net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/atomic.h>
  21 #include <linux/bpf_verifier.h>
  22 #include <linux/module.h>
  23 #include <linux/types.h>
  24 #include <linux/mm.h>
  25 #include <linux/fcntl.h>
  26 #include <linux/socket.h>
  27 #include <linux/sock_diag.h>
  28 #include <linux/in.h>
  29 #include <linux/inet.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/if_packet.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/gfp.h>
  34 #include <net/inet_common.h>
  35 #include <net/ip.h>
  36 #include <net/protocol.h>
  37 #include <net/netlink.h>
  38 #include <linux/skbuff.h>
  39 #include <linux/skmsg.h>
  40 #include <net/sock.h>
  41 #include <net/flow_dissector.h>
  42 #include <linux/errno.h>
  43 #include <linux/timer.h>
  44 #include <linux/uaccess.h>
  45 #include <asm/unaligned.h>
  46 #include <linux/filter.h>
  47 #include <linux/ratelimit.h>
  48 #include <linux/seccomp.h>
  49 #include <linux/if_vlan.h>
  50 #include <linux/bpf.h>
  51 #include <linux/btf.h>
  52 #include <net/sch_generic.h>
  53 #include <net/cls_cgroup.h>
  54 #include <net/dst_metadata.h>
  55 #include <net/dst.h>
  56 #include <net/sock_reuseport.h>
  57 #include <net/busy_poll.h>
  58 #include <net/tcp.h>
  59 #include <net/xfrm.h>
  60 #include <net/udp.h>
  61 #include <linux/bpf_trace.h>
  62 #include <net/xdp_sock.h>
  63 #include <linux/inetdevice.h>
  64 #include <net/inet_hashtables.h>
  65 #include <net/inet6_hashtables.h>
  66 #include <net/ip_fib.h>
  67 #include <net/nexthop.h>
  68 #include <net/flow.h>
  69 #include <net/arp.h>
  70 #include <net/ipv6.h>
  71 #include <net/net_namespace.h>
  72 #include <linux/seg6_local.h>
  73 #include <net/seg6.h>
  74 #include <net/seg6_local.h>
  75 #include <net/lwtunnel.h>
  76 #include <net/ipv6_stubs.h>
  77 #include <net/bpf_sk_storage.h>
  78 #include <net/transp_v6.h>
  79 #include <linux/btf_ids.h>
  80 #include <net/tls.h>
  81 #include <net/xdp.h>
  82 #include <net/mptcp.h>
  83 #include <net/netfilter/nf_conntrack_bpf.h>
  84
  85 static const struct bpf_func_proto *
  86 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  87
  88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  89 {
  90         if (in_compat_syscall()) {
  91                 struct compat_sock_fprog f32;
  92
  93                 if (len != sizeof(f32))
  94                         return -EINVAL;
  95                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  96                         return -EFAULT;
  97                 memset(dst, 0, sizeof(*dst));
  98                 dst->len = f32.len;
  99                 dst->filter = compat_ptr(f32.filter);
 100         } else {
 101                 if (len != sizeof(*dst))
 102                         return -EINVAL;
 103                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 104                         return -EFAULT;
 105         }
 106
 107         return 0;
 108 }
 109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 110
 111 /**
 112  *      sk_filter_trim_cap - run a packet through a socket filter
 113  *      @sk: sock associated with &sk_buff
 114  *      @skb: buffer to filter
 115  *      @cap: limit on how short the eBPF program may trim the packet
 116  *
 117  * Run the eBPF program and then cut skb->data to correct size returned by
 118  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 119  * than pkt_len we keep whole skb->data. This is the socket level
 120  * wrapper to bpf_prog_run. It returns 0 if the packet should
 121  * be accepted or -EPERM if the packet should be tossed.
 122  *
 123  */
 124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 125 {
 126         int err;
 127         struct sk_filter *filter;
 128
 129         /*
 130          * If the skb was allocated from pfmemalloc reserves, only
 131          * allow SOCK_MEMALLOC sockets to use it as this socket is
 132          * helping free memory
 133          */
 134         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 135                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 136                 return -ENOMEM;
 137         }
 138         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 139         if (err)
 140                 return err;
 141
 142         err = security_sock_rcv_skb(sk, skb);
 143         if (err)
 144                 return err;
 145
 146         rcu_read_lock();
 147         filter = rcu_dereference(sk->sk_filter);
 148         if (filter) {
 149                 struct sock *save_sk = skb->sk;
 150                 unsigned int pkt_len;
 151
 152                 skb->sk = sk;
 153                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 154                 skb->sk = save_sk;
 155                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 156         }
 157         rcu_read_unlock();
 158
 159         return err;
 160 }
 161 EXPORT_SYMBOL(sk_filter_trim_cap);
 162
 163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 164 {
 165         return skb_get_poff(skb);
 166 }
 167
 168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 169 {
 170         struct nlattr *nla;
 171
 172         if (skb_is_nonlinear(skb))
 173                 return 0;
 174
 175         if (skb->len < sizeof(struct nlattr))
 176                 return 0;
 177
 178         if (a > skb->len - sizeof(struct nlattr))
 179                 return 0;
 180
 181         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 182         if (nla)
 183                 return (void *) nla - (void *) skb->data;
 184
 185         return 0;
 186 }
 187
 188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 189 {
 190         struct nlattr *nla;
 191
 192         if (skb_is_nonlinear(skb))
 193                 return 0;
 194
 195         if (skb->len < sizeof(struct nlattr))
 196                 return 0;
 197
 198         if (a > skb->len - sizeof(struct nlattr))
 199                 return 0;
 200
 201         nla = (struct nlattr *) &skb->data[a];
 202         if (nla->nla_len > skb->len - a)
 203                 return 0;
 204
 205         nla = nla_find_nested(nla, x);
 206         if (nla)
 207                 return (void *) nla - (void *) skb->data;
 208
 209         return 0;
 210 }
 211
 212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 213            data, int, headlen, int, offset)
 214 {
 215         u8 tmp, *ptr;
 216         const int len = sizeof(tmp);
 217
 218         if (offset >= 0) {
 219                 if (headlen - offset >= len)
 220                         return *(u8 *)(data + offset);
 221                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 222                         return tmp;
 223         } else {
 224                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 225                 if (likely(ptr))
 226                         return *(u8 *)ptr;
 227         }
 228
 229         return -EFAULT;
 230 }
 231
 232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 233            int, offset)
 234 {
 235         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 236                                          offset);
 237 }
 238
 239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 240            data, int, headlen, int, offset)
 241 {
 242         __be16 tmp, *ptr;
 243         const int len = sizeof(tmp);
 244
 245         if (offset >= 0) {
 246                 if (headlen - offset >= len)
 247                         return get_unaligned_be16(data + offset);
 248                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 249                         return be16_to_cpu(tmp);
 250         } else {
 251                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 252                 if (likely(ptr))
 253                         return get_unaligned_be16(ptr);
 254         }
 255
 256         return -EFAULT;
 257 }
 258
 259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 260            int, offset)
 261 {
 262         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 263                                           offset);
 264 }
 265
 266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 267            data, int, headlen, int, offset)
 268 {
 269         __be32 tmp, *ptr;
 270         const int len = sizeof(tmp);
 271
 272         if (likely(offset >= 0)) {
 273                 if (headlen - offset >= len)
 274                         return get_unaligned_be32(data + offset);
 275                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 276                         return be32_to_cpu(tmp);
 277         } else {
 278                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 279                 if (likely(ptr))
 280                         return get_unaligned_be32(ptr);
 281         }
 282
 283         return -EFAULT;
 284 }
 285
 286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 287            int, offset)
 288 {
 289         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 290                                           offset);
 291 }
 292
 293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 294                               struct bpf_insn *insn_buf)
 295 {
 296         struct bpf_insn *insn = insn_buf;
 297
 298         switch (skb_field) {
 299         case SKF_AD_MARK:
 300                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 301
 302                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 303                                       offsetof(struct sk_buff, mark));
 304                 break;
 305
 306         case SKF_AD_PKTTYPE:
 307                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
 308                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 309 #ifdef __BIG_ENDIAN_BITFIELD
 310                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 311 #endif
 312                 break;
 313
 314         case SKF_AD_QUEUE:
 315                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 316
 317                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 318                                       offsetof(struct sk_buff, queue_mapping));
 319                 break;
 320
 321         case SKF_AD_VLAN_TAG:
 322                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 323
 324                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 325                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 326                                       offsetof(struct sk_buff, vlan_tci));
 327                 break;
 328         case SKF_AD_VLAN_TAG_PRESENT:
 329                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4);
 330                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 331                                       offsetof(struct sk_buff, vlan_all));
 332                 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1);
 333                 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1);
 334                 break;
 335         }
 336
 337         return insn - insn_buf;
 338 }
 339
 340 static bool convert_bpf_extensions(struct sock_filter *fp,
 341                                    struct bpf_insn **insnp)
 342 {
 343         struct bpf_insn *insn = *insnp;
 344         u32 cnt;
 345
 346         switch (fp->k) {
 347         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 348                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 349
 350                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 351                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 352                                       offsetof(struct sk_buff, protocol));
 353                 /* A = ntohs(A) [emitting a nop or swap16] */
 354                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 355                 break;
 356
 357         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 358                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 359                 insn += cnt - 1;
 360                 break;
 361
 362         case SKF_AD_OFF + SKF_AD_IFINDEX:
 363         case SKF_AD_OFF + SKF_AD_HATYPE:
 364                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 365                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 366
 367                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 368                                       BPF_REG_TMP, BPF_REG_CTX,
 369                                       offsetof(struct sk_buff, dev));
 370                 /* if (tmp != 0) goto pc + 1 */
 371                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 372                 *insn++ = BPF_EXIT_INSN();
 373                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 374                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 375                                             offsetof(struct net_device, ifindex));
 376                 else
 377                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 378                                             offsetof(struct net_device, type));
 379                 break;
 380
 381         case SKF_AD_OFF + SKF_AD_MARK:
 382                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 383                 insn += cnt - 1;
 384                 break;
 385
 386         case SKF_AD_OFF + SKF_AD_RXHASH:
 387                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 388
 389                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 390                                     offsetof(struct sk_buff, hash));
 391                 break;
 392
 393         case SKF_AD_OFF + SKF_AD_QUEUE:
 394                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 395                 insn += cnt - 1;
 396                 break;
 397
 398         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 399                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 400                                          BPF_REG_A, BPF_REG_CTX, insn);
 401                 insn += cnt - 1;
 402                 break;
 403
 404         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 405                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 406                                          BPF_REG_A, BPF_REG_CTX, insn);
 407                 insn += cnt - 1;
 408                 break;
 409
 410         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 411                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 412
 413                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 414                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 415                                       offsetof(struct sk_buff, vlan_proto));
 416                 /* A = ntohs(A) [emitting a nop or swap16] */
 417                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 418                 break;
 419
 420         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 421         case SKF_AD_OFF + SKF_AD_NLATTR:
 422         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 423         case SKF_AD_OFF + SKF_AD_CPU:
 424         case SKF_AD_OFF + SKF_AD_RANDOM:
 425                 /* arg1 = CTX */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 427                 /* arg2 = A */
 428                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 429                 /* arg3 = X */
 430                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 431                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 432                 switch (fp->k) {
 433                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 434                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 435                         break;
 436                 case SKF_AD_OFF + SKF_AD_NLATTR:
 437                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 438                         break;
 439                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 440                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 441                         break;
 442                 case SKF_AD_OFF + SKF_AD_CPU:
 443                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 444                         break;
 445                 case SKF_AD_OFF + SKF_AD_RANDOM:
 446                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 447                         bpf_user_rnd_init_once();
 448                         break;
 449                 }
 450                 break;
 451
 452         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 453                 /* A ^= X */
 454                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 455                 break;
 456
 457         default:
 458                 /* This is just a dummy call to avoid letting the compiler
 459                  * evict __bpf_call_base() as an optimization. Placed here
 460                  * where no-one bothers.
 461                  */
 462                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 463                 return false;
 464         }
 465
 466         *insnp = insn;
 467         return true;
 468 }
 469
 470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 471 {
 472         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 473         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 474         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 475                       BPF_SIZE(fp->code) == BPF_W;
 476         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 477         const int ip_align = NET_IP_ALIGN;
 478         struct bpf_insn *insn = *insnp;
 479         int offset = fp->k;
 480
 481         if (!indirect &&
 482             ((unaligned_ok && offset >= 0) ||
 483              (!unaligned_ok && offset >= 0 &&
 484               offset + ip_align >= 0 &&
 485               offset + ip_align % size == 0))) {
 486                 bool ldx_off_ok = offset <= S16_MAX;
 487
 488                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 489                 if (offset)
 490                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 491                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 492                                       size, 2 + endian + (!ldx_off_ok * 2));
 493                 if (ldx_off_ok) {
 494                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 495                                               BPF_REG_D, offset);
 496                 } else {
 497                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 498                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 499                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 500                                               BPF_REG_TMP, 0);
 501                 }
 502                 if (endian)
 503                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 504                 *insn++ = BPF_JMP_A(8);
 505         }
 506
 507         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 508         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 509         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 510         if (!indirect) {
 511                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 512         } else {
 513                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 514                 if (fp->k)
 515                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 516         }
 517
 518         switch (BPF_SIZE(fp->code)) {
 519         case BPF_B:
 520                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 521                 break;
 522         case BPF_H:
 523                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 524                 break;
 525         case BPF_W:
 526                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 527                 break;
 528         default:
 529                 return false;
 530         }
 531
 532         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 533         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 534         *insn   = BPF_EXIT_INSN();
 535
 536         *insnp = insn;
 537         return true;
 538 }
 539
 540 /**
 541  *      bpf_convert_filter - convert filter program
 542  *      @prog: the user passed filter program
 543  *      @len: the length of the user passed filter program
 544  *      @new_prog: allocated 'struct bpf_prog' or NULL
 545  *      @new_len: pointer to store length of converted program
 546  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 547  *
 548  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 549  * style extended BPF (eBPF).
 550  * Conversion workflow:
 551  *
 552  * 1) First pass for calculating the new program length:
 553  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 554  *
 555  * 2) 2nd pass to remap in two passes: 1st pass finds new
 556  *    jump offsets, 2nd pass remapping:
 557  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 558  */
 559 static int bpf_convert_filter(struct sock_filter *prog, int len,
 560                               struct bpf_prog *new_prog, int *new_len,
 561                               bool *seen_ld_abs)
 562 {
 563         int new_flen = 0, pass = 0, target, i, stack_off;
 564         struct bpf_insn *new_insn, *first_insn = NULL;
 565         struct sock_filter *fp;
 566         int *addrs = NULL;
 567         u8 bpf_src;
 568
 569         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 570         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 571
 572         if (len <= 0 || len > BPF_MAXINSNS)
 573                 return -EINVAL;
 574
 575         if (new_prog) {
 576                 first_insn = new_prog->insnsi;
 577                 addrs = kcalloc(len, sizeof(*addrs),
 578                                 GFP_KERNEL | __GFP_NOWARN);
 579                 if (!addrs)
 580                         return -ENOMEM;
 581         }
 582
 583 do_pass:
 584         new_insn = first_insn;
 585         fp = prog;
 586
 587         /* Classic BPF related prologue emission. */
 588         if (new_prog) {
 589                 /* Classic BPF expects A and X to be reset first. These need
 590                  * to be guaranteed to be the first two instructions.
 591                  */
 592                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 593                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 594
 595                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 596                  * In eBPF case it's done by the compiler, here we need to
 597                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 598                  */
 599                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 600                 if (*seen_ld_abs) {
 601                         /* For packet access in classic BPF, cache skb->data
 602                          * in callee-saved BPF R8 and skb->len - skb->data_len
 603                          * (headlen) in BPF R9. Since classic BPF is read-only
 604                          * on CTX, we only need to cache it once.
