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