk));\
10203                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10204                                                        OBJ_FIELD),            \
10205                                       si->dst_reg, si->dst_reg,               \
10206                                       offsetof(OBJ, OBJ_FIELD));              \
10207                 if (si->dst_reg == si->src_reg) {                             \
10208                         *insn++ = BPF_JMP_A(1);                               \
10209                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10210                                       offsetof(struct bpf_sock_ops_kern,      \
10211                                       temp));                                 \
10212                 }                                                             \
10213         } while (0)
10214
10215 #define SOCK_OPS_GET_SK()                                                             \
10216         do {                                                                  \
10217                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10218                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10219                         reg--;                                                \
10220                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10221                         reg--;                                                \
10222                 if (si->dst_reg == si->src_reg) {                             \
10223                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10224                                           offsetof(struct bpf_sock_ops_kern,  \
10225                                           temp));                             \
10226                         fullsock_reg = reg;                                   \
10227                         jmp += 2;                                             \
10228                 }                                                             \
10229                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10230                                                 struct bpf_sock_ops_kern,     \
10231                                                 is_fullsock),                 \
10232                                       fullsock_reg, si->src_reg,              \
10233                                       offsetof(struct bpf_sock_ops_kern,      \
10234                                                is_fullsock));                 \
10235                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10236                 if (si->dst_reg == si->src_reg)                               \
10237                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10238                                       offsetof(struct bpf_sock_ops_kern,      \
10239                                       temp));                                 \
10240                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10241                                                 struct bpf_sock_ops_kern, sk),\
10242                                       si->dst_reg, si->src_reg,               \
10243                                       offsetof(struct bpf_sock_ops_kern, sk));\
10244                 if (si->dst_reg == si->src_reg) {                             \
10245                         *insn++ = BPF_JMP_A(1);                               \
10246                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10247                                       offsetof(struct bpf_sock_ops_kern,      \
10248                                       temp));                                 \
10249                 }                                                             \
10250         } while (0)
10251
10252 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10253                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10254
10255 /* Helper macro for adding write access to tcp_sock or sock fields.
10256  * The macro is called with two registers, dst_reg which contains a pointer
10257  * to ctx (context) and src_reg which contains the value that should be
10258  * stored. However, we need an additional register since we cannot overwrite
10259  * dst_reg because it may be used later in the program.
10260  * Instead we "borrow" one of the other register. We first save its value
10261  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10262  * it at the end of the macro.
10263  */
10264 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10265         do {                                                                  \
10266                 int reg = BPF_REG_9;                                          \
10267                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10268                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10269                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10270                         reg--;                                                \
10271                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10272                         reg--;                                                \
10273                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10274                                       offsetof(struct bpf_sock_ops_kern,      \
10275                                                temp));                        \
10276                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10277                                                 struct bpf_sock_ops_kern,     \
10278                                                 is_fullsock),                 \
10279                                       reg, si->dst_reg,                       \
10280                                       offsetof(struct bpf_sock_ops_kern,      \
10281                                                is_fullsock));                 \
10282                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10283                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10284                                                 struct bpf_sock_ops_kern, sk),\
10285                                       reg, si->dst_reg,                       \
10286                                       offsetof(struct bpf_sock_ops_kern, sk));\
10287                 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) |     \
10288                                        BPF_MEM | BPF_CLASS(si->code),         \
10289                                        reg, si->src_reg,                      \
10290                                        offsetof(OBJ, OBJ_FIELD),              \
10291                                        si->imm);                              \
10292                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10293                                       offsetof(struct bpf_sock_ops_kern,      \
10294                                                temp));                        \
10295         } while (0)
10296
10297 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10298         do {                                                                  \
10299                 if (TYPE == BPF_WRITE)                                        \
10300                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10301                 else                                                          \
10302                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10303         } while (0)
10304
10305         switch (si->off) {
10306         case offsetof(struct bpf_sock_ops, op):
10307                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10308                                                        op),
10309                                       si->dst_reg, si->src_reg,
10310                                       offsetof(struct bpf_sock_ops_kern, op));
10311                 break;
10312
10313         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10314              offsetof(struct bpf_sock_ops, replylong[3]):
10315                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10316                              sizeof_field(struct bpf_sock_ops_kern, reply));
10317                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10318                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10319                 off = si->off;
10320                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10321                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10322                 if (type == BPF_WRITE)
10323                         *insn++ = BPF_EMIT_STORE(BPF_W, si, off);
10324                 else
10325                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10326                                               off);
10327                 break;
10328
10329         case offsetof(struct bpf_sock_ops, family):
10330                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10331
10332                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10333                                               struct bpf_sock_ops_kern, sk),
10334                                       si->dst_reg, si->src_reg,
10335                                       offsetof(struct bpf_sock_ops_kern, sk));
10336                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10337                                       offsetof(struct sock_common, skc_family));
10338                 break;
10339
10340         case offsetof(struct bpf_sock_ops, remote_ip4):
10341                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10342
10343                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10344                                                 struct bpf_sock_ops_kern, sk),
10345                                       si->dst_reg, si->src_reg,
10346                                       offsetof(struct bpf_sock_ops_kern, sk));
10347                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10348                                       offsetof(struct sock_common, skc_daddr));
10349                 break;
10350
10351         case offsetof(struct bpf_sock_ops, local_ip4):
10352                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10353                                           skc_rcv_saddr) != 4);
10354
10355                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10356                                               struct bpf_sock_ops_kern, sk),
10357                                       si->dst_reg, si->src_reg,
10358                                       offsetof(struct bpf_sock_ops_kern, sk));
10359                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10360                                       offsetof(struct sock_common,
10361                                                skc_rcv_saddr));