 605                          */
 606                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 607                                                   BPF_REG_D, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, data));
 609                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 610                                                   offsetof(struct sk_buff, len));
 611                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 612                                                   offsetof(struct sk_buff, data_len));
 613                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 614                 }
 615         } else {
 616                 new_insn += 3;
 617         }
 618
 619         for (i = 0; i < len; fp++, i++) {
 620                 struct bpf_insn tmp_insns[32] = { };
 621                 struct bpf_insn *insn = tmp_insns;
 622
 623                 if (addrs)
 624                         addrs[i] = new_insn - first_insn;
 625
 626                 switch (fp->code) {
 627                 /* All arithmetic insns and skb loads map as-is. */
 628                 case BPF_ALU | BPF_ADD | BPF_X:
 629                 case BPF_ALU | BPF_ADD | BPF_K:
 630                 case BPF_ALU | BPF_SUB | BPF_X:
 631                 case BPF_ALU | BPF_SUB | BPF_K:
 632                 case BPF_ALU | BPF_AND | BPF_X:
 633                 case BPF_ALU | BPF_AND | BPF_K:
 634                 case BPF_ALU | BPF_OR | BPF_X:
 635                 case BPF_ALU | BPF_OR | BPF_K:
 636                 case BPF_ALU | BPF_LSH | BPF_X:
 637                 case BPF_ALU | BPF_LSH | BPF_K:
 638                 case BPF_ALU | BPF_RSH | BPF_X:
 639                 case BPF_ALU | BPF_RSH | BPF_K:
 640                 case BPF_ALU | BPF_XOR | BPF_X:
 641                 case BPF_ALU | BPF_XOR | BPF_K:
 642                 case BPF_ALU | BPF_MUL | BPF_X:
 643                 case BPF_ALU | BPF_MUL | BPF_K:
 644                 case BPF_ALU | BPF_DIV | BPF_X:
 645                 case BPF_ALU | BPF_DIV | BPF_K:
 646                 case BPF_ALU | BPF_MOD | BPF_X:
 647                 case BPF_ALU | BPF_MOD | BPF_K:
 648                 case BPF_ALU | BPF_NEG:
 649                 case BPF_LD | BPF_ABS | BPF_W:
 650                 case BPF_LD | BPF_ABS | BPF_H:
 651                 case BPF_LD | BPF_ABS | BPF_B:
 652                 case BPF_LD | BPF_IND | BPF_W:
 653                 case BPF_LD | BPF_IND | BPF_H:
 654                 case BPF_LD | BPF_IND | BPF_B:
 655                         /* Check for overloaded BPF extension and
 656                          * directly convert it if found, otherwise
 657                          * just move on with mapping.
 658                          */
 659                         if (BPF_CLASS(fp->code) == BPF_LD &&
 660                             BPF_MODE(fp->code) == BPF_ABS &&
 661                             convert_bpf_extensions(fp, &insn))
 662                                 break;
 663                         if (BPF_CLASS(fp->code) == BPF_LD &&
 664                             convert_bpf_ld_abs(fp, &insn)) {
 665                                 *seen_ld_abs = true;
 666                                 break;
 667                         }
 668
 669                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 670                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 671                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 672                                 /* Error with exception code on div/mod by 0.
 673                                  * For cBPF programs, this was always return 0.
 674                                  */
 675                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 676                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 677                                 *insn++ = BPF_EXIT_INSN();
 678                         }
 679
 680                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 681                         break;
 682
 683                 /* Jump transformation cannot use BPF block macros
 684                  * everywhere as offset calculation and target updates
 685                  * require a bit more work than the rest, i.e. jump
 686                  * opcodes map as-is, but offsets need adjustment.
 687                  */
 688
 689 #define BPF_EMIT_JMP                                                    \
 690         do {                                                            \
 691                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 692                 s32 off;                                                \
 693                                                                         \
 694                 if (target >= len || target < 0)                        \
 695                         goto err;                                       \
 696                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 697                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 698                 off -= insn - tmp_insns;                                \
 699                 /* Reject anything not fitting into insn->off. */       \
 700                 if (off < off_min || off > off_max)                     \
 701                         goto err;                                       \
 702                 insn->off = off;                                        \
 703         } while (0)
 704
 705                 case BPF_JMP | BPF_JA:
 706                         target = i + fp->k + 1;
 707                         insn->code = fp->code;
 708                         BPF_EMIT_JMP;
 709                         break;
 710
 711                 case BPF_JMP | BPF_JEQ | BPF_K:
 712                 case BPF_JMP | BPF_JEQ | BPF_X:
 713                 case BPF_JMP | BPF_JSET | BPF_K:
 714                 case BPF_JMP | BPF_JSET | BPF_X:
 715                 case BPF_JMP | BPF_JGT | BPF_K:
 716                 case BPF_JMP | BPF_JGT | BPF_X:
 717                 case BPF_JMP | BPF_JGE | BPF_K:
 718                 case BPF_JMP | BPF_JGE | BPF_X:
 719                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 720                                 /* BPF immediates are signed, zero extend
 721                                  * immediate into tmp register and use it
 722                                  * in compare insn.
 723                                  */
 724                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 725
 726                                 insn->dst_reg = BPF_REG_A;
 727                                 insn->src_reg = BPF_REG_TMP;
 728                                 bpf_src = BPF_X;
 729                         } else {
 730                                 insn->dst_reg = BPF_REG_A;
 731                                 insn->imm = fp->k;
 732                                 bpf_src = BPF_SRC(fp->code);
 733                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 734                         }
 735
 736                         /* Common case where 'jump_false' is next insn. */
 737                         if (fp->jf == 0) {
 738                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 739                                 target = i + fp->jt + 1;
 740                                 BPF_EMIT_JMP;
 741                                 break;
 742                         }
 743
 744                         /* Convert some jumps when 'jump_true' is next insn. */
 745                         if (fp->jt == 0) {
 746                                 switch (BPF_OP(fp->code)) {
 747                                 case BPF_JEQ:
 748                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 749                                         break;
 750                                 case BPF_JGT:
 751                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 752                                         break;
 753                                 case BPF_JGE:
 754                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 755                                         break;
 756                                 default:
 757                                         goto jmp_rest;
 758                                 }
 759
 760                                 target = i + fp->jf + 1;
 761                                 BPF_EMIT_JMP;
 762                                 break;
 763                         }
 764 jmp_rest:
 765                         /* Other jumps are mapped into two insns: Jxx and JA. */
 766                         target = i + fp->jt + 1;
 767                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 768                         BPF_EMIT_JMP;
 769                         insn++;
 770
 771                         insn->code = BPF_JMP | BPF_JA;
 772                         target = i + fp->jf + 1;
 773                         BPF_EMIT_JMP;
 774                         break;
 775
 776                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 777                 case BPF_LDX | BPF_MSH | BPF_B: {
 778                         struct sock_filter tmp = {
 779                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 780                                 .k      = fp->k,
 781                         };
 782
 783                         *seen_ld_abs = true;
 784
 785                         /* X = A */
 786                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 787                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 788                         convert_bpf_ld_abs(&tmp, &insn);
 789                         insn++;
 790                         /* A &= 0xf */
 791                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 792                         /* A <<= 2 */
 793                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 794                         /* tmp = X */
 795                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 796                         /* X = A */
 797                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 798                         /* A = tmp */
 799                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 800                         break;
 801                 }
 802                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 803                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 804                  */
 805                 case BPF_RET | BPF_A:
 806                 case BPF_RET | BPF_K:
 807                         if (BPF_RVAL(fp->code) == BPF_K)
 808                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 809                                                         0, fp->k);
 810                         *insn = BPF_EXIT_INSN();
 811                         break;
 812
 813                 /* Store to stack. */
 814                 case BPF_ST:
 815                 case BPF_STX:
 816                         stack_off = fp->k * 4  + 4;
 817                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 818                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 819                                             -stack_off);
 820                         /* check_load_and_stores() verifies that classic BPF can
 821                          * load from stack only after write, so tracking
 822                          * stack_depth for ST|STX insns is enough
 823                          */
 824                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 825                                 new_prog->aux->stack_depth = stack_off;
 826                         break;
 827
 828                 /* Load from stack. */
 829                 case BPF_LD | BPF_MEM:
 830                 case BPF_LDX | BPF_MEM:
 831                         stack_off = fp->k * 4  + 4;
 832                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 833                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 834                                             -stack_off);
 835                         break;
 836
 837                 /* A = K or X = K */
 838                 case BPF_LD | BPF_IMM:
 839                 case BPF_LDX | BPF_IMM:
 840                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 841                                               BPF_REG_A : BPF_REG_X, fp->k);
 842                         break;
 843
 844                 /* X = A */
 845                 case BPF_MISC | BPF_TAX:
 846                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 847                         break;
 848
 849                 /* A = X */
 850                 case BPF_MISC | BPF_TXA:
 851                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 852                         break;
 853
 854                 /* A = skb->len or X = skb->len */
 855                 case BPF_LD | BPF_W | BPF_LEN:
 856                 case BPF_LDX | BPF_W | BPF_LEN:
 857                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 858                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 859                                             offsetof(struct sk_buff, len));
 860                         break;
 861
 862                 /* Access seccomp_data fields. */
 863                 case BPF_LDX | BPF_ABS | BPF_W:
 864                         /* A = *(u32 *) (ctx + K) */
 865                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 866                         break;
 867
 868                 /* Unknown instruction. */
 869                 default:
 870                         goto err;
 871                 }
 872
 873                 insn++;
 874                 if (new_prog)
 875                         memcpy(new_insn, tmp_insns,
 876                                sizeof(*insn) * (insn - tmp_insns));
 877                 new_insn += insn - tmp_insns;
 878         }
 879
 880         if (!new_prog) {
 881                 /* Only calculating new length. */
 882                 *new_len = new_insn - first_insn;
 883                 if (*seen_ld_abs)
 884                         *new_len += 4; /* Prologue bits. */
 885                 return 0;
 886         }
 887
 888         pass++;
 889         if (new_flen != new_insn - first_insn) {
 890                 new_flen = new_insn - first_insn;
 891                 if (pass > 2)
 892                         goto err;
 893                 goto do_pass;
 894         }
 895
 896         kfree(addrs);
 897         BUG_ON(*new_len != new_flen);
 898         return 0;
 899 err:
 900         kfree(addrs);
 901         return -EINVAL;
 902 }
 903
 904 /* Security:
 905  *
 906  * As we dont want to clear mem[] array for each packet going through
 907  * __bpf_prog_run(), we check that filter loaded by user never try to read
 908  * a cell if not previously written, and we check all branches to be sure
 909  * a malicious user doesn't try to abuse us.
 910  */
 911 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 912 {
 913         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 914         int pc, ret = 0;
 915
 916         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 917
 918         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 919         if (!masks)
 920                 return -ENOMEM;
 921
 922         memset(masks, 0xff, flen * sizeof(*masks));
 923
 924         for (pc = 0; pc < flen; pc++) {
 925                 memvalid &= masks[pc];
 926
 927                 switch (filter[pc].code) {
 928                 case BPF_ST:
 929                 case BPF_STX:
 930                         memvalid |= (1 << filter[pc].k);
 931                         break;
 932                 case BPF_LD | BPF_MEM:
 933                 case BPF_LDX | BPF_MEM:
 934                         if (!(memvalid & (1 << filter[pc].k))) {
 935                                 ret = -EINVAL;
 936                                 goto error;
 937                         }
 938                         break;
 939                 case BPF_JMP | BPF_JA:
 940                         /* A jump must set masks on target */
 941                         masks[pc + 1 + filter[pc].k] &= memvalid;
 942                         memvalid = ~0;
 943                         break;
 944                 case BPF_JMP | BPF_JEQ | BPF_K:
 945                 case BPF_JMP | BPF_JEQ | BPF_X:
 946                 case BPF_JMP | BPF_JGE | BPF_K:
 947                 case BPF_JMP | BPF_JGE | BPF_X:
 948                 case BPF_JMP | BPF_JGT | BPF_K:
 949                 case BPF_JMP | BPF_JGT | BPF_X:
 950                 case BPF_JMP | BPF_JSET | BPF_K:
 951                 case BPF_JMP | BPF_JSET | BPF_X:
 952                         /* A jump must set masks on targets */
 953                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 954                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 955                         memvalid = ~0;
 956                         break;
 957                 }
 958         }
 959 error:
 960         kfree(masks);
 961         return ret;
 962 }
 963
 964 static bool chk_code_allowed(u16 code_to_probe)
 965 {
 966         static const bool codes[] = {
 967                 /* 32 bit ALU operations */
 968                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 969                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 970                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 971                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 972                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 973                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 974                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 975                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 976                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 977                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 978                 [BPF_ALU | BPF_AND | BPF_K] = true,
 979                 [BPF_ALU | BPF_AND | BPF_X] = true,
 980                 [BPF_ALU | BPF_OR | BPF_K] = true,
 981                 [BPF_ALU | BPF_OR | BPF_X] = true,
 982                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 983                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 984                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 985                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 986                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 987                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 988                 [BPF_ALU | BPF_NEG] = true,
 989                 /* Load instructions */
 990                 [BPF_LD | BPF_W | BPF_ABS] = true,
 991                 [BPF_LD | BPF_H | BPF_ABS] = true,
 992                 [BPF_LD | BPF_B | BPF_ABS] = true,
 993                 [BPF_LD | BPF_W | BPF_LEN] = true,
 994                 [BPF_LD | BPF_W | BPF_IND] = true,
 995                 [BPF_LD | BPF_H | BPF_IND] = true,
 996                 [BPF_LD | BPF_B | BPF_IND] = true,
 997                 [BPF_LD | BPF_IMM] = true,
 998                 [BPF_LD | BPF_MEM] = true,
 999                 [BPF_LDX | BPF_W | BPF_LEN] = true,
1000                 [BPF_LDX | BPF_B | BPF_MSH] = true,
1001                 [BPF_LDX | BPF_IMM] = true,
1002                 [BPF_LDX | BPF_MEM] = true,
1003                 /* Store instructions */
1004                 [BPF_ST] = true,
1005                 [BPF_STX] = true,
1006                 /* Misc instructions */
1007                 [BPF_MISC | BPF_TAX] = true,
1008                 [BPF_MISC | BPF_TXA] = true,
1009                 /* Return instructions */
1010                 [BPF_RET | BPF_K] = true,
1011                 [BPF_RET | BPF_A] = true,
1012                 /* Jump instructions */
1013                 [BPF_JMP | BPF_JA] = true,
1014                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1015                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1016                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1017                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1018                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1019                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1020                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1021                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1022         };
1023
1024         if (code_to_probe >= ARRAY_SIZE(codes))
1025                 return false;
1026
1027         return codes[code_to_probe];
1028 }
1029
1030 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1031                                 unsigned int flen)
1032 {
1033         if (filter == NULL)
1034                 return false;
1035         if (flen == 0 || flen > BPF_MAXINSNS)
1036                 return false;
1037
1038         return true;
1039 }
1040
1041 /**
1042  *      bpf_check_classic - verify socket filter code
1043  *      @filter: filter to verify
1044  *      @flen: length of filter
1045  *
1046  * Check the user's filter code. If we let some ugly
1047  * filter code slip through kaboom! The filter must contain
1048  * no references or jumps that are out of range, no illegal
1049  * instructions, and must end with a RET instruction.
1050  *
1051  * All jumps are forward as they are not signed.
1052  *
1053  * Returns 0 if the rule set is legal or -EINVAL if not.
1054  */
1055 static int bpf_check_classic(const struct sock_filter *filter,
1056                              unsigned int flen)
1057 {
1058         bool anc_found;
1059         int pc;
1060
1061         /* Check the filter code now */
1062         for (pc = 0; pc < flen; pc++) {
1063                 const struct sock_filter *ftest = &filter[pc];
1064
1065                 /* May we actually operate on this code? */
1066                 if (!chk_code_allowed(ftest->code))
1067                         return -EINVAL;
1068
1069                 /* Some instructions need special checks */
1070                 switch (ftest->code) {
1071                 case BPF_ALU | BPF_DIV | BPF_K:
1072                 case BPF_ALU | BPF_MOD | BPF_K:
1073                         /* Check for division by zero */
1074                         if (ftest->k == 0)
1075                                 return -EINVAL;
1076                         break;
1077                 case BPF_ALU | BPF_LSH | BPF_K:
1078                 case BPF_ALU | BPF_RSH | BPF_K:
1079                         if (ftest->k >= 32)
1080                                 return -EINVAL;
1081                         break;
1082                 case BPF_LD | BPF_MEM:
1083                 case BPF_LDX | BPF_MEM:
1084                 case BPF_ST:
1085                 case BPF_STX:
1086                         /* Check for invalid memory addresses */
1087                         if (ftest->k >= BPF_MEMWORDS)
1088                                 return -EINVAL;
1089                         break;
1090                 case BPF_JMP | BPF_JA:
1091                         /* Note, the large ftest->k might cause loops.
