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