10362                 break;
10363
10364         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10365              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10366 #if IS_ENABLED(CONFIG_IPV6)
10367                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10368                                           skc_v6_daddr.s6_addr32[0]) != 4);
10369
10370                 off = si->off;
10371                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10372                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10373                                                 struct bpf_sock_ops_kern, sk),
10374                                       si->dst_reg, si->src_reg,
10375                                       offsetof(struct bpf_sock_ops_kern, sk));
10376                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10377                                       offsetof(struct sock_common,
10378                                                skc_v6_daddr.s6_addr32[0]) +
10379                                       off);
10380 #else
10381                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10382 #endif
10383                 break;
10384
10385         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10386              offsetof(struct bpf_sock_ops, local_ip6[3]):
10387 #if IS_ENABLED(CONFIG_IPV6)
10388                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10389                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10390
10391                 off = si->off;
10392                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10393                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10394                                                 struct bpf_sock_ops_kern, sk),
10395                                       si->dst_reg, si->src_reg,
10396                                       offsetof(struct bpf_sock_ops_kern, sk));
10397                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10398                                       offsetof(struct sock_common,
10399                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10400                                       off);
10401 #else
10402                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10403 #endif
10404                 break;
10405
10406         case offsetof(struct bpf_sock_ops, remote_port):
10407                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10408
10409                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10410                                                 struct bpf_sock_ops_kern, sk),
10411                                       si->dst_reg, si->src_reg,
10412                                       offsetof(struct bpf_sock_ops_kern, sk));
10413                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10414                                       offsetof(struct sock_common, skc_dport));
10415 #ifndef __BIG_ENDIAN_BITFIELD
10416                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10417 #endif
10418                 break;
10419
10420         case offsetof(struct bpf_sock_ops, local_port):
10421                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10422
10423                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10424                                                 struct bpf_sock_ops_kern, sk),
10425                                       si->dst_reg, si->src_reg,
10426                                       offsetof(struct bpf_sock_ops_kern, sk));
10427                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10428                                       offsetof(struct sock_common, skc_num));
10429                 break;
10430
10431         case offsetof(struct bpf_sock_ops, is_fullsock):
10432                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10433                                                 struct bpf_sock_ops_kern,
10434                                                 is_fullsock),
10435                                       si->dst_reg, si->src_reg,
10436                                       offsetof(struct bpf_sock_ops_kern,
10437                                                is_fullsock));
10438                 break;
10439
10440         case offsetof(struct bpf_sock_ops, state):
10441                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10442
10443                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10444                                                 struct bpf_sock_ops_kern, sk),
10445                                       si->dst_reg, si->src_reg,
10446                                       offsetof(struct bpf_sock_ops_kern, sk));
10447                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10448                                       offsetof(struct sock_common, skc_state));
10449                 break;
10450
10451         case offsetof(struct bpf_sock_ops, rtt_min):
10452                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10453                              sizeof(struct minmax));
10454                 BUILD_BUG_ON(sizeof(struct minmax) <
10455                              sizeof(struct minmax_sample));
10456
10457                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10458                                                 struct bpf_sock_ops_kern, sk),
10459                                       si->dst_reg, si->src_reg,
10460                                       offsetof(struct bpf_sock_ops_kern, sk));
10461                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10462                                       offsetof(struct tcp_sock, rtt_min) +
10463                                       sizeof_field(struct minmax_sample, t));
10464                 break;
10465
10466         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10467                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10468                                    struct tcp_sock);
10469                 break;
10470
10471         case offsetof(struct bpf_sock_ops, sk_txhash):
10472                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10473                                           struct sock, type);
10474                 break;
10475         case offsetof(struct bpf_sock_ops, snd_cwnd):
10476                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10477                 break;
10478         case offsetof(struct bpf_sock_ops, srtt_us):
10479                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10480                 break;
10481         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10482                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10483                 break;
10484         case offsetof(struct bpf_sock_ops, rcv_nxt):
10485                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10486                 break;
10487         case offsetof(struct bpf_sock_ops, snd_nxt):
10488                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10489                 break;
10490         case offsetof(struct bpf_sock_ops, snd_una):
10491                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10492                 break;
10493         case offsetof(struct bpf_sock_ops, mss_cache):
10494                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10495                 break;
10496         case offsetof(struct bpf_sock_ops, ecn_flags):
10497                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10498                 break;
10499         case offsetof(struct bpf_sock_ops, rate_delivered):
10500                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10501                 break;
10502         case offsetof(struct bpf_sock_ops, rate_interval_us):
10503                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10504                 break;
10505         case offsetof(struct bpf_sock_ops, packets_out):
10506                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10507                 break;
10508         case offsetof(struct bpf_sock_ops, retrans_out):
10509                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10510                 break;
10511         case offsetof(struct bpf_sock_ops, total_retrans):
10512                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10513                 break;
10514         case offsetof(struct bpf_sock_ops, segs_in):
10515                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10516                 break;
10517         case offsetof(struct bpf_sock_ops, data_segs_in):
10518                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10519                 break;
10520         case offsetof(struct bpf_sock_ops, segs_out):
10521                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10522                 break;
10523         case offsetof(struct bpf_sock_ops, data_segs_out):
10524                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10525                 break;
10526         case offsetof(struct bpf_sock_ops, lost_out):
10527                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10528                 break;
10529         case offsetof(struct bpf_sock_ops, sacked_out):
10530                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10531                 break;
10532         case offsetof(struct bpf_sock_ops, bytes_received):
10533                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10534                 break;
10535         case offsetof(struct bpf_sock_ops, bytes_acked):
10536                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10537                 break;
10538         case offsetof(struct bpf_sock_ops, sk):
10539                 SOCK_OPS_GET_SK();
10540                 break;
10541         case offsetof(struct bpf_sock_ops, skb_data_end):
10542                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10543                                                        skb_data_end),
10544                                       si->dst_reg, si->src_reg,
10545                                       offsetof(struct bpf_sock_ops_kern,
10546                                                skb_data_end));
10547                 break;
10548         case offsetof(struct bpf_sock_ops, skb_data):
10549                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10550                                                        skb),
10551                                       si->dst_reg, si->src_reg,
10552                                       offsetof(struct bpf_sock_ops_kern,
10553                                                skb));