1092                          * Compare this with conditional jumps below,
1093                          * where offsets are limited. --ANK (981016)
1094                          */
1095                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1096                                 return -EINVAL;
1097                         break;
1098                 case BPF_JMP | BPF_JEQ | BPF_K:
1099                 case BPF_JMP | BPF_JEQ | BPF_X:
1100                 case BPF_JMP | BPF_JGE | BPF_K:
1101                 case BPF_JMP | BPF_JGE | BPF_X:
1102                 case BPF_JMP | BPF_JGT | BPF_K:
1103                 case BPF_JMP | BPF_JGT | BPF_X:
1104                 case BPF_JMP | BPF_JSET | BPF_K:
1105                 case BPF_JMP | BPF_JSET | BPF_X:
1106                         /* Both conditionals must be safe */
1107                         if (pc + ftest->jt + 1 >= flen ||
1108                             pc + ftest->jf + 1 >= flen)
1109                                 return -EINVAL;
1110                         break;
1111                 case BPF_LD | BPF_W | BPF_ABS:
1112                 case BPF_LD | BPF_H | BPF_ABS:
1113                 case BPF_LD | BPF_B | BPF_ABS:
1114                         anc_found = false;
1115                         if (bpf_anc_helper(ftest) & BPF_ANC)
1116                                 anc_found = true;
1117                         /* Ancillary operation unknown or unsupported */
1118                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1119                                 return -EINVAL;
1120                 }
1121         }
1122
1123         /* Last instruction must be a RET code */
1124         switch (filter[flen - 1].code) {
1125         case BPF_RET | BPF_K:
1126         case BPF_RET | BPF_A:
1127                 return check_load_and_stores(filter, flen);
1128         }
1129
1130         return -EINVAL;
1131 }
1132
1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1134                                       const struct sock_fprog *fprog)
1135 {
1136         unsigned int fsize = bpf_classic_proglen(fprog);
1137         struct sock_fprog_kern *fkprog;
1138
1139         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1140         if (!fp->orig_prog)
1141                 return -ENOMEM;
1142
1143         fkprog = fp->orig_prog;
1144         fkprog->len = fprog->len;
1145
1146         fkprog->filter = kmemdup(fp->insns, fsize,
1147                                  GFP_KERNEL | __GFP_NOWARN);
1148         if (!fkprog->filter) {
1149                 kfree(fp->orig_prog);
1150                 return -ENOMEM;
1151         }
1152
1153         return 0;
1154 }
1155
1156 static void bpf_release_orig_filter(struct bpf_prog *fp)
1157 {
1158         struct sock_fprog_kern *fprog = fp->orig_prog;
1159
1160         if (fprog) {
1161                 kfree(fprog->filter);
1162                 kfree(fprog);
1163         }
1164 }
1165
1166 static void __bpf_prog_release(struct bpf_prog *prog)
1167 {
1168         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1169                 bpf_prog_put(prog);
1170         } else {
1171                 bpf_release_orig_filter(prog);
1172                 bpf_prog_free(prog);
1173         }
1174 }
1175
1176 static void __sk_filter_release(struct sk_filter *fp)
1177 {
1178         __bpf_prog_release(fp->prog);
1179         kfree(fp);
1180 }
1181
1182 /**
1183  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1184  *      @rcu: rcu_head that contains the sk_filter to free
1185  */
1186 static void sk_filter_release_rcu(struct rcu_head *rcu)
1187 {
1188         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1189
1190         __sk_filter_release(fp);
1191 }
1192
1193 /**
1194  *      sk_filter_release - release a socket filter
1195  *      @fp: filter to remove
1196  *
1197  *      Remove a filter from a socket and release its resources.
1198  */
1199 static void sk_filter_release(struct sk_filter *fp)
1200 {
1201         if (refcount_dec_and_test(&fp->refcnt))
1202                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1203 }
1204
1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1206 {
1207         u32 filter_size = bpf_prog_size(fp->prog->len);
1208
1209         atomic_sub(filter_size, &sk->sk_omem_alloc);
1210         sk_filter_release(fp);
1211 }
1212
1213 /* try to charge the socket memory if there is space available
1214  * return true on success
1215  */
1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1217 {
1218         u32 filter_size = bpf_prog_size(fp->prog->len);
1219         int optmem_max = READ_ONCE(sysctl_optmem_max);
1220
1221         /* same check as in sock_kmalloc() */
1222         if (filter_size <= optmem_max &&
1223             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1224                 atomic_add(filter_size, &sk->sk_omem_alloc);
1225                 return true;
1226         }
1227         return false;
1228 }
1229
1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1231 {
1232         if (!refcount_inc_not_zero(&fp->refcnt))
1233                 return false;
1234
1235         if (!__sk_filter_charge(sk, fp)) {
1236                 sk_filter_release(fp);
1237                 return false;
1238         }
1239         return true;
1240 }
1241
1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1243 {
1244         struct sock_filter *old_prog;
1245         struct bpf_prog *old_fp;
1246         int err, new_len, old_len = fp->len;
1247         bool seen_ld_abs = false;
1248
1249         /* We are free to overwrite insns et al right here as it won't be used at
1250          * this point in time anymore internally after the migration to the eBPF
1251          * instruction representation.
1252          */
1253         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1254                      sizeof(struct bpf_insn));
1255
1256         /* Conversion cannot happen on overlapping memory areas,
1257          * so we need to keep the user BPF around until the 2nd
1258          * pass. At this time, the user BPF is stored in fp->insns.
1259          */
1260         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1261                            GFP_KERNEL | __GFP_NOWARN);
1262         if (!old_prog) {
1263                 err = -ENOMEM;
1264                 goto out_err;
1265         }
1266
1267         /* 1st pass: calculate the new program length. */
1268         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1269                                  &seen_ld_abs);
1270         if (err)
1271                 goto out_err_free;
1272
1273         /* Expand fp for appending the new filter representation. */
1274         old_fp = fp;
1275         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1276         if (!fp) {
1277                 /* The old_fp is still around in case we couldn't
1278                  * allocate new memory, so uncharge on that one.
1279                  */
1280                 fp = old_fp;
1281                 err = -ENOMEM;
1282                 goto out_err_free;
1283         }
1284
1285         fp->len = new_len;
1286
1287         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1288         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1289                                  &seen_ld_abs);
1290         if (err)
1291                 /* 2nd bpf_convert_filter() can fail only if it fails
1292                  * to allocate memory, remapping must succeed. Note,
1293                  * that at this time old_fp has already been released
1294                  * by krealloc().
1295                  */
1296                 goto out_err_free;
1297
1298         fp = bpf_prog_select_runtime(fp, &err);
1299         if (err)
1300                 goto out_err_free;
1301
1302         kfree(old_prog);
1303         return fp;
1304
1305 out_err_free:
1306         kfree(old_prog);
1307 out_err:
1308         __bpf_prog_release(fp);
1309         return ERR_PTR(err);
1310 }
1311
1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1313                                            bpf_aux_classic_check_t trans)
1314 {
1315         int err;
1316
1317         fp->bpf_func = NULL;
1318         fp->jited = 0;
1319
1320         err = bpf_check_classic(fp->insns, fp->len);
1321         if (err) {
1322                 __bpf_prog_release(fp);
1323                 return ERR_PTR(err);
1324         }
1325
1326         /* There might be additional checks and transformations
1327          * needed on classic filters, f.e. in case of seccomp.
1328          */
1329         if (trans) {
1330                 err = trans(fp->insns, fp->len);
1331                 if (err) {
1332                         __bpf_prog_release(fp);
1333                         return ERR_PTR(err);
1334                 }
1335         }
1336
1337         /* Probe if we can JIT compile the filter and if so, do
1338          * the compilation of the filter.
1339          */
1340         bpf_jit_compile(fp);
1341
1342         /* JIT compiler couldn't process this filter, so do the eBPF translation
1343          * for the optimized interpreter.
1344          */
1345         if (!fp->jited)
1346                 fp = bpf_migrate_filter(fp);
1347
1348         return fp;
1349 }
1350
1351 /**
1352  *      bpf_prog_create - create an unattached filter
1353  *      @pfp: the unattached filter that is created
1354  *      @fprog: the filter program
1355  *
1356  * Create a filter independent of any socket. We first run some
1357  * sanity checks on it to make sure it does not explode on us later.
1358  * If an error occurs or there is insufficient memory for the filter
1359  * a negative errno code is returned. On success the return is zero.
1360  */
1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1362 {
1363         unsigned int fsize = bpf_classic_proglen(fprog);
1364         struct bpf_prog *fp;
1365
1366         /* Make sure new filter is there and in the right amounts. */
1367         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1368                 return -EINVAL;
1369
1370         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1371         if (!fp)
1372                 return -ENOMEM;
1373
1374         memcpy(fp->insns, fprog->filter, fsize);
1375
1376         fp->len = fprog->len;
1377         /* Since unattached filters are not copied back to user
1378          * space through sk_get_filter(), we do not need to hold
1379          * a copy here, and can spare us the work.
1380          */
1381         fp->orig_prog = NULL;
1382
1383         /* bpf_prepare_filter() already takes care of freeing
1384          * memory in case something goes wrong.
1385          */
1386         fp = bpf_prepare_filter(fp, NULL);
1387         if (IS_ERR(fp))
1388                 return PTR_ERR(fp);
1389
1390         *pfp = fp;
1391         return 0;
1392 }
1393 EXPORT_SYMBOL_GPL(bpf_prog_create);
1394
1395 /**
1396  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1397  *      @pfp: the unattached filter that is created
1398  *      @fprog: the filter program
1399  *      @trans: post-classic verifier transformation handler
1400  *      @save_orig: save classic BPF program
1401  *
1402  * This function effectively does the same as bpf_prog_create(), only
1403  * that it builds up its insns buffer from user space provided buffer.
1404  * It also allows for passing a bpf_aux_classic_check_t handler.
1405  */
1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1407                               bpf_aux_classic_check_t trans, bool save_orig)
1408 {
1409         unsigned int fsize = bpf_classic_proglen(fprog);
1410         struct bpf_prog *fp;
1411         int err;
1412
1413         /* Make sure new filter is there and in the right amounts. */
1414         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1415                 return -EINVAL;
1416
1417         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1418         if (!fp)
1419                 return -ENOMEM;
1420
1421         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1422                 __bpf_prog_free(fp);
1423                 return -EFAULT;
1424         }
1425
1426         fp->len = fprog->len;
1427         fp->orig_prog = NULL;
1428
1429         if (save_orig) {
1430                 err = bpf_prog_store_orig_filter(fp, fprog);
1431                 if (err) {
1432                         __bpf_prog_free(fp);
1433                         return -ENOMEM;
1434                 }
1435         }
1436
1437         /* bpf_prepare_filter() already takes care of freeing
1438          * memory in case something goes wrong.
1439          */
1440         fp = bpf_prepare_filter(fp, trans);
1441         if (IS_ERR(fp))
1442                 return PTR_ERR(fp);
1443
1444         *pfp = fp;
1445         return 0;
1446 }
1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1448
1449 void bpf_prog_destroy(struct bpf_prog *fp)
1450 {
1451         __bpf_prog_release(fp);
1452 }
1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1454
1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1456 {
1457         struct sk_filter *fp, *old_fp;
1458
1459         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1460         if (!fp)
1461                 return -ENOMEM;
1462
1463         fp->prog = prog;
1464
1465         if (!__sk_filter_charge(sk, fp)) {
1466                 kfree(fp);
1467                 return -ENOMEM;
1468         }
1469         refcount_set(&fp->refcnt, 1);
1470
1471         old_fp = rcu_dereference_protected(sk->sk_filter,
1472                                            lockdep_sock_is_held(sk));
1473         rcu_assign_pointer(sk->sk_filter, fp);
1474
1475         if (old_fp)
1476                 sk_filter_uncharge(sk, old_fp);
1477
1478         return 0;
1479 }
1480
1481 static
1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1483 {
1484         unsigned int fsize = bpf_classic_proglen(fprog);
1485         struct bpf_prog *prog;
1486         int err;
1487
1488         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1489                 return ERR_PTR(-EPERM);
1490
1491         /* Make sure new filter is there and in the right amounts. */
1492         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1493                 return ERR_PTR(-EINVAL);
1494
1495         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1496         if (!prog)
1497                 return ERR_PTR(-ENOMEM);
1498
1499         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1500                 __bpf_prog_free(prog);
1501                 return ERR_PTR(-EFAULT);
1502         }
1503
1504         prog->len = fprog->len;
1505
1506         err = bpf_prog_store_orig_filter(prog, fprog);
1507         if (err) {
1508                 __bpf_prog_free(prog);
1509                 return ERR_PTR(-ENOMEM);
1510         }
1511
1512         /* bpf_prepare_filter() already takes care of freeing
1513          * memory in case something goes wrong.
1514          */
1515         return bpf_prepare_filter(prog, NULL);
1516 }
1517
1518 /**
1519  *      sk_attach_filter - attach a socket filter
1520  *      @fprog: the filter program
1521  *      @sk: the socket to use
1522  *
1523  * Attach the user's filter code. We first run some sanity checks on
1524  * it to make sure it does not explode on us later. If an error
1525  * occurs or there is insufficient memory for the filter a negative
1526  * errno code is returned. On success the return is zero.
1527  */
1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1529 {
1530         struct bpf_prog *prog = __get_filter(fprog, sk);
1531         int err;
1532
1533         if (IS_ERR(prog))
1534                 return PTR_ERR(prog);
1535
1536         err = __sk_attach_prog(prog, sk);
1537         if (err < 0) {
1538                 __bpf_prog_release(prog);
1539                 return err;
1540         }
1541
1542         return 0;
1543 }
1544 EXPORT_SYMBOL_GPL(sk_attach_filter);
1545
1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1547 {
1548         struct bpf_prog *prog = __get_filter(fprog, sk);
1549         int err;
1550
1551         if (IS_ERR(prog))
1552                 return PTR_ERR(prog);
1553
1554         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1555                 err = -ENOMEM;
1556         else
1557                 err = reuseport_attach_prog(sk, prog);
1558
1559         if (err)
1560                 __bpf_prog_release(prog);
1561
1562         return err;
1563 }
1564
1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1566 {
1567         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1568                 return ERR_PTR(-EPERM);
1569
1570         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1571 }
1572
1573 int sk_attach_bpf(u32 ufd, struct sock *sk)
1574 {
1575         struct bpf_prog *prog = __get_bpf(ufd, sk);
1576         int err;
1577
1578         if (IS_ERR(prog))
1579                 return PTR_ERR(prog);
1580
1581         err = __sk_attach_prog(prog, sk);
1582         if (err < 0) {
1583                 bpf_prog_put(prog);
1584                 return err;
1585         }
1586
1587         return 0;
1588 }
1589
1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1591 {
1592         struct bpf_prog *prog;
1593         int err;
1594
1595         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1596                 return -EPERM;
1597
1598         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1599         if (PTR_ERR(prog) == -EINVAL)
1600                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1601         if (IS_ERR(prog))
1602                 return PTR_ERR(prog);
1603
1604         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1605                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1606                  * bpf prog (e.g. sockmap).  It depends on the
1607                  * limitation imposed by bpf_prog_load().
1608                  * Hence, sysctl_optmem_max is not checked.
1609                  */
1610                 if ((sk->sk_type != SOCK_STREAM &&
1611                      sk->sk_type != SOCK_DGRAM) ||
1612                     (sk->sk_protocol != IPPROTO_UDP &&
1613                      sk->sk_protocol != IPPROTO_TCP) ||
1614                     (sk->sk_family != AF_INET &&
1615                      sk->sk_family != AF_INET6)) {
1616                         err = -ENOTSUPP;
1617                         goto err_prog_put;
1618                 }
1619         } else {
1620                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1621                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1622                         err = -ENOMEM;
1623                         goto err_prog_put;
1624                 }
1625         }
1626
1627         err = reuseport_attach_prog(sk, prog);
1628 err_prog_put:
1629         if (err)
1630                 bpf_prog_put(prog);
1631
1632         return err;
1633 }
1634
1635 void sk_reuseport_prog_free(struct bpf_prog *prog)
1636 {
1637         if (!prog)
1638                 return;
1639
1640         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1641                 bpf_prog_put(prog);
1642         else
1643                 bpf_prog_destroy(prog);
1644 }
1645
1646 struct bpf_scratchpad {
1647         union {
1648                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1649                 u8     buff[MAX_BPF_STACK];
1650         };
1651 };
1652
1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1654
1655 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1656                                           unsigned int write_len)
1657 {
1658         return skb_ensure_writable(skb, write_len);
1659 }
1660
1661 static inline int bpf_try_make_writable(struct sk_buff *skb,
1662                                         unsigned int write_len)
1663 {
1664         int err = __bpf_try_make_writable(skb, write_len);
1665
1666         bpf_compute_data_pointers(skb);
1667         return err;
1668 }
1669
1670 static int bpf_try_make_head_writable(struct sk_buff *skb)
1671 {
1672         return bpf_try_make_writable(skb, skb_headlen(skb));
1673 }
1674
1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1676 {
1677         if (skb_at_tc_ingress(skb))
1678                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1679 }
1680
1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1682 {
1683         if (skb_at_tc_ingress(skb))
1684                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1685 }
1686
1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1688            const void *, from, u32, len, u64, flags)
1689 {
1690         void *ptr;
1691
1692         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1693                 return -EINVAL;
1694         if (unlikely(offset > INT_MAX))
1695                 return -EFAULT;
1696         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1697                 return -EFAULT;
1698
1699         ptr = skb->data + offset;
1700         if (flags & BPF_F_RECOMPUTE_CSUM)
1701                 __skb_postpull_rcsum(skb, ptr, len, offset);
1702
1703         memcpy(ptr, from, len);
1704
1705         if (flags & BPF_F_RECOMPUTE_CSUM)
1706                 __skb_postpush_rcsum(skb, ptr, len, offset);
1707         if (flags & BPF_F_INVALIDATE_HASH)
1708                 skb_clear_hash(skb);
1709
1710         return 0;
1711 }
1712
1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1714         .func           = bpf_skb_store_bytes,
1715         .gpl_only       = false,
1716         .ret_type       = RET_INTEGER,
1717         .arg1_type      = ARG_PTR_TO_CTX,
1718         .arg2_type      = ARG_ANYTHING,
1719         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1720         .arg4_type      = ARG_CONST_SIZE,
1721         .arg5_type      = ARG_ANYTHING,
1722 };
1723
1724 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1725            void *, to, u32, len)
1726 {
1727         void *ptr;
1728
1729         if (unlikely(offset > INT_MAX))
1730                 goto err_clear;
1731
1732         ptr = skb_header_pointer(skb, offset, len, to);
1733         if (unlikely(!ptr))
1734                 goto err_clear;
1735         if (ptr != to)
1736                 memcpy(to, ptr, len);
1737
1738         return 0;
1739 err_clear:
1740         memset(to, 0, len);
1741         return -EFAULT;
1742 }
1743
1744 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1745         .func           = bpf_skb_load_bytes,
1746         .gpl_only       = false,
1747         .ret_type       = RET_INTEGER,
1748         .arg1_type      = ARG_PTR_TO_CTX,
1749         .arg2_type      = ARG_ANYTHING,
1750         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1751         .arg4_type      = ARG_CONST_SIZE,
1752 };
1753
1754 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1755            const struct bpf_flow_dissector *, ctx, u32, offset,
1756            void *, to, u32, len)
1757 {
1758         void *ptr;
1759
1760         if (unlikely(offset > 0xffff))
1761                 goto err_clear;
1762
1763         if (unlikely(!ctx->skb))
1764                 goto err_clear;
1765
1766         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1767         if (unlikely(!ptr))
1768                 goto err_clear;
1769         if (ptr != to)
1770                 memcpy(to, ptr, len);
1771
1772         return 0;
1773 err_clear:
1774         memset(to, 0, len);
1775         return -EFAULT;
1776 }
1777
1778 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1779         .func           = bpf_flow_dissector_load_bytes,
1780         .gpl_only       = false,
1781         .ret_type       = RET_INTEGER,
1782         .arg1_type      = ARG_PTR_TO_CTX,
1783         .arg2_type      = ARG_ANYTHING,
1784         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1785         .arg4_type      = ARG_CONST_SIZE,
1786 };
1787
1788 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1789            u32, offset, void *, to, u32, len, u32, start_header)
1790 {
1791         u8 *end = skb_tail_pointer(skb);
1792         u8 *start, *ptr;
1793
1794         if (unlikely(offset > 0xffff))
1795                 goto err_clear;
1796
1797         switch (start_header) {
1798         case BPF_HDR_START_MAC:
1799                 if (unlikely(!skb_mac_header_was_set(skb)))
1800                         goto err_clear;
1801                 start = skb_mac_header(skb);
1802                 break;
1803         case BPF_HDR_START_NET:
1804                 start = skb_network_header(skb);
1805                 break;
1806         default:
1807                 goto err_clear;
1808         }
1809
1810         ptr = start + offset;
1811
1812         if (likely(ptr + len <= end)) {
1813                 memcpy(to, ptr, len);
1814                 return 0;
1815         }
1816
1817 err_clear:
1818         memset(to, 0, len);
1819         return -EFAULT;
1820 }
1821
1822 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1823         .func           = bpf_skb_load_bytes_relative,
1824         .gpl_only       = false,
1825         .ret_type       = RET_INTEGER,
1826         .arg1_type      = ARG_PTR_TO_CTX,
1827         .arg2_type      = ARG_ANYTHING,
1828         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1829         .arg4_type      = ARG_CONST_SIZE,
1830         .arg5_type      = ARG_ANYTHING,
1831 };
1832
1833 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1834 {
1835         /* Idea is the following: should the needed direct read/write
1836          * test fail during runtime, we can pull in more data and redo
1837          * again, since implicitly, we invalidate previous checks here.
1838          *
1839          * Or, since we know how much we need to make read/writeable,
1840          * this can be done once at the program beginning for direct
1841          * access case. By this we overcome limitations of only current
1842          * headroom being accessible.
1843          */
1844         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1845 }
1846
1847 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1848         .func           = bpf_skb_pull_data,
1849         .gpl_only       = false,
1850         .ret_type       = RET_INTEGER,
1851         .arg1_type      = ARG_PTR_TO_CTX,
1852         .arg2_type      = ARG_ANYTHING,
1853 };
1854
1855 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1856 {
1857         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1858 }
1859
1860 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1861         .func           = bpf_sk_fullsock,
1862         .gpl_only       = false,
1863         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1864         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1865 };
1866
1867 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1868                                            unsigned int write_len)
1869 {
1870         return __bpf_try_make_writable(skb, write_len);
1871 }
1872
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1874 {
1875         /* Idea is the following: should the needed direct read/write
1876          * test fail during runtime, we can pull in more data and redo
1877          * again, since implicitly, we invalidate previous checks here.
1878          *
1879          * Or, since we know how much we need to make read/writeable,
1880          * this can be done once at the program beginning for direct
1881          * access case. By this we overcome limitations of only current
1882          * headroom being accessible.
1883          */
1884         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1885 }
1886
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888         .func           = sk_skb_pull_data,
1889         .gpl_only       = false,
1890         .ret_type       = RET_INTEGER,
1891         .arg1_type      = ARG_PTR_TO_CTX,
1892         .arg2_type      = ARG_ANYTHING,
1893 };
1894
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896            u64, from, u64, to, u64, flags)
1897 {
1898         __sum16 *ptr;
1899
1900         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1901                 return -EINVAL;
1902         if (unlikely(offset > 0xffff || offset & 1))
1903                 return -EFAULT;
1904         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1905                 return -EFAULT;
1906
1907         ptr = (__sum16 *)(skb->data + offset);
1908         switch (flags & BPF_F_HDR_FIELD_MASK) {
1909         case 0:
1910                 if (unlikely(from != 0))
1911                         return -EINVAL;
1912
1913                 csum_replace_by_diff(ptr, to);
1914                 break;
1915         case 2:
1916                 csum_replace2(ptr, from, to);
1917                 break;
1918         case 4:
1919                 csum_replace4(ptr, from, to);
1920                 break;
1921         default:
1922                 return -EINVAL;
1923         }
1924
1925         return 0;
1926 }
1927
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929         .func           = bpf_l3_csum_replace,
1930         .gpl_only       = false,
1931         .ret_type       = RET_INTEGER,
1932         .arg1_type      = ARG_PTR_TO_CTX,
1933         .arg2_type      = ARG_ANYTHING,
1934         .arg3_type      = ARG_ANYTHING,
1935         .arg4_type      = ARG_ANYTHING,
1936         .arg5_type      = ARG_ANYTHING,
1937 };
1938
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940            u64, from, u64, to, u64, flags)
1941 {
1942         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1945         __sum16 *ptr;
1946
1947         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1949                 return -EINVAL;
1950         if (unlikely(offset > 0xffff || offset & 1))
1951                 return -EFAULT;
1952         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1953                 return -EFAULT;
1954
1955         ptr = (__sum16 *)(skb->data + offset);
1956         if (is_mmzero && !do_mforce && !*ptr)
1957                 return 0;
1958
1959         switch (flags & BPF_F_HDR_FIELD_MASK) {
1960         case 0:
1961                 if (unlikely(from != 0))
1962                         return -EINVAL;
1963
1964                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1965                 break;
1966         case 2:
1967                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1968                 break;
1969         case 4:
1970                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1971                 break;
1972         default:
1973                 return -EINVAL;
1974         }
1975
1976         if (is_mmzero && !*ptr)
1977                 *ptr = CSUM_MANGLED_0;
1978         return 0;
1979 }
1980
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982         .func           = bpf_l4_csum_replace,
1983         .gpl_only       = false,
1984         .ret_type       = RET_INTEGER,
1985         .arg1_type      = ARG_PTR_TO_CTX,
1986         .arg2_type      = ARG_ANYTHING,
1987         .arg3_type      = ARG_ANYTHING,
1988         .arg4_type      = ARG_ANYTHING,
1989         .arg5_type      = ARG_ANYTHING,
1990 };
1991
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993            __be32 *, to, u32, to_size, __wsum, seed)
1994 {
1995         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996         u32 diff_size = from_size + to_size;
1997         int i, j = 0;
1998
1999         /* This is quite flexible, some examples:
2000          *
2001          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2002          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2003          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2004          *
2005          * Even for diffing, from_size and to_size don't need to be equal.
2006          */
2007         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008                      diff_size > sizeof(sp->diff)))
2009                 return -EINVAL;
2010
2011         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012                 sp->diff[j] = ~from[i];
2013         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2014                 sp->diff[j] = to[i];
2015
2016         return csum_partial(sp->diff, diff_size, seed);
2017 }
2018
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020         .func           = bpf_csum_diff,
2021         .gpl_only       = false,
2022         .pkt_access     = true,
2023         .ret_type       = RET_INTEGER,
2024         .arg1_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2026         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2027         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2028         .arg5_type      = ARG_ANYTHING,
2029 };
2030
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2032 {
2033         /* The interface is to be used in combination with bpf_csum_diff()
2034          * for direct packet writes. csum rotation for alignment as well
2035          * as emulating csum_sub() can be done from the eBPF program.
2036          */
2037         if (skb->ip_summed == CHECKSUM_COMPLETE)
2038                 return (skb->csum = csum_add(skb->csum, csum));
2039
2040         return -ENOTSUPP;
2041 }
2042
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044         .func           = bpf_csum_update,
2045         .gpl_only       = false,
2046         .ret_type       = RET_INTEGER,
2047         .arg1_type      = ARG_PTR_TO_CTX,
2048         .arg2_type      = ARG_ANYTHING,
2049 };
2050
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2052 {
2053         /* The interface is to be used in combination with bpf_skb_adjust_room()
2054          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055          * is passed as flags, for example.
2056          */
2057         switch (level) {
2058         case BPF_CSUM_LEVEL_INC:
2059                 __skb_incr_checksum_unnecessary(skb);
2060                 break;
2061         case BPF_CSUM_LEVEL_DEC:
2062                 __skb_decr_checksum_unnecessary(skb);
2063                 break;
2064         case BPF_CSUM_LEVEL_RESET:
2065                 __skb_reset_checksum_unnecessary(skb);
2066                 break;
2067         case BPF_CSUM_LEVEL_QUERY:
2068                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069                        skb->csum_level : -EACCES;
2070         default:
2071                 return -EINVAL;
2072         }
2073
2074         return 0;
2075 }
2076
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078         .func           = bpf_csum_level,
2079         .gpl_only       = false,
2080         .ret_type       = RET_INTEGER,
2081         .arg1_type      = ARG_PTR_TO_CTX,
2082         .arg2_type      = ARG_ANYTHING,
2083 };
2084
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2086 {
2087         return dev_forward_skb_nomtu(dev, skb);
2088 }
2089
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091                                       struct sk_buff *skb)
2092 {
2093         int ret = ____dev_forward_skb(dev, skb, false);
2094
2095         if (likely(!ret)) {
2096                 skb->dev = dev;
2097                 ret = netif_rx(skb);
2098         }
2099
2100         return ret;
2101 }
2102
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2104 {
2105         int ret;
2106
2107         if (dev_xmit_recursion()) {
2108                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2109                 kfree_skb(skb);
2110                 return -ENETDOWN;
2111         }
2112
2113         skb->dev = dev;
2114         skb_clear_tstamp(skb);
2115
2116         dev_xmit_recursion_inc();
2117         ret = dev_queue_xmit(skb);
2118         dev_xmit_recursion_dec();
2119
2120         return ret;
2121 }
2122
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2124                                  u32 flags)
2125 {
2126         unsigned int mlen = skb_network_offset(skb);
2127
2128         if (unlikely(skb->len <= mlen)) {
2129                 kfree_skb(skb);
2130                 return -ERANGE;
2131         }
2132
2133         if (mlen) {
2134                 __skb_pull(skb, mlen);
2135
2136                 /* At ingress, the mac header has already been pulled once.
2137                  * At egress, skb_pospull_rcsum has to be done in case that
2138                  * the skb is originated from ingress (i.e. a forwarded skb)
2139                  * to ensure that rcsum starts at net header.
2140                  */
2141                 if (!skb_at_tc_ingress(skb))
2142                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2143         }
2144         skb_pop_mac_header(skb);
2145         skb_reset_mac_len(skb);
2146         return flags & BPF_F_INGRESS ?
2147                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2148 }
2149
2150 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2151                                  u32 flags)
2152 {
2153         /* Verify that a link layer header is carried */
2154         if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) {
2155                 kfree_skb(skb);
2156                 return -ERANGE;
2157         }
2158
2159         bpf_push_mac_rcsum(skb);
2160         return flags & BPF_F_INGRESS ?