10554                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10555                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10556                                       si->dst_reg, si->dst_reg,
10557                                       offsetof(struct sk_buff, data));
10558                 break;
10559         case offsetof(struct bpf_sock_ops, skb_len):
10560                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10561                                                        skb),
10562                                       si->dst_reg, si->src_reg,
10563                                       offsetof(struct bpf_sock_ops_kern,
10564                                                skb));
10565                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10566                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10567                                       si->dst_reg, si->dst_reg,
10568                                       offsetof(struct sk_buff, len));
10569                 break;
10570         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10571                 off = offsetof(struct sk_buff, cb);
10572                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10573                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10574                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10575                                                        skb),
10576                                       si->dst_reg, si->src_reg,
10577                                       offsetof(struct bpf_sock_ops_kern,
10578                                                skb));
10579                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10580                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10581                                                        tcp_flags),
10582                                       si->dst_reg, si->dst_reg, off);
10583                 break;
10584         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10585                 struct bpf_insn *jmp_on_null_skb;
10586
10587                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10588                                                        skb),
10589                                       si->dst_reg, si->src_reg,
10590                                       offsetof(struct bpf_sock_ops_kern,
10591                                                skb));
10592                 /* Reserve one insn to test skb == NULL */
10593                 jmp_on_null_skb = insn++;
10594                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10595                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10596                                       bpf_target_off(struct skb_shared_info,
10597                                                      hwtstamps, 8,
10598                                                      target_size));
10599                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10600                                                insn - jmp_on_null_skb - 1);
10601                 break;
10602         }
10603         }
10604         return insn - insn_buf;
10605 }
10606
10607 /* data_end = skb->data + skb_headlen() */
10608 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10609                                                     struct bpf_insn *insn)
10610 {
10611         int reg;
10612         int temp_reg_off = offsetof(struct sk_buff, cb) +
10613                            offsetof(struct sk_skb_cb, temp_reg);
10614
10615         if (si->src_reg == si->dst_reg) {
10616                 /* We need an extra register, choose and save a register. */
10617                 reg = BPF_REG_9;
10618                 if (si->src_reg == reg || si->dst_reg == reg)
10619                         reg--;
10620                 if (si->src_reg == reg || si->dst_reg == reg)
10621                         reg--;
10622                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10623         } else {
10624                 reg = si->dst_reg;
10625         }
10626
10627         /* reg = skb->data */
10628         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10629                               reg, si->src_reg,
10630                               offsetof(struct sk_buff, data));
10631         /* AX = skb->len */
10632         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10633                               BPF_REG_AX, si->src_reg,
10634                               offsetof(struct sk_buff, len));
10635         /* reg = skb->data + skb->len */
10636         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10637         /* AX = skb->data_len */
10638         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10639                               BPF_REG_AX, si->src_reg,
10640                               offsetof(struct sk_buff, data_len));
10641
10642         /* reg = skb->data + skb->len - skb->data_len */
10643         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10644
10645         if (si->src_reg == si->dst_reg) {
10646                 /* Restore the saved register */
10647                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10648                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10649                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10650         }
10651
10652         return insn;
10653 }
10654
10655 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10656                                      const struct bpf_insn *si,
10657                                      struct bpf_insn *insn_buf,
10658                                      struct bpf_prog *prog, u32 *target_size)
10659 {
10660         struct bpf_insn *insn = insn_buf;
10661         int off;
10662
10663         switch (si->off) {
10664         case offsetof(struct __sk_buff, data_end):
10665                 insn = bpf_convert_data_end_access(si, insn);
10666                 break;
10667         case offsetof(struct __sk_buff, cb[0]) ...
10668              offsetofend(struct __sk_buff, cb[4]) - 1:
10669                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10670                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10671                               offsetof(struct sk_skb_cb, data)) %
10672                              sizeof(__u64));
10673
10674                 prog->cb_access = 1;
10675                 off  = si->off;
10676                 off -= offsetof(struct __sk_buff, cb[0]);
10677                 off += offsetof(struct sk_buff, cb);
10678                 off += offsetof(struct sk_skb_cb, data);
10679                 if (type == BPF_WRITE)
10680                         *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off);
10681                 else
10682                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10683                                               si->src_reg, off);
10684                 break;
10685
10686
10687         default:
10688                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10689                                               target_size);
10690         }
10691
10692         return insn - insn_buf;
10693 }
10694
10695 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10696                                      const struct bpf_insn *si,
10697                                      struct bpf_insn *insn_buf,
10698                                      struct bpf_prog *prog, u32 *target_size)
10699 {
10700         struct bpf_insn *insn = insn_buf;
10701 #if IS_ENABLED(CONFIG_IPV6)
10702         int off;
10703 #endif
10704
10705         /* convert ctx uses the fact sg element is first in struct */
10706         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10707
10708         switch (si->off) {
10709         case offsetof(struct sk_msg_md, data):
10710                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10711                                       si->dst_reg, si->src_reg,
10712                                       offsetof(struct sk_msg, data));
10713                 break;
10714         case offsetof(struct sk_msg_md, data_end):
10715                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10716                                       si->dst_reg, si->src_reg,
10717                                       offsetof(struct sk_msg, data_end));
10718                 break;
10719         case offsetof(struct sk_msg_md, family):
10720                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10721
10722                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10723                                               struct sk_msg, sk),
10724                                       si->dst_reg, si->src_reg,
10725                                       offsetof(struct sk_msg, sk));
10726                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10727                                       offsetof(struct sock_common, skc_family));
10728                 break;
10729
10730         case offsetof(struct sk_msg_md, remote_ip4):
10731                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10732
10733                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10734                                                 struct sk_msg, sk),
10735                                       si->dst_reg, si->src_reg,
10736                                       offsetof(struct sk_msg, sk));
10737                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10738                                       offsetof(struct sock_common, skc_daddr));
10739                 break;
10740
10741         case offsetof(struct sk_msg_md, local_ip4):
10742                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10743                                           skc_rcv_saddr) != 4);
10744
10745                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10746                                               struct sk_msg, sk),
10747                                       si->dst_reg, si->src_reg,
10748                                       offsetof(struct sk_msg, sk));
10749                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10750                                       offsetof(struct sock_common,
10751                                                skc_rcv_saddr));
10752                 break;
10753
10754         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10755              offsetof(struct sk_msg_md, remote_ip6[3]):
10756 #if IS_ENABLED(CONFIG_IPV6)
10757                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10758                                           skc_v6_daddr.s6_addr32[0]) != 4);
10759
10760                 off = si->off;
10761                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10762                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10763                                                 struct sk_msg, sk),