2161                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2162 }
2163
2164 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2165                           u32 flags)
2166 {
2167         if (dev_is_mac_header_xmit(dev))
2168                 return __bpf_redirect_common(skb, dev, flags);
2169         else
2170                 return __bpf_redirect_no_mac(skb, dev, flags);
2171 }
2172
2173 #if IS_ENABLED(CONFIG_IPV6)
2174 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2175                             struct net_device *dev, struct bpf_nh_params *nh)
2176 {
2177         u32 hh_len = LL_RESERVED_SPACE(dev);
2178         const struct in6_addr *nexthop;
2179         struct dst_entry *dst = NULL;
2180         struct neighbour *neigh;
2181
2182         if (dev_xmit_recursion()) {
2183                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2184                 goto out_drop;
2185         }
2186
2187         skb->dev = dev;
2188         skb_clear_tstamp(skb);
2189
2190         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2191                 skb = skb_expand_head(skb, hh_len);
2192                 if (!skb)
2193                         return -ENOMEM;
2194         }
2195
2196         rcu_read_lock_bh();
2197         if (!nh) {
2198                 dst = skb_dst(skb);
2199                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2200                                       &ipv6_hdr(skb)->daddr);
2201         } else {
2202                 nexthop = &nh->ipv6_nh;
2203         }
2204         neigh = ip_neigh_gw6(dev, nexthop);
2205         if (likely(!IS_ERR(neigh))) {
2206                 int ret;
2207
2208                 sock_confirm_neigh(skb, neigh);
2209                 dev_xmit_recursion_inc();
2210                 ret = neigh_output(neigh, skb, false);
2211                 dev_xmit_recursion_dec();
2212                 rcu_read_unlock_bh();
2213                 return ret;
2214         }
2215         rcu_read_unlock_bh();
2216         if (dst)
2217                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2218 out_drop:
2219         kfree_skb(skb);
2220         return -ENETDOWN;
2221 }
2222
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224                                    struct bpf_nh_params *nh)
2225 {
2226         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227         struct net *net = dev_net(dev);
2228         int err, ret = NET_XMIT_DROP;
2229
2230         if (!nh) {
2231                 struct dst_entry *dst;
2232                 struct flowi6 fl6 = {
2233                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234                         .flowi6_mark  = skb->mark,
2235                         .flowlabel    = ip6_flowinfo(ip6h),
2236                         .flowi6_oif   = dev->ifindex,
2237                         .flowi6_proto = ip6h->nexthdr,
2238                         .daddr        = ip6h->daddr,
2239                         .saddr        = ip6h->saddr,
2240                 };
2241
2242                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2243                 if (IS_ERR(dst))
2244                         goto out_drop;
2245
2246                 skb_dst_set(skb, dst);
2247         } else if (nh->nh_family != AF_INET6) {
2248                 goto out_drop;
2249         }
2250
2251         err = bpf_out_neigh_v6(net, skb, dev, nh);
2252         if (unlikely(net_xmit_eval(err)))
2253                 dev->stats.tx_errors++;
2254         else
2255                 ret = NET_XMIT_SUCCESS;
2256         goto out_xmit;
2257 out_drop:
2258         dev->stats.tx_errors++;
2259         kfree_skb(skb);
2260 out_xmit:
2261         return ret;
2262 }
2263 #else
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265                                    struct bpf_nh_params *nh)
2266 {
2267         kfree_skb(skb);
2268         return NET_XMIT_DROP;
2269 }
2270 #endif /* CONFIG_IPV6 */
2271
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274                             struct net_device *dev, struct bpf_nh_params *nh)
2275 {
2276         u32 hh_len = LL_RESERVED_SPACE(dev);
2277         struct neighbour *neigh;
2278         bool is_v6gw = false;
2279
2280         if (dev_xmit_recursion()) {
2281                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2282                 goto out_drop;
2283         }
2284
2285         skb->dev = dev;
2286         skb_clear_tstamp(skb);
2287
2288         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289                 skb = skb_expand_head(skb, hh_len);
2290                 if (!skb)
2291                         return -ENOMEM;
2292         }
2293
2294         rcu_read_lock_bh();
2295         if (!nh) {
2296                 struct dst_entry *dst = skb_dst(skb);
2297                 struct rtable *rt = container_of(dst, struct rtable, dst);
2298
2299                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2300         } else if (nh->nh_family == AF_INET6) {
2301                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2302                 is_v6gw = true;
2303         } else if (nh->nh_family == AF_INET) {
2304                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2305         } else {
2306                 rcu_read_unlock_bh();
2307                 goto out_drop;
2308         }
2309
2310         if (likely(!IS_ERR(neigh))) {
2311                 int ret;
2312
2313                 sock_confirm_neigh(skb, neigh);
2314                 dev_xmit_recursion_inc();
2315                 ret = neigh_output(neigh, skb, is_v6gw);
2316                 dev_xmit_recursion_dec();
2317                 rcu_read_unlock_bh();
2318                 return ret;
2319         }
2320         rcu_read_unlock_bh();
2321 out_drop:
2322         kfree_skb(skb);
2323         return -ENETDOWN;
2324 }
2325
2326 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2327                                    struct bpf_nh_params *nh)
2328 {
2329         const struct iphdr *ip4h = ip_hdr(skb);
2330         struct net *net = dev_net(dev);
2331         int err, ret = NET_XMIT_DROP;
2332
2333         if (!nh) {
2334                 struct flowi4 fl4 = {
2335                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2336                         .flowi4_mark  = skb->mark,
2337                         .flowi4_tos   = RT_TOS(ip4h->tos),
2338                         .flowi4_oif   = dev->ifindex,
2339                         .flowi4_proto = ip4h->protocol,
2340                         .daddr        = ip4h->daddr,
2341                         .saddr        = ip4h->saddr,
2342                 };
2343                 struct rtable *rt;
2344
2345                 rt = ip_route_output_flow(net, &fl4, NULL);
2346                 if (IS_ERR(rt))
2347                         goto out_drop;
2348                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2349                         ip_rt_put(rt);
2350                         goto out_drop;
2351                 }
2352
2353                 skb_dst_set(skb, &rt->dst);
2354         }
2355
2356         err = bpf_out_neigh_v4(net, skb, dev, nh);
2357         if (unlikely(net_xmit_eval(err)))
2358                 dev->stats.tx_errors++;
2359         else
2360                 ret = NET_XMIT_SUCCESS;
2361         goto out_xmit;
2362 out_drop:
2363         dev->stats.tx_errors++;
2364         kfree_skb(skb);
2365 out_xmit:
2366         return ret;
2367 }
2368 #else
2369 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2370                                    struct bpf_nh_params *nh)
2371 {
2372         kfree_skb(skb);
2373         return NET_XMIT_DROP;
2374 }
2375 #endif /* CONFIG_INET */
2376
2377 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2378                                 struct bpf_nh_params *nh)
2379 {
2380         struct ethhdr *ethh = eth_hdr(skb);
2381
2382         if (unlikely(skb->mac_header >= skb->network_header))
2383                 goto out;
2384         bpf_push_mac_rcsum(skb);
2385         if (is_multicast_ether_addr(ethh->h_dest))
2386                 goto out;
2387
2388         skb_pull(skb, sizeof(*ethh));
2389         skb_unset_mac_header(skb);
2390         skb_reset_network_header(skb);
2391
2392         if (skb->protocol == htons(ETH_P_IP))
2393                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2394         else if (skb->protocol == htons(ETH_P_IPV6))
2395                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2396 out:
2397         kfree_skb(skb);
2398         return -ENOTSUPP;
2399 }
2400
2401 /* Internal, non-exposed redirect flags. */
2402 enum {
2403         BPF_F_NEIGH     = (1ULL << 1),
2404         BPF_F_PEER      = (1ULL << 2),
2405         BPF_F_NEXTHOP   = (1ULL << 3),
2406 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2407 };
2408
2409 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2410 {
2411         struct net_device *dev;
2412         struct sk_buff *clone;
2413         int ret;
2414
2415         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2416                 return -EINVAL;
2417
2418         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2419         if (unlikely(!dev))
2420                 return -EINVAL;
2421
2422         clone = skb_clone(skb, GFP_ATOMIC);
2423         if (unlikely(!clone))
2424                 return -ENOMEM;
2425
2426         /* For direct write, we need to keep the invariant that the skbs
2427          * we're dealing with need to be uncloned. Should uncloning fail
2428          * here, we need to free the just generated clone to unclone once
2429          * again.
2430          */
2431         ret = bpf_try_make_head_writable(skb);
2432         if (unlikely(ret)) {
2433                 kfree_skb(clone);
2434                 return -ENOMEM;
2435         }
2436
2437         return __bpf_redirect(clone, dev, flags);
2438 }
2439
2440 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2441         .func           = bpf_clone_redirect,
2442         .gpl_only       = false,
2443         .ret_type       = RET_INTEGER,
2444         .arg1_type      = ARG_PTR_TO_CTX,
2445         .arg2_type      = ARG_ANYTHING,
2446         .arg3_type      = ARG_ANYTHING,
2447 };
2448
2449 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2450 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2451
2452 int skb_do_redirect(struct sk_buff *skb)
2453 {
2454         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2455         struct net *net = dev_net(skb->dev);
2456         struct net_device *dev;
2457         u32 flags = ri->flags;
2458
2459         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2460         ri->tgt_index = 0;
2461         ri->flags = 0;
2462         if (unlikely(!dev))
2463                 goto out_drop;
2464         if (flags & BPF_F_PEER) {
2465                 const struct net_device_ops *ops = dev->netdev_ops;
2466
2467                 if (unlikely(!ops->ndo_get_peer_dev ||
2468                              !skb_at_tc_ingress(skb)))
2469                         goto out_drop;
2470                 dev = ops->ndo_get_peer_dev(dev);
2471                 if (unlikely(!dev ||
2472                              !(dev->flags & IFF_UP) ||
2473                              net_eq(net, dev_net(dev))))
2474                         goto out_drop;
2475                 skb->dev = dev;
2476                 return -EAGAIN;
2477         }
2478         return flags & BPF_F_NEIGH ?
2479                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2480                                     &ri->nh : NULL) :
2481                __bpf_redirect(skb, dev, flags);
2482 out_drop:
2483         kfree_skb(skb);
2484         return -EINVAL;
2485 }
2486
2487 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2488 {
2489         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2490
2491         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2492                 return TC_ACT_SHOT;
2493
2494         ri->flags = flags;
2495         ri->tgt_index = ifindex;
2496
2497         return TC_ACT_REDIRECT;
2498 }
2499
2500 static const struct bpf_func_proto bpf_redirect_proto = {
2501         .func           = bpf_redirect,
2502         .gpl_only       = false,
2503         .ret_type       = RET_INTEGER,
2504         .arg1_type      = ARG_ANYTHING,
2505         .arg2_type      = ARG_ANYTHING,
2506 };
2507
2508 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2509 {
2510         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2511
2512         if (unlikely(flags))
2513                 return TC_ACT_SHOT;
2514
2515         ri->flags = BPF_F_PEER;
2516         ri->tgt_index = ifindex;
2517
2518         return TC_ACT_REDIRECT;
2519 }
2520
2521 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2522         .func           = bpf_redirect_peer,
2523         .gpl_only       = false,
2524         .ret_type       = RET_INTEGER,
2525         .arg1_type      = ARG_ANYTHING,
2526         .arg2_type      = ARG_ANYTHING,
2527 };
2528
2529 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2530            int, plen, u64, flags)
2531 {
2532         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2533
2534         if (unlikely((plen && plen < sizeof(*params)) || flags))
2535                 return TC_ACT_SHOT;
2536
2537         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2538         ri->tgt_index = ifindex;
2539
2540         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2541         if (plen)
2542                 memcpy(&ri->nh, params, sizeof(ri->nh));
2543
2544         return TC_ACT_REDIRECT;
2545 }
2546
2547 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2548         .func           = bpf_redirect_neigh,
2549         .gpl_only       = false,
2550         .ret_type       = RET_INTEGER,
2551         .arg1_type      = ARG_ANYTHING,
2552         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2553         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2554         .arg4_type      = ARG_ANYTHING,
2555 };
2556
2557 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2558 {
2559         msg->apply_bytes = bytes;
2560         return 0;
2561 }
2562
2563 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2564         .func           = bpf_msg_apply_bytes,
2565         .gpl_only       = false,
2566         .ret_type       = RET_INTEGER,
2567         .arg1_type      = ARG_PTR_TO_CTX,
2568         .arg2_type      = ARG_ANYTHING,
2569 };
2570
2571 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2572 {
2573         msg->cork_bytes = bytes;
2574         return 0;
2575 }
2576
2577 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2578         .func           = bpf_msg_cork_bytes,
2579         .gpl_only       = false,
2580         .ret_type       = RET_INTEGER,
2581         .arg1_type      = ARG_PTR_TO_CTX,
2582         .arg2_type      = ARG_ANYTHING,
2583 };
2584
2585 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2586            u32, end, u64, flags)
2587 {
2588         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2589         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2590         struct scatterlist *sge;
2591         u8 *raw, *to, *from;
2592         struct page *page;
2593
2594         if (unlikely(flags || end <= start))
2595                 return -EINVAL;
2596
2597         /* First find the starting scatterlist element */
2598         i = msg->sg.start;
2599         do {
2600                 offset += len;
2601                 len = sk_msg_elem(msg, i)->length;
2602                 if (start < offset + len)
2603                         break;
2604                 sk_msg_iter_var_next(i);
2605         } while (i != msg->sg.end);
2606
2607         if (unlikely(start >= offset + len))
2608                 return -EINVAL;
2609
2610         first_sge = i;
2611         /* The start may point into the sg element so we need to also
2612          * account for the headroom.
2613          */
2614         bytes_sg_total = start - offset + bytes;
2615         if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2616                 goto out;
2617
2618         /* At this point we need to linearize multiple scatterlist
2619          * elements or a single shared page. Either way we need to
2620          * copy into a linear buffer exclusively owned by BPF. Then
2621          * place the buffer in the scatterlist and fixup the original
2622          * entries by removing the entries now in the linear buffer
2623          * and shifting the remaining entries. For now we do not try
2624          * to copy partial entries to avoid complexity of running out
2625          * of sg_entry slots. The downside is reading a single byte
2626          * will copy the entire sg entry.
2627          */
2628         do {
2629                 copy += sk_msg_elem(msg, i)->length;
2630                 sk_msg_iter_var_next(i);
2631                 if (bytes_sg_total <= copy)
2632                         break;
2633         } while (i != msg->sg.end);
2634         last_sge = i;
2635
2636         if (unlikely(bytes_sg_total > copy))
2637                 return -EINVAL;
2638
2639         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2640                            get_order(copy));
2641         if (unlikely(!page))
2642                 return -ENOMEM;
2643
2644         raw = page_address(page);
2645         i = first_sge;
2646         do {
2647                 sge = sk_msg_elem(msg, i);
2648                 from = sg_virt(sge);
2649                 len = sge->length;
2650                 to = raw + poffset;
2651
2652                 memcpy(to, from, len);
2653                 poffset += len;
2654                 sge->length = 0;
2655                 put_page(sg_page(sge));
2656
2657                 sk_msg_iter_var_next(i);
2658         } while (i != last_sge);
2659
2660         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2661
2662         /* To repair sg ring we need to shift entries. If we only
2663          * had a single entry though we can just replace it and
2664          * be done. Otherwise walk the ring and shift the entries.
2665          */
2666         WARN_ON_ONCE(last_sge == first_sge);
2667         shift = last_sge > first_sge ?
2668                 last_sge - first_sge - 1 :
2669                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2670         if (!shift)
2671                 goto out;
2672
2673         i = first_sge;
2674         sk_msg_iter_var_next(i);
2675         do {
2676                 u32 move_from;
2677
2678                 if (i + shift >= NR_MSG_FRAG_IDS)
2679                         move_from = i + shift - NR_MSG_FRAG_IDS;
2680                 else
2681                         move_from = i + shift;
2682                 if (move_from == msg->sg.end)
2683                         break;
2684
2685                 msg->sg.data[i] = msg->sg.data[move_from];
2686                 msg->sg.data[move_from].length = 0;
2687                 msg->sg.data[move_from].page_link = 0;
2688                 msg->sg.data[move_from].offset = 0;
2689                 sk_msg_iter_var_next(i);
2690         } while (1);
2691
2692         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2693                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2694                       msg->sg.end - shift;
2695 out:
2696         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2697         msg->data_end = msg->data + bytes;
2698         return 0;
2699 }
2700
2701 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2702         .func           = bpf_msg_pull_data,
2703         .gpl_only       = false,
2704         .ret_type       = RET_INTEGER,
2705         .arg1_type      = ARG_PTR_TO_CTX,
2706         .arg2_type      = ARG_ANYTHING,
2707         .arg3_type      = ARG_ANYTHING,
2708         .arg4_type      = ARG_ANYTHING,
2709 };
2710
2711 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2712            u32, len, u64, flags)
2713 {
2714         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2715         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2716         u8 *raw, *to, *from;
2717         struct page *page;
2718
2719         if (unlikely(flags))
2720                 return -EINVAL;
2721
2722         if (unlikely(len == 0))
2723                 return 0;
2724
2725         /* First find the starting scatterlist element */
2726         i = msg->sg.start;
2727         do {
2728                 offset += l;
2729                 l = sk_msg_elem(msg, i)->length;
2730
2731                 if (start < offset + l)
2732                         break;
2733                 sk_msg_iter_var_next(i);
2734         } while (i != msg->sg.end);
2735
2736         if (start >= offset + l)
2737                 return -EINVAL;
2738
2739         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2740
2741         /* If no space available will fallback to copy, we need at
2742          * least one scatterlist elem available to push data into
2743          * when start aligns to the beginning of an element or two
2744          * when it falls inside an element. We handle the start equals
2745          * offset case because its the common case for inserting a
2746          * header.
2747          */
2748         if (!space || (space == 1 && start != offset))
2749                 copy = msg->sg.data[i].length;
2750
2751         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2752                            get_order(copy + len));
2753         if (unlikely(!page))
2754                 return -ENOMEM;
2755
2756         if (copy) {
2757                 int front, back;
2758
2759                 raw = page_address(page);
2760
2761                 psge = sk_msg_elem(msg, i);
2762                 front = start - offset;
2763                 back = psge->length - front;
2764                 from = sg_virt(psge);
2765
2766                 if (front)
2767                         memcpy(raw, from, front);
2768
2769                 if (back) {
2770                         from += front;
2771                         to = raw + front + len;
2772
2773                         memcpy(to, from, back);
2774                 }
2775
2776                 put_page(sg_page(psge));
2777         } else if (start - offset) {
2778                 psge = sk_msg_elem(msg, i);
2779                 rsge = sk_msg_elem_cpy(msg, i);
2780
2781                 psge->length = start - offset;
2782                 rsge.length -= psge->length;
2783                 rsge.offset += start;
2784
2785                 sk_msg_iter_var_next(i);
2786                 sg_unmark_end(psge);
2787                 sg_unmark_end(&rsge);
2788                 sk_msg_iter_next(msg, end);
2789         }
2790
2791         /* Slot(s) to place newly allocated data */
2792         new = i;
2793
2794         /* Shift one or two slots as needed */
2795         if (!copy) {
2796                 sge = sk_msg_elem_cpy(msg, i);
2797
2798                 sk_msg_iter_var_next(i);
2799                 sg_unmark_end(&sge);
2800                 sk_msg_iter_next(msg, end);
2801
2802                 nsge = sk_msg_elem_cpy(msg, i);
2803                 if (rsge.length) {
2804                         sk_msg_iter_var_next(i);
2805                         nnsge = sk_msg_elem_cpy(msg, i);
2806                 }
2807
2808                 while (i != msg->sg.end) {
2809                         msg->sg.data[i] = sge;
2810                         sge = nsge;
2811                         sk_msg_iter_var_next(i);
2812                         if (rsge.length) {
2813                                 nsge = nnsge;
2814                                 nnsge = sk_msg_elem_cpy(msg, i);
2815                         } else {
2816                                 nsge = sk_msg_elem_cpy(msg, i);
2817                         }
2818                 }
2819         }
2820
2821         /* Place newly allocated data buffer */
2822         sk_mem_charge(msg->sk, len);
2823         msg->sg.size += len;
2824         __clear_bit(new, msg->sg.copy);
2825         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2826         if (rsge.length) {
2827                 get_page(sg_page(&rsge));
2828                 sk_msg_iter_var_next(new);
2829                 msg->sg.data[new] = rsge;
2830         }
2831
2832         sk_msg_compute_data_pointers(msg);
2833         return 0;
2834 }
2835
2836 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2837         .func           = bpf_msg_push_data,
2838         .gpl_only       = false,
2839         .ret_type       = RET_INTEGER,
2840         .arg1_type      = ARG_PTR_TO_CTX,
2841         .arg2_type      = ARG_ANYTHING,
2842         .arg3_type      = ARG_ANYTHING,
2843         .arg4_type      = ARG_ANYTHING,
2844 };
2845
2846 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2847 {
2848         int prev;
2849
2850         do {
2851                 prev = i;
2852                 sk_msg_iter_var_next(i);
2853                 msg->sg.data[prev] = msg->sg.data[i];
2854         } while (i != msg->sg.end);
2855
2856         sk_msg_iter_prev(msg, end);
2857 }
2858
2859 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2860 {
2861         struct scatterlist tmp, sge;
2862
2863         sk_msg_iter_next(msg, end);
2864         sge = sk_msg_elem_cpy(msg, i);
2865         sk_msg_iter_var_next(i);
2866         tmp = sk_msg_elem_cpy(msg, i);
2867
2868         while (i != msg->sg.end) {
2869                 msg->sg.data[i] = sge;
2870                 sk_msg_iter_var_next(i);
2871                 sge = tmp;
2872                 tmp = sk_msg_elem_cpy(msg, i);
2873         }
2874 }
2875
2876 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2877            u32, len, u64, flags)
2878 {
2879         u32 i = 0, l = 0, space, offset = 0;
2880         u64 last = start + len;
2881         int pop;
2882
2883         if (unlikely(flags))
2884                 return -EINVAL;
2885
2886         /* First find the starting scatterlist element */
2887         i = msg->sg.start;
2888         do {
2889                 offset += l;
2890                 l = sk_msg_elem(msg, i)->length;
2891
2892                 if (start < offset + l)
2893                         break;
2894                 sk_msg_iter_var_next(i);
2895         } while (i != msg->sg.end);
2896
2897         /* Bounds checks: start and pop must be inside message */
2898         if (start >= offset + l || last >= msg->sg.size)
2899                 return -EINVAL;
2900
2901         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2902
2903         pop = len;
2904         /* --------------| offset
2905          * -| start      |-------- len -------|
2906          *
2907          *  |----- a ----|-------- pop -------|----- b ----|
2908          *  |______________________________________________| length
2909          *
2910          *
2911          * a:   region at front of scatter element to save
2912          * b:   region at back of scatter element to save when length > A + pop
2913          * pop: region to pop from element, same as input 'pop' here will be
2914          *      decremented below per iteration.
2915          *
2916          * Two top-level cases to handle when start != offset, first B is non
2917          * zero and second B is zero corresponding to when a pop includes more
2918          * than one element.
2919          *
2920          * Then if B is non-zero AND there is no space allocate space and
2921          * compact A, B regions into page. If there is space shift ring to
2922          * the rigth free'ing the next element in ring to place B, leaving
2923          * A untouched except to reduce length.
2924          */
2925         if (start != offset) {
2926                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2927                 int a = start;
2928                 int b = sge->length - pop - a;
2929
2930                 sk_msg_iter_var_next(i);
2931
2932                 if (pop < sge->length - a) {
2933                         if (space) {
2934                                 sge->length = a;
2935                                 sk_msg_shift_right(msg, i);
2936                                 nsge = sk_msg_elem(msg, i);
2937                                 get_page(sg_page(sge));
2938                                 sg_set_page(nsge,
2939                                             sg_page(sge),
2940                                             b, sge->offset + pop + a);
2941                         } else {
2942                                 struct page *page, *orig;
2943                                 u8 *to, *from;
2944
2945                                 page = alloc_pages(__GFP_NOWARN |
2946                                                    __GFP_COMP   | GFP_ATOMIC,
2947                                                    get_order(a + b));
2948                                 if (unlikely(!page))
2949                                         return -ENOMEM;
2950
2951                                 sge->length = a;
2952                                 orig = sg_page(sge);
2953                                 from = sg_virt(sge);
2954                                 to = page_address(page);
2955                                 memcpy(to, from, a);
2956                                 memcpy(to + a, from + a + pop, b);
2957                                 sg_set_page(sge, page, a + b, 0);
2958                                 put_page(orig);
2959                         }
2960                         pop = 0;
2961                 } else if (pop >= sge->length - a) {
2962                         pop -= (sge->length - a);
2963                         sge->length = a;
2964                 }
2965         }
2966
2967         /* From above the current layout _must_ be as follows,
2968          *
2969          * -| offset
2970          * -| start
2971          *
2972          *  |---- pop ---|---------------- b ------------|
2973          *  |____________________________________________| length
2974          *
2975          * Offset and start of the current msg elem are equal because in the
2976          * previous case we handled offset != start and either consumed the
2977          * entire element and advanced to the next element OR pop == 0.
2978          *
2979          * Two cases to handle here are first pop is less than the length
2980          * leaving some remainder b above. Simply adjust the element's layout
2981          * in this case. Or pop >= length of the element so that b = 0. In this
2982          * case advance to next element decrementing pop.
2983          */
2984         while (pop) {
2985                 struct scatterlist *sge = sk_msg_elem(msg, i);
2986
2987                 if (pop < sge->length) {
2988                         sge->length -= pop;
2989                         sge->offset += pop;
2990                         pop = 0;
2991                 } else {
2992                         pop -= sge->length;
2993                         sk_msg_shift_left(msg, i);
2994                 }
2995                 sk_msg_iter_var_next(i);
2996         }
2997
2998         sk_mem_uncharge(msg->sk, len - pop);
2999         msg->sg.size -= (len - pop);
3000         sk_msg_compute_data_pointers(msg);
3001         return 0;
3002 }
3003
3004 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3005         .func           = bpf_msg_pop_data,
3006         .gpl_only       = false,
3007         .ret_type       = RET_INTEGER,
3008         .arg1_type      = ARG_PTR_TO_CTX,
3009         .arg2_type      = ARG_ANYTHING,
3010         .arg3_type      = ARG_ANYTHING,
3011         .arg4_type      = ARG_ANYTHING,
3012 };
3013
3014 #ifdef CONFIG_CGROUP_NET_CLASSID
3015 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3016 {
3017         return __task_get_classid(current);
3018 }
3019
3020 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3021         .func           = bpf_get_cgroup_classid_curr,
3022         .gpl_only       = false,
3023         .ret_type       = RET_INTEGER,
3024 };
3025
3026 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3027 {
3028         struct sock *sk = skb_to_full_sk(skb);
3029
3030         if (!sk || !sk_fullsock(sk))
3031                 return 0;
3032
3033         return sock_cgroup_classid(&sk->sk_cgrp_data);
3034 }
3035
3036 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3037         .func           = bpf_skb_cgroup_classid,
3038         .gpl_only       = false,
3039         .ret_type       = RET_INTEGER,
3040         .arg1_type      = ARG_PTR_TO_CTX,
3041 };
3042 #endif
3043
3044 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3045 {
3046         return task_get_classid(skb);
3047 }
3048
3049 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3050         .func           = bpf_get_cgroup_classid,
3051         .gpl_only       = false,
3052         .ret_type       = RET_INTEGER,
3053         .arg1_type      = ARG_PTR_TO_CTX,
3054 };
3055
3056 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3057 {
3058         return dst_tclassid(skb);
3059 }
3060
3061 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3062         .func           = bpf_get_route_realm,
3063         .gpl_only       = false,
3064         .ret_type       = RET_INTEGER,
3065         .arg1_type      = ARG_PTR_TO_CTX,
3066 };
3067
3068 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3069 {
3070         /* If skb_clear_hash() was called due to mangling, we can
3071          * trigger SW recalculation here. Later access to hash
3072          * can then use the inline skb->hash via context directly
3073          * instead of calling this helper again.
3074          */
3075         return skb_get_hash(skb);
3076 }
3077
3078 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3079         .func           = bpf_get_hash_recalc,
3080         .gpl_only       = false,
3081         .ret_type       = RET_INTEGER,
3082         .arg1_type      = ARG_PTR_TO_CTX,
3083 };
3084
3085 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3086 {
3087         /* After all direct packet write, this can be used once for
3088          * triggering a lazy recalc on next skb_get_hash() invocation.
3089          */
3090         skb_clear_hash(skb);
3091         return 0;
3092 }
3093
3094 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3095         .func           = bpf_set_hash_invalid,
3096         .gpl_only       = false,
3097         .ret_type       = RET_INTEGER,
3098         .arg1_type      = ARG_PTR_TO_CTX,
3099 };
3100
3101 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3102 {
3103         /* Set user specified hash as L4(+), so that it gets returned
3104          * on skb_get_hash() call unless BPF prog later on triggers a
3105          * skb_clear_hash().
3106          */
3107         __skb_set_sw_hash(skb, hash, true);
3108         return 0;
3109 }
3110
3111 static const struct bpf_func_proto bpf_set_hash_proto = {
3112         .func           = bpf_set_hash,
3113         .gpl_only       = false,
3114         .ret_type       = RET_INTEGER,
3115         .arg1_type      = ARG_PTR_TO_CTX,
3116         .arg2_type      = ARG_ANYTHING,
3117 };
3118
3119 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3120            u16, vlan_tci)
3121 {
3122         int ret;
3123
3124         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3125                      vlan_proto != htons(ETH_P_8021AD)))
3126                 vlan_proto = htons(ETH_P_8021Q);
3127
3128         bpf_push_mac_rcsum(skb);
3129         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3130         bpf_pull_mac_rcsum(skb);
3131
3132         bpf_compute_data_pointers(skb);
3133         return ret;
3134 }
3135
3136 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3137         .func           = bpf_skb_vlan_push,
3138         .gpl_only       = false,
3139         .ret_type       = RET_INTEGER,
3140         .arg1_type      = ARG_PTR_TO_CTX,
3141         .arg2_type      = ARG_ANYTHING,
3142         .arg3_type      = ARG_ANYTHING,
3143 };
3144
3145 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3146 {
3147         int ret;
3148
3149         bpf_push_mac_rcsum(skb);
3150         ret = skb_vlan_pop(skb);
3151         bpf_pull_mac_rcsum(skb);
3152
3153         bpf_compute_data_pointers(skb);
3154         return ret;
3155 }
3156
3157 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3158         .func           = bpf_skb_vlan_pop,
3159         .gpl_only       = false,
3160         .ret_type       = RET_INTEGER,
3161         .arg1_type      = ARG_PTR_TO_CTX,
3162 };
3163
3164 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3165 {
3166         /* Caller already did skb_cow() with len as headroom,
3167          * so no need to do it here.
3168          */
3169         skb_push(skb, len);
3170         memmove(skb->data, skb->data + len, off);
3171         memset(skb->data + off, 0, len);
3172
3173         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3174          * needed here as it does not change the skb->csum
3175          * result for checksum complete when summing over
3176          * zeroed blocks.
3177          */
3178         return 0;
3179 }
3180
3181 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3182 {
3183         void *old_data;
3184
3185         /* skb_ensure_writable() is not needed here, as we're
3186          * already working on an uncloned skb.
3187          */
3188         if (unlikely(!pskb_may_pull(skb, off + len)))
3189                 return -ENOMEM;
3190
3191         old_data = skb->data;
3192         __skb_pull(skb, len);
3193         skb_postpull_rcsum(skb, old_data + off, len);
3194         memmove(skb->data, old_data, off);
3195
3196         return 0;
3197 }
3198
3199 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3200 {
3201         bool trans_same = skb->transport_header == skb->network_header;
3202         int ret;
3203
3204         /* There's no need for __skb_push()/__skb_pull() pair to
3205          * get to the start of the mac header as we're guaranteed
3206          * to always start from here under eBPF.