10764                                       si->dst_reg, si->src_reg,
10765                                       offsetof(struct sk_msg, sk));
10766                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10767                                       offsetof(struct sock_common,
10768                                                skc_v6_daddr.s6_addr32[0]) +
10769                                       off);
10770 #else
10771                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10772 #endif
10773                 break;
10774
10775         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10776              offsetof(struct sk_msg_md, local_ip6[3]):
10777 #if IS_ENABLED(CONFIG_IPV6)
10778                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10779                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10780
10781                 off = si->off;
10782                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10783                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10784                                                 struct sk_msg, sk),
10785                                       si->dst_reg, si->src_reg,
10786                                       offsetof(struct sk_msg, sk));
10787                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10788                                       offsetof(struct sock_common,
10789                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10790                                       off);
10791 #else
10792                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10793 #endif
10794                 break;
10795
10796         case offsetof(struct sk_msg_md, remote_port):
10797                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10798
10799                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10800                                                 struct sk_msg, sk),
10801                                       si->dst_reg, si->src_reg,
10802                                       offsetof(struct sk_msg, sk));
10803                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10804                                       offsetof(struct sock_common, skc_dport));
10805 #ifndef __BIG_ENDIAN_BITFIELD
10806                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10807 #endif
10808                 break;
10809
10810         case offsetof(struct sk_msg_md, local_port):
10811                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10812
10813                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10814                                                 struct sk_msg, sk),
10815                                       si->dst_reg, si->src_reg,
10816                                       offsetof(struct sk_msg, sk));
10817                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10818                                       offsetof(struct sock_common, skc_num));
10819                 break;
10820
10821         case offsetof(struct sk_msg_md, size):
10822                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10823                                       si->dst_reg, si->src_reg,
10824                                       offsetof(struct sk_msg_sg, size));
10825                 break;
10826
10827         case offsetof(struct sk_msg_md, sk):
10828                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10829                                       si->dst_reg, si->src_reg,
10830                                       offsetof(struct sk_msg, sk));
10831                 break;
10832         }
10833
10834         return insn - insn_buf;
10835 }
10836
10837 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10838         .get_func_proto         = sk_filter_func_proto,
10839         .is_valid_access        = sk_filter_is_valid_access,
10840         .convert_ctx_access     = bpf_convert_ctx_access,
10841         .gen_ld_abs             = bpf_gen_ld_abs,
10842 };
10843
10844 const struct bpf_prog_ops sk_filter_prog_ops = {
10845         .test_run               = bpf_prog_test_run_skb,
10846 };
10847
10848 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10849         .get_func_proto         = tc_cls_act_func_proto,
10850         .is_valid_access        = tc_cls_act_is_valid_access,
10851         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10852         .gen_prologue           = tc_cls_act_prologue,
10853         .gen_ld_abs             = bpf_gen_ld_abs,
10854         .btf_struct_access      = tc_cls_act_btf_struct_access,
10855 };
10856
10857 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10858         .test_run               = bpf_prog_test_run_skb,
10859 };
10860
10861 const struct bpf_verifier_ops xdp_verifier_ops = {
10862         .get_func_proto         = xdp_func_proto,
10863         .is_valid_access        = xdp_is_valid_access,
10864         .convert_ctx_access     = xdp_convert_ctx_access,
10865         .gen_prologue           = bpf_noop_prologue,
10866         .btf_struct_access      = xdp_btf_struct_access,
10867 };
10868
10869 const struct bpf_prog_ops xdp_prog_ops = {
10870         .test_run               = bpf_prog_test_run_xdp,
10871 };
10872
10873 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10874         .get_func_proto         = cg_skb_func_proto,
10875         .is_valid_access        = cg_skb_is_valid_access,
10876         .convert_ctx_access     = bpf_convert_ctx_access,
10877 };
10878
10879 const struct bpf_prog_ops cg_skb_prog_ops = {
10880         .test_run               = bpf_prog_test_run_skb,
10881 };
10882
10883 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10884         .get_func_proto         = lwt_in_func_proto,
10885         .is_valid_access        = lwt_is_valid_access,
10886         .convert_ctx_access     = bpf_convert_ctx_access,
10887 };
10888
10889 const struct bpf_prog_ops lwt_in_prog_ops = {
10890         .test_run               = bpf_prog_test_run_skb,
10891 };
10892
10893 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10894         .get_func_proto         = lwt_out_func_proto,
10895         .is_valid_access        = lwt_is_valid_access,
10896         .convert_ctx_access     = bpf_convert_ctx_access,
10897 };
10898
10899 const struct bpf_prog_ops lwt_out_prog_ops = {
10900         .test_run               = bpf_prog_test_run_skb,
10901 };
10902
10903 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10904         .get_func_proto         = lwt_xmit_func_proto,
10905         .is_valid_access        = lwt_is_valid_access,
10906         .convert_ctx_access     = bpf_convert_ctx_access,
10907         .gen_prologue           = tc_cls_act_prologue,
10908 };
10909
10910 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10911         .test_run               = bpf_prog_test_run_skb,
10912 };
10913
10914 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10915         .get_func_proto         = lwt_seg6local_func_proto,
10916         .is_valid_access        = lwt_is_valid_access,
10917         .convert_ctx_access     = bpf_convert_ctx_access,
10918 };
10919
10920 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10921         .test_run               = bpf_prog_test_run_skb,
10922 };
10923
10924 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10925         .get_func_proto         = sock_filter_func_proto,
10926         .is_valid_access        = sock_filter_is_valid_access,
10927         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10928 };
10929
10930 const struct bpf_prog_ops cg_sock_prog_ops = {
10931 };
10932
10933 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10934         .get_func_proto         = sock_addr_func_proto,
10935         .is_valid_access        = sock_addr_is_valid_access,
10936         .convert_ctx_access     = sock_addr_convert_ctx_access,
10937 };
10938
10939 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10940 };
10941
10942 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10943         .get_func_proto         = sock_ops_func_proto,
10944         .is_valid_access        = sock_ops_is_valid_access,
10945         .convert_ctx_access     = sock_ops_convert_ctx_access,
10946 };
10947
10948 const struct bpf_prog_ops sock_ops_prog_ops = {
10949 };
10950
10951 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10952         .get_func_proto         = sk_skb_func_proto,
10953         .is_valid_access        = sk_skb_is_valid_access,
10954         .convert_ctx_access     = sk_skb_convert_ctx_access,
10955         .gen_prologue           = sk_skb_prologue,
10956 };
10957
10958 const struct bpf_prog_ops sk_skb_prog_ops = {
10959 };
10960
10961 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10962         .get_func_proto         = sk_msg_func_proto,
10963         .is_valid_access        = sk_msg_is_valid_access,
10964         .convert_ctx_access     = sk_msg_convert_ctx_access,
10965         .gen_prologue           = bpf_noop_prologue,
10966 };
10967
10968 const struct bpf_prog_ops sk_msg_prog_ops = {
10969 };
10970
10971 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10972         .get_func_proto         = flow_dissector_func_proto,
10973         .is_valid_access        = flow_dissector_is_valid_access,
10974         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10975 };
10976
10977 const struct bpf_prog_ops flow_dissector_prog_ops = {
10978         .test_run               = bpf_prog_test_run_flow_dissector,
10979 };
10980
10981 int sk_detach_filter(struct sock *sk)
10982 {
10983         int ret = -ENOENT;
10984         struct sk_filter *filter;
10985
10986         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10987                 return -EPERM;
10988
10989         filter = rcu_dereference_protected(sk->sk_filter,
10990                                            lockdep_sock_is_held(sk));
10991         if (filter) {
10992                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10993                 sk_filter_uncharge(sk, filter);
10994                 ret = 0;
10995         }
10996
10997         return ret;
10998 }
10999 EXPORT_SYMBOL_GPL(sk_detach_filter);
11000
11001 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
11002 {
11003         struct sock_fprog_kern *fprog;
11004         struct sk_filter *filter;
11005         int ret = 0;
11006
11007         sockopt_lock_sock(sk);
11008         filter = rcu_dereference_protected(sk->sk_filter,
11009                                            lockdep_sock_is_held(sk));
11010         if (!filter)
11011                 goto out;
11012
11013         /* We're copying the filter that has been originally attached,
11014          * so no conversion/decode needed anymore. eBPF programs that
11015          * have no original program cannot be dumped through this.