3207          */
3208         ret = bpf_skb_generic_push(skb, off, len);
3209         if (likely(!ret)) {
3210                 skb->mac_header -= len;
3211                 skb->network_header -= len;
3212                 if (trans_same)
3213                         skb->transport_header = skb->network_header;
3214         }
3215
3216         return ret;
3217 }
3218
3219 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3220 {
3221         bool trans_same = skb->transport_header == skb->network_header;
3222         int ret;
3223
3224         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3225         ret = bpf_skb_generic_pop(skb, off, len);
3226         if (likely(!ret)) {
3227                 skb->mac_header += len;
3228                 skb->network_header += len;
3229                 if (trans_same)
3230                         skb->transport_header = skb->network_header;
3231         }
3232
3233         return ret;
3234 }
3235
3236 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3237 {
3238         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3239         u32 off = skb_mac_header_len(skb);
3240         int ret;
3241
3242         ret = skb_cow(skb, len_diff);
3243         if (unlikely(ret < 0))
3244                 return ret;
3245
3246         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3247         if (unlikely(ret < 0))
3248                 return ret;
3249
3250         if (skb_is_gso(skb)) {
3251                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3252
3253                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3254                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3255                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3256                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3257                 }
3258         }
3259
3260         skb->protocol = htons(ETH_P_IPV6);
3261         skb_clear_hash(skb);
3262
3263         return 0;
3264 }
3265
3266 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3267 {
3268         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3269         u32 off = skb_mac_header_len(skb);
3270         int ret;
3271
3272         ret = skb_unclone(skb, GFP_ATOMIC);
3273         if (unlikely(ret < 0))
3274                 return ret;
3275
3276         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3277         if (unlikely(ret < 0))
3278                 return ret;
3279
3280         if (skb_is_gso(skb)) {
3281                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3282
3283                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3284                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3285                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3286                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3287                 }
3288         }
3289
3290         skb->protocol = htons(ETH_P_IP);
3291         skb_clear_hash(skb);
3292
3293         return 0;
3294 }
3295
3296 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3297 {
3298         __be16 from_proto = skb->protocol;
3299
3300         if (from_proto == htons(ETH_P_IP) &&
3301               to_proto == htons(ETH_P_IPV6))
3302                 return bpf_skb_proto_4_to_6(skb);
3303
3304         if (from_proto == htons(ETH_P_IPV6) &&
3305               to_proto == htons(ETH_P_IP))
3306                 return bpf_skb_proto_6_to_4(skb);
3307
3308         return -ENOTSUPP;
3309 }
3310
3311 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3312            u64, flags)
3313 {
3314         int ret;
3315
3316         if (unlikely(flags))
3317                 return -EINVAL;
3318
3319         /* General idea is that this helper does the basic groundwork
3320          * needed for changing the protocol, and eBPF program fills the
3321          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3322          * and other helpers, rather than passing a raw buffer here.
3323          *
3324          * The rationale is to keep this minimal and without a need to
3325          * deal with raw packet data. F.e. even if we would pass buffers
3326          * here, the program still needs to call the bpf_lX_csum_replace()
3327          * helpers anyway. Plus, this way we keep also separation of
3328          * concerns, since f.e. bpf_skb_store_bytes() should only take
3329          * care of stores.
3330          *
3331          * Currently, additional options and extension header space are
3332          * not supported, but flags register is reserved so we can adapt
3333          * that. For offloads, we mark packet as dodgy, so that headers
3334          * need to be verified first.
3335          */
3336         ret = bpf_skb_proto_xlat(skb, proto);
3337         bpf_compute_data_pointers(skb);
3338         return ret;
3339 }
3340
3341 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3342         .func           = bpf_skb_change_proto,
3343         .gpl_only       = false,
3344         .ret_type       = RET_INTEGER,
3345         .arg1_type      = ARG_PTR_TO_CTX,
3346         .arg2_type      = ARG_ANYTHING,
3347         .arg3_type      = ARG_ANYTHING,
3348 };
3349
3350 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3351 {
3352         /* We only allow a restricted subset to be changed for now. */
3353         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3354                      !skb_pkt_type_ok(pkt_type)))
3355                 return -EINVAL;
3356
3357         skb->pkt_type = pkt_type;
3358         return 0;
3359 }
3360
3361 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3362         .func           = bpf_skb_change_type,
3363         .gpl_only       = false,
3364         .ret_type       = RET_INTEGER,
3365         .arg1_type      = ARG_PTR_TO_CTX,
3366         .arg2_type      = ARG_ANYTHING,
3367 };
3368
3369 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3370 {
3371         switch (skb->protocol) {
3372         case htons(ETH_P_IP):
3373                 return sizeof(struct iphdr);
3374         case htons(ETH_P_IPV6):
3375                 return sizeof(struct ipv6hdr);
3376         default:
3377                 return ~0U;
3378         }
3379 }
3380
3381 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3382                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3383
3384 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK    (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \
3385                                          BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3386
3387 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3388                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3389                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3390                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3391                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3392                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3393                                           BPF_ADJ_ROOM_ENCAP_L2_MASK) | \
3394                                          BPF_F_ADJ_ROOM_DECAP_L3_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_DECAP_L3_MASK |
3509                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3510                 return -EINVAL;
3511
3512         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3513                 /* udp gso_size delineates datagrams, only allow if fixed */
3514                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3515                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3516                         return -ENOTSUPP;
3517         }
3518
3519         ret = skb_unclone(skb, GFP_ATOMIC);
3520         if (unlikely(ret < 0))
3521                 return ret;
3522
3523         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3524         if (unlikely(ret < 0))
3525                 return ret;
3526
3527         /* Match skb->protocol to new outer l3 protocol */
3528         if (skb->protocol == htons(ETH_P_IP) &&
3529             flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6)
3530                 skb->protocol = htons(ETH_P_IPV6);
3531         else if (skb->protocol == htons(ETH_P_IPV6) &&
3532                  flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4)
3533                 skb->protocol = htons(ETH_P_IP);
3534
3535         if (skb_is_gso(skb)) {
3536                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3537
3538                 /* Due to header shrink, MSS can be upgraded. */
3539                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3540                         skb_increase_gso_size(shinfo, len_diff);
3541
3542                 /* Header must be checked, and gso_segs recomputed. */
3543                 shinfo->gso_type |= SKB_GSO_DODGY;
3544                 shinfo->gso_segs = 0;
3545         }
3546
3547         return 0;
3548 }
3549
3550 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3551
3552 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3553            u32, mode, u64, flags)
3554 {
3555         u32 len_diff_abs = abs(len_diff);
3556         bool shrink = len_diff < 0;
3557         int ret = 0;
3558
3559         if (unlikely(flags || mode))
3560                 return -EINVAL;
3561         if (unlikely(len_diff_abs > 0xfffU))
3562                 return -EFAULT;
3563
3564         if (!shrink) {
3565                 ret = skb_cow(skb, len_diff);
3566                 if (unlikely(ret < 0))
3567                         return ret;
3568                 __skb_push(skb, len_diff_abs);
3569                 memset(skb->data, 0, len_diff_abs);
3570         } else {
3571                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3572                         return -ENOMEM;
3573                 __skb_pull(skb, len_diff_abs);
3574         }
3575         if (tls_sw_has_ctx_rx(skb->sk)) {
3576                 struct strp_msg *rxm = strp_msg(skb);
3577
3578                 rxm->full_len += len_diff;
3579         }
3580         return ret;
3581 }
3582
3583 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3584         .func           = sk_skb_adjust_room,
3585         .gpl_only       = false,
3586         .ret_type       = RET_INTEGER,
3587         .arg1_type      = ARG_PTR_TO_CTX,
3588         .arg2_type      = ARG_ANYTHING,
3589         .arg3_type      = ARG_ANYTHING,
3590         .arg4_type      = ARG_ANYTHING,
3591 };
3592
3593 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3594            u32, mode, u64, flags)
3595 {
3596         u32 len_cur, len_diff_abs = abs(len_diff);
3597         u32 len_min = bpf_skb_net_base_len(skb);
3598         u32 len_max = BPF_SKB_MAX_LEN;
3599         __be16 proto = skb->protocol;
3600         bool shrink = len_diff < 0;
3601         u32 off;
3602         int ret;
3603
3604         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3605                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3606                 return -EINVAL;
3607         if (unlikely(len_diff_abs > 0xfffU))
3608                 return -EFAULT;
3609         if (unlikely(proto != htons(ETH_P_IP) &&
3610                      proto != htons(ETH_P_IPV6)))
3611                 return -ENOTSUPP;
3612
3613         off = skb_mac_header_len(skb);
3614         switch (mode) {
3615         case BPF_ADJ_ROOM_NET:
3616                 off += bpf_skb_net_base_len(skb);
3617                 break;
3618         case BPF_ADJ_ROOM_MAC:
3619                 break;
3620         default:
3621                 return -ENOTSUPP;
3622         }
3623
3624         if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3625                 if (!shrink)
3626                         return -EINVAL;
3627
3628                 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) {
3629                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4:
3630                         len_min = sizeof(struct iphdr);
3631                         break;
3632                 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6:
3633                         len_min = sizeof(struct ipv6hdr);
3634                         break;
3635                 default:
3636                         return -EINVAL;
3637                 }
3638         }
3639
3640         len_cur = skb->len - skb_network_offset(skb);
3641         if ((shrink && (len_diff_abs >= len_cur ||
3642                         len_cur - len_diff_abs < len_min)) ||
3643             (!shrink && (skb->len + len_diff_abs > len_max &&
3644                          !skb_is_gso(skb))))
3645                 return -ENOTSUPP;
3646
3647         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3648                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3649         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3650                 __skb_reset_checksum_unnecessary(skb);
3651
3652         bpf_compute_data_pointers(skb);
3653         return ret;
3654 }
3655
3656 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3657         .func           = bpf_skb_adjust_room,
3658         .gpl_only       = false,
3659         .ret_type       = RET_INTEGER,
3660         .arg1_type      = ARG_PTR_TO_CTX,
3661         .arg2_type      = ARG_ANYTHING,
3662         .arg3_type      = ARG_ANYTHING,
3663         .arg4_type      = ARG_ANYTHING,
3664 };
3665
3666 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3667 {
3668         u32 min_len = skb_network_offset(skb);
3669
3670         if (skb_transport_header_was_set(skb))
3671                 min_len = skb_transport_offset(skb);
3672         if (skb->ip_summed == CHECKSUM_PARTIAL)
3673                 min_len = skb_checksum_start_offset(skb) +
3674                           skb->csum_offset + sizeof(__sum16);
3675         return min_len;
3676 }
3677
3678 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3679 {
3680         unsigned int old_len = skb->len;
3681         int ret;
3682
3683         ret = __skb_grow_rcsum(skb, new_len);
3684         if (!ret)
3685                 memset(skb->data + old_len, 0, new_len - old_len);
3686         return ret;
3687 }
3688
3689 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3690 {
3691         return __skb_trim_rcsum(skb, new_len);
3692 }
3693
3694 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3695                                         u64 flags)
3696 {
3697         u32 max_len = BPF_SKB_MAX_LEN;
3698         u32 min_len = __bpf_skb_min_len(skb);
3699         int ret;
3700
3701         if (unlikely(flags || new_len > max_len || new_len < min_len))
3702                 return -EINVAL;
3703         if (skb->encapsulation)
3704                 return -ENOTSUPP;
3705
3706         /* The basic idea of this helper is that it's performing the
3707          * needed work to either grow or trim an skb, and eBPF program
3708          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3709          * bpf_lX_csum_replace() and others rather than passing a raw
3710          * buffer here. This one is a slow path helper and intended
3711          * for replies with control messages.
3712          *
3713          * Like in bpf_skb_change_proto(), we want to keep this rather
3714          * minimal and without protocol specifics so that we are able
3715          * to separate concerns as in bpf_skb_store_bytes() should only
3716          * be the one responsible for writing buffers.
3717          *
3718          * It's really expected to be a slow path operation here for
3719          * control message replies, so we're implicitly linearizing,
3720          * uncloning and drop offloads from the skb by this.
3721          */
3722         ret = __bpf_try_make_writable(skb, skb->len);
3723         if (!ret) {
3724                 if (new_len > skb->len)
3725                         ret = bpf_skb_grow_rcsum(skb, new_len);
3726                 else if (new_len < skb->len)
3727                         ret = bpf_skb_trim_rcsum(skb, new_len);
3728                 if (!ret && skb_is_gso(skb))
3729                         skb_gso_reset(skb);
3730         }
3731         return ret;
3732 }
3733
3734 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3735            u64, flags)
3736 {
3737         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3738
3739         bpf_compute_data_pointers(skb);
3740         return ret;
3741 }
3742
3743 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3744         .func           = bpf_skb_change_tail,
3745         .gpl_only       = false,
3746         .ret_type       = RET_INTEGER,
3747         .arg1_type      = ARG_PTR_TO_CTX,
3748         .arg2_type      = ARG_ANYTHING,
3749         .arg3_type      = ARG_ANYTHING,
3750 };
3751
3752 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3753            u64, flags)
3754 {
3755         return __bpf_skb_change_tail(skb, new_len, flags);
3756 }
3757
3758 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3759         .func           = sk_skb_change_tail,
3760         .gpl_only       = false,
3761         .ret_type       = RET_INTEGER,
3762         .arg1_type      = ARG_PTR_TO_CTX,
3763         .arg2_type      = ARG_ANYTHING,
3764         .arg3_type      = ARG_ANYTHING,
3765 };
3766
3767 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3768                                         u64 flags)
3769 {
3770         u32 max_len = BPF_SKB_MAX_LEN;
3771         u32 new_len = skb->len + head_room;
3772         int ret;
3773
3774         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3775                      new_len < skb->len))
3776                 return -EINVAL;
3777
3778         ret = skb_cow(skb, head_room);
3779         if (likely(!ret)) {
3780                 /* Idea for this helper is that we currently only
3781                  * allow to expand on mac header. This means that
3782                  * skb->protocol network header, etc, stay as is.
3783                  * Compared to bpf_skb_change_tail(), we're more
3784                  * flexible due to not needing to linearize or
3785                  * reset GSO. Intention for this helper is to be
3786                  * used by an L3 skb that needs to push mac header
3787                  * for redirection into L2 device.