11016          */
11017         ret = -EACCES;
11018         fprog = filter->prog->orig_prog;
11019         if (!fprog)
11020                 goto out;
11021
11022         ret = fprog->len;
11023         if (!len)
11024                 /* User space only enquires number of filter blocks. */
11025                 goto out;
11026
11027         ret = -EINVAL;
11028         if (len < fprog->len)
11029                 goto out;
11030
11031         ret = -EFAULT;
11032         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
11033                 goto out;
11034
11035         /* Instead of bytes, the API requests to return the number
11036          * of filter blocks.
11037          */
11038         ret = fprog->len;
11039 out:
11040         sockopt_release_sock(sk);
11041         return ret;
11042 }
11043
11044 #ifdef CONFIG_INET
11045 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
11046                                     struct sock_reuseport *reuse,
11047                                     struct sock *sk, struct sk_buff *skb,
11048                                     struct sock *migrating_sk,
11049                                     u32 hash)
11050 {
11051         reuse_kern->skb = skb;
11052         reuse_kern->sk = sk;
11053         reuse_kern->selected_sk = NULL;
11054         reuse_kern->migrating_sk = migrating_sk;
11055         reuse_kern->data_end = skb->data + skb_headlen(skb);
11056         reuse_kern->hash = hash;
11057         reuse_kern->reuseport_id = reuse->reuseport_id;
11058         reuse_kern->bind_inany = reuse->bind_inany;
11059 }
11060
11061 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
11062                                   struct bpf_prog *prog, struct sk_buff *skb,
11063                                   struct sock *migrating_sk,
11064                                   u32 hash)
11065 {
11066         struct sk_reuseport_kern reuse_kern;
11067         enum sk_action action;
11068
11069         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
11070         action = bpf_prog_run(prog, &reuse_kern);
11071
11072         if (action == SK_PASS)
11073                 return reuse_kern.selected_sk;
11074         else
11075                 return ERR_PTR(-ECONNREFUSED);
11076 }
11077
11078 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
11079            struct bpf_map *, map, void *, key, u32, flags)
11080 {
11081         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
11082         struct sock_reuseport *reuse;
11083         struct sock *selected_sk;
11084
11085         selected_sk = map->ops->map_lookup_elem(map, key);
11086         if (!selected_sk)
11087                 return -ENOENT;
11088
11089         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
11090         if (!reuse) {
11091                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
11092                 if (sk_is_refcounted(selected_sk))
11093                         sock_put(selected_sk);
11094
11095                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
11096                  * The only (!reuse) case here is - the sk has already been
11097                  * unhashed (e.g. by close()), so treat it as -ENOENT.
11098                  *
11099                  * Other maps (e.g. sock_map) do not provide this guarantee and
11100                  * the sk may never be in the reuseport group to begin with.
11101                  */
11102                 return is_sockarray ? -ENOENT : -EINVAL;
11103         }
11104
11105         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
11106                 struct sock *sk = reuse_kern->sk;
11107
11108                 if (sk->sk_protocol != selected_sk->sk_protocol)
11109                         return -EPROTOTYPE;
11110                 else if (sk->sk_family != selected_sk->sk_family)
11111                         return -EAFNOSUPPORT;
11112
11113                 /* Catch all. Likely bound to a different sockaddr. */
11114                 return -EBADFD;
11115         }
11116
11117         reuse_kern->selected_sk = selected_sk;
11118
11119         return 0;
11120 }
11121
11122 static const struct bpf_func_proto sk_select_reuseport_proto = {
11123         .func           = sk_select_reuseport,
11124         .gpl_only       = false,
11125         .ret_type       = RET_INTEGER,
11126         .arg1_type      = ARG_PTR_TO_CTX,
11127         .arg2_type      = ARG_CONST_MAP_PTR,
11128         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11129         .arg4_type      = ARG_ANYTHING,
11130 };
11131
11132 BPF_CALL_4(sk_reuseport_load_bytes,
11133            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11134            void *, to, u32, len)
11135 {
11136         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11137 }
11138
11139 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11140         .func           = sk_reuseport_load_bytes,
11141         .gpl_only       = false,
11142         .ret_type       = RET_INTEGER,
11143         .arg1_type      = ARG_PTR_TO_CTX,
11144         .arg2_type      = ARG_ANYTHING,
11145         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11146         .arg4_type      = ARG_CONST_SIZE,
11147 };
11148
11149 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11150            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11151            void *, to, u32, len, u32, start_header)
11152 {
11153         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11154                                                len, start_header);
11155 }
11156
11157 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11158         .func           = sk_reuseport_load_bytes_relative,
11159         .gpl_only       = false,
11160         .ret_type       = RET_INTEGER,
11161         .arg1_type      = ARG_PTR_TO_CTX,
11162         .arg2_type      = ARG_ANYTHING,
11163         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11164         .arg4_type      = ARG_CONST_SIZE,
11165         .arg5_type      = ARG_ANYTHING,
11166 };
11167
11168 static const struct bpf_func_proto *
11169 sk_reuseport_func_proto(enum bpf_func_id func_id,
11170                         const struct bpf_prog *prog)
11171 {
11172         switch (func_id) {
11173         case BPF_FUNC_sk_select_reuseport:
11174                 return &sk_select_reuseport_proto;
11175         case BPF_FUNC_skb_load_bytes:
11176                 return &sk_reuseport_load_bytes_proto;
11177         case BPF_FUNC_skb_load_bytes_relative:
11178                 return &sk_reuseport_load_bytes_relative_proto;
11179         case BPF_FUNC_get_socket_cookie:
11180                 return &bpf_get_socket_ptr_cookie_proto;
11181         case BPF_FUNC_ktime_get_coarse_ns:
11182                 return &bpf_ktime_get_coarse_ns_proto;
11183         default:
11184                 return bpf_base_func_proto(func_id);
11185         }
11186 }
11187
11188 static bool
11189 sk_reuseport_is_valid_access(int off, int size,
11190                              enum bpf_access_type type,
11191                              const struct bpf_prog *prog,
11192                              struct bpf_insn_access_aux *info)
11193 {
11194         const u32 size_default = sizeof(__u32);
11195
11196         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11197             off % size || type != BPF_READ)
11198                 return false;
11199
11200         switch (off) {
11201         case offsetof(struct sk_reuseport_md, data):
11202                 info->reg_type = PTR_TO_PACKET;
11203                 return size == sizeof(__u64);
11204
11205         case offsetof(struct sk_reuseport_md, data_end):
11206                 info->reg_type = PTR_TO_PACKET_END;
11207                 return size == sizeof(__u64);
11208
11209         case offsetof(struct sk_reuseport_md, hash):
11210                 return size == size_default;
11211
11212         case offsetof(struct sk_reuseport_md, sk):
11213                 info->reg_type = PTR_TO_SOCKET;
11214                 return size == sizeof(__u64);
11215
11216         case offsetof(struct sk_reuseport_md, migrating_sk):
11217                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11218                 return size == sizeof(__u64);
11219
11220         /* Fields that allow narrowing */
11221         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11222                 if (size < sizeof_field(struct sk_buff, protocol))
11223                         return false;
11224                 fallthrough;
11225         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11226         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11227         case bpf_ctx_range(struct sk_reuseport_md, len):
11228                 bpf_ctx_record_field_size(info, size_default);
11229                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11230
11231         default:
11232                 return false;
11233         }
11234 }
11235
11236 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11237         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11238                               si->dst_reg, si->src_reg,                 \
11239                               bpf_target_off(struct sk_reuseport_kern, F, \
11240                                              sizeof_field(struct sk_reuseport_kern, F), \
11241                                              target_size));             \
11242         })
11243
11244 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11245         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11246                                     struct sk_buff,                     \
11247                                     skb,                                \
11248                                     SKB_FIELD)
11249
11250 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11251         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11252                                     struct sock,                        \
11253                                     sk,                                 \
11254                                     SK_FIELD)
11255
11256 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11257                                            const struct bpf_insn *si,
11258                                            struct bpf_insn *insn_buf,
11259                                            struct bpf_prog *prog,
11260                                            u32 *target_size)
11261 {
11262         struct bpf_insn *insn = insn_buf;
11263
11264         switch (si->off) {
11265         case offsetof(struct sk_reuseport_md, data):
11266                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11267                 break;
11268
11269         case offsetof(struct sk_reuseport_md, len):
11270                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11271                 break;
11272
11273         case offsetof(struct sk_reuseport_md, eth_protocol):
11274                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11275                 break;
11276
11277         case offsetof(struct sk_reuseport_md, ip_protocol):
11278                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11279                 break;
11280
11281         case offsetof(struct sk_reuseport_md, data_end):
11282                 SK_REUSEPORT_LOAD_FIELD(data_end);
11283                 break;
11284
11285         case offsetof(struct sk_reuseport_md, hash):
11286                 SK_REUSEPORT_LOAD_FIELD(hash);
11287                 break;
11288
11289         case offsetof(struct sk_reuseport_md, bind_inany):
11290                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11291                 break;
11292
11293         case offsetof(struct sk_reuseport_md, sk):
11294                 SK_REUSEPORT_LOAD_FIELD(sk);
11295                 break;
11296
11297         case offsetof(struct sk_reuseport_md, migrating_sk):
11298                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11299                 break;
11300         }
11301
11302         return insn - insn_buf;
11303 }
11304
11305 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11306         .get_func_proto         = sk_reuseport_func_proto,
11307         .is_valid_access        = sk_reuseport_is_valid_access,
11308         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11309 };
11310
11311 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11312 };
11313
11314 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11315 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11316
11317 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11318            struct sock *, sk, u64, flags)
11319 {
11320         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11321                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11322                 return -EINVAL;
11323         if (unlikely(sk && sk_is_refcounted(sk)))
11324                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11325         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11326                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11327         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11328                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11329
11330         /* Check if socket is suitable for packet L3/L4 protocol */
11331         if (sk && sk->sk_protocol != ctx->protocol)
11332                 return -EPROTOTYPE;
11333         if (sk && sk->sk_family != ctx->family &&
11334             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11335                 return -EAFNOSUPPORT;
11336
11337         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11338                 return -EEXIST;
11339
11340         /* Select socket as lookup result */
11341         ctx->selected_sk = sk;
11342         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11343         return 0;
11344 }
11345
11346 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11347         .func           = bpf_sk_lookup_assign,
11348         .gpl_only       = false,
11349         .ret_type       = RET_INTEGER,
11350         .arg1_type      = ARG_PTR_TO_CTX,
11351         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11352         .arg3_type      = ARG_ANYTHING,
11353 };
11354
11355 static const struct bpf_func_proto *
11356 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11357 {
11358         switch (func_id) {
11359         case BPF_FUNC_perf_event_output:
11360                 return &bpf_event_output_data_proto;
11361         case BPF_FUNC_sk_assign:
11362                 return &bpf_sk_lookup_assign_proto;
11363         case BPF_FUNC_sk_release:
11364                 return &bpf_sk_release_proto;
11365         default:
11366                 return bpf_sk_base_func_proto(func_id);
11367         }
11368 }
11369
11370 static bool sk_lookup_is_valid_access(int off, int size,
11371                                       enum bpf_access_type type,
11372                                       const struct bpf_prog *prog,
11373                                       struct bpf_insn_access_aux *info)
11374 {
11375         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11376                 return false;
11377         if (off % size != 0)
11378                 return false;
11379         if (type != BPF_READ)
11380                 return false;
11381
11382         switch (off) {
11383         case offsetof(struct bpf_sk_lookup, sk):
11384                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11385                 return size == sizeof(__u64);
11386
11387         case bpf_ctx_range(struct bpf_sk_lookup, family):
11388         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11389         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11390         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11391         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11392         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11393         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11394         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11395                 bpf_ctx_record_field_size(info, sizeof(__u32));
11396                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11397
11398         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11399                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11400                 if (size == sizeof(__u32))
11401                         return true;
11402                 bpf_ctx_record_field_size(info, sizeof(__be16));
11403                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11404
11405         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11406              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11407                 /* Allow access to zero padding for backward compatibility */
11408                 bpf_ctx_record_field_size(info, sizeof(__u16));
11409                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11410
11411         default:
11412                 return false;
11413         }
11414 }
11415
11416 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11417                                         const struct bpf_insn *si,
11418                                         struct bpf_insn *insn_buf,
11419                                         struct bpf_prog *prog,
11420                                         u32 *target_size)
11421 {
11422         struct bpf_insn *insn = insn_buf;
11423
11424         switch (si->off) {
11425         case offsetof(struct bpf_sk_lookup, sk):
11426                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11427                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11428                 break;
11429
11430         case offsetof(struct bpf_sk_lookup, family):
11431                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11432                                       bpf_target_off(struct bpf_sk_lookup_kern,
11433                                                      family, 2, target_size));
11434                 break;
11435
11436         case offsetof(struct bpf_sk_lookup, protocol):
11437                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11438                                       bpf_target_off(struct bpf_sk_lookup_kern,
11439                                                      protocol, 2, target_size));
11440                 break;
11441
11442         case offsetof(struct bpf_sk_lookup, remote_ip4):
11443                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11444                                       bpf_target_off(struct bpf_sk_lookup_kern,
11445                                                      v4.saddr, 4, target_size));
11446                 break;
11447
11448         case offsetof(struct bpf_sk_lookup, local_ip4):
11449                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11450                                       bpf_target_off(struct bpf_sk_lookup_kern,
11451                                                      v4.daddr, 4, target_size));
11452                 break;
11453
11454         case bpf_ctx_range_till(struct bpf_sk_lookup,
11455                                 remote_ip6[0], remote_ip6[3]): {
11456 #if IS_ENABLED(CONFIG_IPV6)
11457                 int off = si->off;
11458
11459                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11460                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11461                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11462                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11463                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11464                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11465 #else
11466                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11467 #endif
11468                 break;
11469         }
11470         case bpf_ctx_range_till(struct bpf_sk_lookup,
11471                                 local_ip6[0], local_ip6[3]): {
11472 #if IS_ENABLED(CONFIG_IPV6)
11473                 int off = si->off;
11474
11475                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11476                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11477                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11478                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11479                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11480                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11481 #else
11482                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11483 #endif
11484                 break;
11485         }
11486         case offsetof(struct bpf_sk_lookup, remote_port):
11487                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11488                                       bpf_target_off(struct bpf_sk_lookup_kern,
11489                                                      sport, 2, target_size));
11490                 break;
11491
11492         case offsetofend(struct bpf_sk_lookup, remote_port):
11493                 *target_size = 2;
11494                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11495                 break;
11496
11497         case offsetof(struct bpf_sk_lookup, local_port):
11498                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11499                                       bpf_target_off(struct bpf_sk_lookup_kern,
11500                                                      dport, 2, target_size));
11501                 break;
11502
11503         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11504                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11505                                       bpf_target_off(struct bpf_sk_lookup_kern,
11506                                                      ingress_ifindex, 4, target_size));
11507                 break;
11508         }
11509
11510         return insn - insn_buf;
11511 }
11512
11513 const struct bpf_prog_ops sk_lookup_prog_ops = {
11514         .test_run = bpf_prog_test_run_sk_lookup,
11515 };
11516
11517 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11518         .get_func_proto         = sk_lookup_func_proto,
11519         .is_valid_access        = sk_lookup_is_valid_access,
11520         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11521 };
11522
11523 #endif /* CONFIG_INET */
11524
11525 DEFINE_BPF_DISPATCHER(xdp)
11526
11527 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11528 {
11529         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11530 }
11531
11532 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11533 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11534 BTF_SOCK_TYPE_xxx
11535 #undef BTF_SOCK_TYPE
11536
11537 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11538 {
11539         /* tcp6_sock type is not generated in dwarf and hence btf,
11540          * trigger an explicit type generation here.
11541          */
11542         BTF_TYPE_EMIT(struct tcp6_sock);
11543         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11544             sk->sk_family == AF_INET6)
11545                 return (unsigned long)sk;
11546
11547         return (unsigned long)NULL;
11548 }
11549
11550 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11551         .func                   = bpf_skc_to_tcp6_sock,
11552         .gpl_only               = false,
11553         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11554         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11555         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11556 };
11557
11558 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11559 {
11560         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11561                 return (unsigned long)sk;
11562
11563         return (unsigned long)NULL;
11564 }
11565
11566 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11567         .func                   = bpf_skc_to_tcp_sock,
11568         .gpl_only               = false,
11569         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11570         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11571         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11572 };
11573
11574 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11575 {
11576         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11577          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11578          */
11579         BTF_TYPE_EMIT(struct inet_timewait_sock);
11580         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11581
11582 #ifdef CONFIG_INET
11583         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11584                 return (unsigned long)sk;
11585 #endif
11586
11587 #if IS_BUILTIN(CONFIG_IPV6)
11588         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11589                 return (unsigned long)sk;
11590 #endif
11591
11592         return (unsigned long)NULL;
11593 }
11594
11595 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11596         .func                   = bpf_skc_to_tcp_timewait_sock,
11597         .gpl_only               = false,
11598         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11599         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11600         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11601 };
11602
11603 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11604 {
11605 #ifdef CONFIG_INET
11606         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11607                 return (unsigned long)sk;
11608 #endif
11609
11610 #if IS_BUILTIN(CONFIG_IPV6)
11611         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11612                 return (unsigned long)sk;
11613 #endif
11614
11615         return (unsigned long)NULL;
11616 }
11617
11618 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11619         .func                   = bpf_skc_to_tcp_request_sock,
11620         .gpl_only               = false,
11621         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11622         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11623         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11624 };
11625
11626 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11627 {
11628         /* udp6_sock type is not generated in dwarf and hence btf,
11629          * trigger an explicit type generation here.
11630          */
11631         BTF_TYPE_EMIT(struct udp6_sock);
11632         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11633             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11634                 return (unsigned long)sk;
11635
11636         return (unsigned long)NULL;
11637 }
11638
11639 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11640         .func                   = bpf_skc_to_udp6_sock,
11641         .gpl_only               = false,
11642         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11643         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11644         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11645 };
11646
11647 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11648 {
11649         /* unix_sock type is not generated in dwarf and hence btf,
11650          * trigger an explicit type generation here.
11651          */
11652         BTF_TYPE_EMIT(struct unix_sock);
11653         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11654                 return (unsigned long)sk;
11655
11656         return (unsigned long)NULL;
11657 }
11658
11659 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11660         .func                   = bpf_skc_to_unix_sock,
11661         .gpl_only               = false,
11662         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11663         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11664         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11665 };
11666
11667 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11668 {
11669         BTF_TYPE_EMIT(struct mptcp_sock);
11670         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11671 }
11672
11673 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11674         .func           = bpf_skc_to_mptcp_sock,
11675         .gpl_only       = false,
11676         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11677         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11678         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11679 };
11680
11681 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11682 {
11683         return (unsigned long)sock_from_file(file);
11684 }
11685
11686 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11687 BTF_ID(struct, socket)
11688 BTF_ID(struct, file)
11689
11690 const struct bpf_func_proto bpf_sock_from_file_proto = {
11691         .func           = bpf_sock_from_file,
11692         .gpl_only       = false,
11693         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11694         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11695         .arg1_type      = ARG_PTR_TO_BTF_ID,
11696         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11697 };
11698
11699 static const struct bpf_func_proto *
11700 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11701 {
11702         const struct bpf_func_proto *func;
11703
11704         switch (func_id) {
11705         case BPF_FUNC_skc_to_tcp6_sock:
11706                 func = &bpf_skc_to_tcp6_sock_proto;
11707                 break;
11708         case BPF_FUNC_skc_to_tcp_sock:
11709                 func = &bpf_skc_to_tcp_sock_proto;
11710                 break;
11711         case BPF_FUNC_skc_to_tcp_timewait_sock:
11712                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11713                 break;
11714         case BPF_FUNC_skc_to_tcp_request_sock:
11715                 func = &bpf_skc_to_tcp_request_sock_proto;
11716                 break;
11717         case BPF_FUNC_skc_to_udp6_sock:
11718                 func = &bpf_skc_to_udp6_sock_proto;
11719                 break;
11720         case BPF_FUNC_skc_to_unix_sock:
11721                 func = &bpf_skc_to_unix_sock_proto;
11722                 break;
11723         case BPF_FUNC_skc_to_mptcp_sock:
11724                 func = &bpf_skc_to_mptcp_sock_proto;
11725                 break;
11726         case BPF_FUNC_ktime_get_coarse_ns:
11727                 return &bpf_ktime_get_coarse_ns_proto;
11728         default:
11729                 return bpf_base_func_proto(func_id);
11730         }
11731
11732         if (!perfmon_capable())
11733                 return NULL;
11734
11735         return func;
11736 }
11737
11738 __diag_push();
11739 __diag_ignore_all("-Wmissing-prototypes",
11740                   "Global functions as their definitions will be in vmlinux BTF");
11741 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags,
11742                                     struct bpf_dynptr_kern *ptr__uninit)
11743 {
11744         if (flags) {
11745                 bpf_dynptr_set_null(ptr__uninit);
11746                 return -EINVAL;
11747         }
11748
11749         bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len);
11750
11751         return 0;
11752 }
11753
11754 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags,
11755                                     struct bpf_dynptr_kern *ptr__uninit)
11756 {
11757         if (flags) {
11758                 bpf_dynptr_set_null(ptr__uninit);
11759                 return -EINVAL;
11760         }
11761
11762         bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp));
11763
11764         return 0;
11765 }
11766 __diag_pop();
11767
11768 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags,
11769                                struct bpf_dynptr_kern *ptr__uninit)
11770 {
11771         int err;
11772
11773         err = bpf_dynptr_from_skb(skb, flags, ptr__uninit);
11774         if (err)
11775                 return err;
11776
11777         bpf_dynptr_set_rdonly(ptr__uninit);
11778
11779         return 0;
11780 }
11781
11782 BTF_SET8_START(bpf_kfunc_check_set_skb)
11783 BTF_ID_FLAGS(func, bpf_dynptr_from_skb)
11784 BTF_SET8_END(bpf_kfunc_check_set_skb)
11785
11786 BTF_SET8_START(bpf_kfunc_check_set_xdp)
11787 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp)
11788 BTF_SET8_END(bpf_kfunc_check_set_xdp)
11789
11790 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = {
11791         .owner = THIS_MODULE,
11792         .set = &bpf_kfunc_check_set_skb,
11793 };
11794
11795 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = {
11796         .owner = THIS_MODULE,
11797         .set = &bpf_kfunc_check_set_xdp,
11798 };
11799
11800 static int __init bpf_kfunc_init(void)
11801 {
11802         int ret;
11803
11804         ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb);
11805         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb);
11806         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb);
11807         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb);
11808         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb);
11809         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb);
11810         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb);
11811         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb);
11812         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb);
11813         ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb);
11814         return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp);
11815 }
11816 late_initcall(bpf_kfunc_init);
11817
11818 /* Disables missing prototype warnings */
11819 __diag_push();
11820 __diag_ignore_all("-Wmissing-prototypes",
11821                   "Global functions as their definitions will be in vmlinux BTF");
11822
11823 /* bpf_sock_destroy: Destroy the given socket with ECONNABORTED error code.
11824  *
11825  * The function expects a non-NULL pointer to a socket, and invokes the
11826  * protocol specific socket destroy handlers.
11827  *
11828  * The helper can only be called from BPF contexts that have acquired the socket
11829  * locks.
11830  *
11831  * Parameters:
11832  * @sock: Pointer to socket to be destroyed
11833  *
11834  * Return:
11835  * On error, may return EPROTONOSUPPORT, EINVAL.
11836  * EPROTONOSUPPORT if protocol specific destroy handler is not supported.
11837  * 0 otherwise
11838  */
11839 __bpf_kfunc int bpf_sock_destroy(struct sock_common *sock)
11840 {
11841         struct sock *sk = (struct sock *)sock;
11842
11843         /* The locking semantics that allow for synchronous execution of the
11844          * destroy handlers are only supported for TCP and UDP.
11845          * Supporting protocols will need to acquire sock lock in the BPF context
11846          * prior to invoking this kfunc.
11847          */
11848         if (!sk->sk_prot->diag_destroy || (sk->sk_protocol != IPPROTO_TCP &&
11849                                            sk->sk_protocol != IPPROTO_UDP))
11850                 return -EOPNOTSUPP;
11851
11852         return sk->sk_prot->diag_destroy(sk, ECONNABORTED);
11853 }
11854
11855 __diag_pop()
11856
11857 BTF_SET8_START(bpf_sk_iter_kfunc_ids)
11858 BTF_ID_FLAGS(func, bpf_sock_destroy, KF_TRUSTED_ARGS)
11859 BTF_SET8_END(bpf_sk_iter_kfunc_ids)
11860
11861 static int tracing_iter_filter(const struct bpf_prog *prog, u32 kfunc_id)
11862 {
11863         if (btf_id_set8_contains(&bpf_sk_iter_kfunc_ids, kfunc_id) &&
11864             prog->expected_attach_type != BPF_TRACE_ITER)
11865                 return -EACCES;
11866         return 0;
11867 }
11868
11869 static const struct btf_kfunc_id_set bpf_sk_iter_kfunc_set = {
11870         .owner = THIS_MODULE,
11871         .set   = &bpf_sk_iter_kfunc_ids,
11872         .filter = tracing_iter_filter,
11873 };
11874
11875 static int init_subsystem(void)
11876 {
11877         return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_sk_iter_kfunc_set);
11878 }
11879 late_initcall(init_subsystem);