3788                  */
3789                 __skb_push(skb, head_room);
3790                 memset(skb->data, 0, head_room);
3791                 skb_reset_mac_header(skb);
3792                 skb_reset_mac_len(skb);
3793         }
3794
3795         return ret;
3796 }
3797
3798 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3799            u64, flags)
3800 {
3801         int ret = __bpf_skb_change_head(skb, head_room, flags);
3802
3803         bpf_compute_data_pointers(skb);
3804         return ret;
3805 }
3806
3807 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3808         .func           = bpf_skb_change_head,
3809         .gpl_only       = false,
3810         .ret_type       = RET_INTEGER,
3811         .arg1_type      = ARG_PTR_TO_CTX,
3812         .arg2_type      = ARG_ANYTHING,
3813         .arg3_type      = ARG_ANYTHING,
3814 };
3815
3816 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3817            u64, flags)
3818 {
3819         return __bpf_skb_change_head(skb, head_room, flags);
3820 }
3821
3822 static const struct bpf_func_proto sk_skb_change_head_proto = {
3823         .func           = sk_skb_change_head,
3824         .gpl_only       = false,
3825         .ret_type       = RET_INTEGER,
3826         .arg1_type      = ARG_PTR_TO_CTX,
3827         .arg2_type      = ARG_ANYTHING,
3828         .arg3_type      = ARG_ANYTHING,
3829 };
3830
3831 BPF_CALL_1(bpf_xdp_get_buff_len, struct  xdp_buff*, xdp)
3832 {
3833         return xdp_get_buff_len(xdp);
3834 }
3835
3836 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3837         .func           = bpf_xdp_get_buff_len,
3838         .gpl_only       = false,
3839         .ret_type       = RET_INTEGER,
3840         .arg1_type      = ARG_PTR_TO_CTX,
3841 };
3842
3843 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3844
3845 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3846         .func           = bpf_xdp_get_buff_len,
3847         .gpl_only       = false,
3848         .arg1_type      = ARG_PTR_TO_BTF_ID,
3849         .arg1_btf_id    = &bpf_xdp_get_buff_len_bpf_ids[0],
3850 };
3851
3852 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3853 {
3854         return xdp_data_meta_unsupported(xdp) ? 0 :
3855                xdp->data - xdp->data_meta;
3856 }
3857
3858 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3859 {
3860         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3861         unsigned long metalen = xdp_get_metalen(xdp);
3862         void *data_start = xdp_frame_end + metalen;
3863         void *data = xdp->data + offset;
3864
3865         if (unlikely(data < data_start ||
3866                      data > xdp->data_end - ETH_HLEN))
3867                 return -EINVAL;
3868
3869         if (metalen)
3870                 memmove(xdp->data_meta + offset,
3871                         xdp->data_meta, metalen);
3872         xdp->data_meta += offset;
3873         xdp->data = data;
3874
3875         return 0;
3876 }
3877
3878 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3879         .func           = bpf_xdp_adjust_head,
3880         .gpl_only       = false,
3881         .ret_type       = RET_INTEGER,
3882         .arg1_type      = ARG_PTR_TO_CTX,
3883         .arg2_type      = ARG_ANYTHING,
3884 };
3885
3886 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3887                              void *buf, unsigned long len, bool flush)
3888 {
3889         unsigned long ptr_len, ptr_off = 0;
3890         skb_frag_t *next_frag, *end_frag;
3891         struct skb_shared_info *sinfo;
3892         void *src, *dst;
3893         u8 *ptr_buf;
3894
3895         if (likely(xdp->data_end - xdp->data >= off + len)) {
3896                 src = flush ? buf : xdp->data + off;
3897                 dst = flush ? xdp->data + off : buf;
3898                 memcpy(dst, src, len);
3899                 return;
3900         }
3901
3902         sinfo = xdp_get_shared_info_from_buff(xdp);
3903         end_frag = &sinfo->frags[sinfo->nr_frags];
3904         next_frag = &sinfo->frags[0];
3905
3906         ptr_len = xdp->data_end - xdp->data;
3907         ptr_buf = xdp->data;
3908
3909         while (true) {
3910                 if (off < ptr_off + ptr_len) {
3911                         unsigned long copy_off = off - ptr_off;
3912                         unsigned long copy_len = min(len, ptr_len - copy_off);
3913
3914                         src = flush ? buf : ptr_buf + copy_off;
3915                         dst = flush ? ptr_buf + copy_off : buf;
3916                         memcpy(dst, src, copy_len);
3917
3918                         off += copy_len;
3919                         len -= copy_len;
3920                         buf += copy_len;
3921                 }
3922
3923                 if (!len || next_frag == end_frag)
3924                         break;
3925
3926                 ptr_off += ptr_len;
3927                 ptr_buf = skb_frag_address(next_frag);
3928                 ptr_len = skb_frag_size(next_frag);
3929                 next_frag++;
3930         }
3931 }
3932
3933 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3934 {
3935         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3936         u32 size = xdp->data_end - xdp->data;
3937         void *addr = xdp->data;
3938         int i;
3939
3940         if (unlikely(offset > 0xffff || len > 0xffff))
3941                 return ERR_PTR(-EFAULT);
3942
3943         if (offset + len > xdp_get_buff_len(xdp))
3944                 return ERR_PTR(-EINVAL);
3945
3946         if (offset < size) /* linear area */
3947                 goto out;
3948
3949         offset -= size;
3950         for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3951                 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3952
3953                 if  (offset < frag_size) {
3954                         addr = skb_frag_address(&sinfo->frags[i]);
3955                         size = frag_size;
3956                         break;
3957                 }
3958                 offset -= frag_size;
3959         }
3960 out:
3961         return offset + len <= size ? addr + offset : NULL;
3962 }
3963
3964 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3965            void *, buf, u32, len)
3966 {
3967         void *ptr;
3968
3969         ptr = bpf_xdp_pointer(xdp, offset, len);
3970         if (IS_ERR(ptr))
3971                 return PTR_ERR(ptr);
3972
3973         if (!ptr)
3974                 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3975         else
3976                 memcpy(buf, ptr, len);
3977
3978         return 0;
3979 }
3980
3981 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3982         .func           = bpf_xdp_load_bytes,
3983         .gpl_only       = false,
3984         .ret_type       = RET_INTEGER,
3985         .arg1_type      = ARG_PTR_TO_CTX,
3986         .arg2_type      = ARG_ANYTHING,
3987         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
3988         .arg4_type      = ARG_CONST_SIZE,
3989 };
3990
3991 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3992            void *, buf, u32, len)
3993 {
3994         void *ptr;
3995
3996         ptr = bpf_xdp_pointer(xdp, offset, len);
3997         if (IS_ERR(ptr))
3998                 return PTR_ERR(ptr);
3999
4000         if (!ptr)
4001                 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
4002         else
4003                 memcpy(ptr, buf, len);
4004
4005         return 0;
4006 }
4007
4008 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
4009         .func           = bpf_xdp_store_bytes,
4010         .gpl_only       = false,
4011         .ret_type       = RET_INTEGER,
4012         .arg1_type      = ARG_PTR_TO_CTX,
4013         .arg2_type      = ARG_ANYTHING,
4014         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4015         .arg4_type      = ARG_CONST_SIZE,
4016 };
4017
4018 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
4019 {
4020         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4021         skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
4022         struct xdp_rxq_info *rxq = xdp->rxq;
4023         unsigned int tailroom;
4024
4025         if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
4026                 return -EOPNOTSUPP;
4027
4028         tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
4029         if (unlikely(offset > tailroom))
4030                 return -EINVAL;
4031
4032         memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
4033         skb_frag_size_add(frag, offset);
4034         sinfo->xdp_frags_size += offset;
4035
4036         return 0;
4037 }
4038
4039 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4040 {
4041         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4042         int i, n_frags_free = 0, len_free = 0;
4043
4044         if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4045                 return -EINVAL;
4046
4047         for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4048                 skb_frag_t *frag = &sinfo->frags[i];
4049                 int shrink = min_t(int, offset, skb_frag_size(frag));
4050
4051                 len_free += shrink;
4052                 offset -= shrink;
4053
4054                 if (skb_frag_size(frag) == shrink) {
4055                         struct page *page = skb_frag_page(frag);
4056
4057                         __xdp_return(page_address(page), &xdp->rxq->mem,
4058                                      false, NULL);
4059                         n_frags_free++;
4060                 } else {
4061                         skb_frag_size_sub(frag, shrink);
4062                         break;
4063                 }
4064         }
4065         sinfo->nr_frags -= n_frags_free;
4066         sinfo->xdp_frags_size -= len_free;
4067
4068         if (unlikely(!sinfo->nr_frags)) {
4069                 xdp_buff_clear_frags_flag(xdp);
4070                 xdp->data_end -= offset;
4071         }
4072
4073         return 0;
4074 }
4075
4076 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4077 {
4078         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4079         void *data_end = xdp->data_end + offset;
4080
4081         if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4082                 if (offset < 0)
4083                         return bpf_xdp_frags_shrink_tail(xdp, -offset);
4084
4085                 return bpf_xdp_frags_increase_tail(xdp, offset);
4086         }
4087
4088         /* Notice that xdp_data_hard_end have reserved some tailroom */
4089         if (unlikely(data_end > data_hard_end))
4090                 return -EINVAL;
4091
4092         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4093         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4094                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4095                 return -EINVAL;
4096         }
4097
4098         if (unlikely(data_end < xdp->data + ETH_HLEN))
4099                 return -EINVAL;
4100
4101         /* Clear memory area on grow, can contain uninit kernel memory */
4102         if (offset > 0)
4103                 memset(xdp->data_end, 0, offset);
4104
4105         xdp->data_end = data_end;
4106
4107         return 0;
4108 }
4109
4110 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4111         .func           = bpf_xdp_adjust_tail,
4112         .gpl_only       = false,
4113         .ret_type       = RET_INTEGER,
4114         .arg1_type      = ARG_PTR_TO_CTX,
4115         .arg2_type      = ARG_ANYTHING,
4116 };
4117
4118 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4119 {
4120         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4121         void *meta = xdp->data_meta + offset;
4122         unsigned long metalen = xdp->data - meta;
4123
4124         if (xdp_data_meta_unsupported(xdp))
4125                 return -ENOTSUPP;
4126         if (unlikely(meta < xdp_frame_end ||
4127                      meta > xdp->data))
4128                 return -EINVAL;
4129         if (unlikely(xdp_metalen_invalid(metalen)))
4130                 return -EACCES;
4131
4132         xdp->data_meta = meta;
4133
4134         return 0;
4135 }
4136
4137 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4138         .func           = bpf_xdp_adjust_meta,
4139         .gpl_only       = false,
4140         .ret_type       = RET_INTEGER,
4141         .arg1_type      = ARG_PTR_TO_CTX,
4142         .arg2_type      = ARG_ANYTHING,
4143 };
4144
4145 /**
4146  * DOC: xdp redirect
4147  *
4148  * XDP_REDIRECT works by a three-step process, implemented in the functions
4149  * below:
4150  *
4151  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4152  *    of the redirect and store it (along with some other metadata) in a per-CPU
4153  *    struct bpf_redirect_info.
4154  *
4155  * 2. When the program returns the XDP_REDIRECT return code, the driver will
4156  *    call xdp_do_redirect() which will use the information in struct
4157  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
4158  *    bulk queue structure.
4159  *
4160  * 3. Before exiting its NAPI poll loop, the driver will call
4161  *    xdp_do_flush(), which will flush all the different bulk queues,
4162  *    thus completing the redirect. Note that xdp_do_flush() must be
4163  *    called before napi_complete_done() in the driver, as the
4164  *    XDP_REDIRECT logic relies on being inside a single NAPI instance
4165  *    through to the xdp_do_flush() call for RCU protection of all
4166  *    in-kernel data structures.
4167  */
4168 /*
4169  * Pointers to the map entries will be kept around for this whole sequence of
4170  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4171  * the core code; instead, the RCU protection relies on everything happening
4172  * inside a single NAPI poll sequence, which means it's between a pair of calls
4173  * to local_bh_disable()/local_bh_enable().
4174  *
4175  * The map entries are marked as __rcu and the map code makes sure to
4176  * dereference those pointers with rcu_dereference_check() in a way that works
4177  * for both sections that to hold an rcu_read_lock() and sections that are
4178  * called from NAPI without a separate rcu_read_lock(). The code below does not
4179  * use RCU annotations, but relies on those in the map code.
4180  */
4181 void xdp_do_flush(void)
4182 {
4183         __dev_flush();
4184         __cpu_map_flush();
4185         __xsk_map_flush();
4186 }
4187 EXPORT_SYMBOL_GPL(xdp_do_flush);
4188
4189 void bpf_clear_redirect_map(struct bpf_map *map)
4190 {
4191         struct bpf_redirect_info *ri;
4192         int cpu;
4193
4194         for_each_possible_cpu(cpu) {
4195                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4196                 /* Avoid polluting remote cacheline due to writes if
4197                  * not needed. Once we pass this test, we need the
4198                  * cmpxchg() to make sure it hasn't been changed in
4199                  * the meantime by remote CPU.
4200                  */
4201                 if (unlikely(READ_ONCE(ri->map) == map))
4202                         cmpxchg(&ri->map, map, NULL);
4203         }
4204 }
4205
4206 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4207 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4208
4209 u32 xdp_master_redirect(struct xdp_buff *xdp)
4210 {
4211         struct net_device *master, *slave;
4212         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4213
4214         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4215         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4216         if (slave && slave != xdp->rxq->dev) {
4217                 /* The target device is different from the receiving device, so
4218                  * redirect it to the new device.
4219                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4220                  * drivers to unmap the packet from their rx ring.
4221                  */
4222                 ri->tgt_index = slave->ifindex;
4223                 ri->map_id = INT_MAX;
4224                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4225                 return XDP_REDIRECT;
4226         }
4227         return XDP_TX;
4228 }
4229 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4230
4231 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4232                                         struct net_device *dev,
4233                                         struct xdp_buff *xdp,
4234                                         struct bpf_prog *xdp_prog)
4235 {
4236         enum bpf_map_type map_type = ri->map_type;
4237         void *fwd = ri->tgt_value;
4238         u32 map_id = ri->map_id;
4239         int err;
4240
4241         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4242         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4243
4244         err = __xsk_map_redirect(fwd, xdp);
4245         if (unlikely(err))
4246                 goto err;
4247
4248         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4249         return 0;
4250 err:
4251         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4252         return err;
4253 }
4254
4255 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4256                                                    struct net_device *dev,
4257                                                    struct xdp_frame *xdpf,
4258                                                    struct bpf_prog *xdp_prog)
4259 {
4260         enum bpf_map_type map_type = ri->map_type;
4261         void *fwd = ri->tgt_value;
4262         u32 map_id = ri->map_id;
4263         struct bpf_map *map;
4264         int err;
4265
4266         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4267         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4268
4269         if (unlikely(!xdpf)) {
4270                 err = -EOVERFLOW;
4271                 goto err;
4272         }
4273
4274         switch (map_type) {
4275         case BPF_MAP_TYPE_DEVMAP:
4276                 fallthrough;
4277         case BPF_MAP_TYPE_DEVMAP_HASH:
4278                 map = READ_ONCE(ri->map);
4279                 if (unlikely(map)) {
4280                         WRITE_ONCE(ri->map, NULL);
4281                         err = dev_map_enqueue_multi(xdpf, dev, map,
4282                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
4283                 } else {
4284                         err = dev_map_enqueue(fwd, xdpf, dev);
4285                 }
4286                 break;
4287         case BPF_MAP_TYPE_CPUMAP:
4288                 err = cpu_map_enqueue(fwd, xdpf, dev);
4289                 break;
4290         case BPF_MAP_TYPE_UNSPEC:
4291                 if (map_id == INT_MAX) {
4292                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4293                         if (unlikely(!fwd)) {
4294                                 err = -EINVAL;
4295                                 break;
4296                         }
4297                         err = dev_xdp_enqueue(fwd, xdpf, dev);
4298                         break;
4299                 }
4300                 fallthrough;
4301         default:
4302                 err = -EBADRQC;
4303         }
4304
4305         if (unlikely(err))
4306                 goto err;
4307
4308         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4309         return 0;
4310 err:
4311         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4312         return err;
4313 }
4314
4315 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4316                     struct bpf_prog *xdp_prog)
4317 {
4318         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4319         enum bpf_map_type map_type = ri->map_type;
4320
4321         if (map_type == BPF_MAP_TYPE_XSKMAP) {
4322                 /* XDP_REDIRECT is not supported AF_XDP yet. */
4323                 if (unlikely(xdp_buff_has_frags(xdp)))
4324                         return -EOPNOTSUPP;
4325
4326                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4327         }
4328
4329         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4330                                        xdp_prog);
4331 }
4332 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4333
4334 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4335                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4336 {
4337         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4338         enum bpf_map_type map_type = ri->map_type;
4339
4340         if (map_type == BPF_MAP_TYPE_XSKMAP)
4341                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4342
4343         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4344 }
4345 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4346
4347 static int xdp_do_generic_redirect_map(struct net_device *dev,
4348                                        struct sk_buff *skb,
4349                                        struct xdp_buff *xdp,
4350                                        struct bpf_prog *xdp_prog,
4351                                        void *fwd,
4352                                        enum bpf_map_type map_type, u32 map_id)
4353 {
4354         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4355         struct bpf_map *map;
4356         int err;
4357
4358         switch (map_type) {
4359         case BPF_MAP_TYPE_DEVMAP:
4360                 fallthrough;
4361         case BPF_MAP_TYPE_DEVMAP_HASH:
4362                 map = READ_ONCE(ri->map);
4363                 if (unlikely(map)) {
4364                         WRITE_ONCE(ri->map, NULL);
4365                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4366                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4367                 } else {
4368                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4369                 }
4370                 if (unlikely(err))
4371                         goto err;
4372                 break;
4373         case BPF_MAP_TYPE_XSKMAP:
4374                 err = xsk_generic_rcv(fwd, xdp);
4375                 if (err)
4376                         goto err;
4377                 consume_skb(skb);
4378                 break;
4379         case BPF_MAP_TYPE_CPUMAP:
4380                 err = cpu_map_generic_redirect(fwd, skb);
4381                 if (unlikely(err))
4382                         goto err;
4383                 break;
4384         default:
4385                 err = -EBADRQC;
4386                 goto err;
4387         }
4388
4389         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4390         return 0;
4391 err:
4392         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4393         return err;
4394 }
4395
4396 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4397                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4398 {
4399         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4400         enum bpf_map_type map_type = ri->map_type;
4401         void *fwd = ri->tgt_value;
4402         u32 map_id = ri->map_id;
4403         int err;
4404
4405         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4406         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4407
4408         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4409                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4410                 if (unlikely(!fwd)) {
4411                         err = -EINVAL;
4412                         goto err;
4413                 }
4414
4415                 err = xdp_ok_fwd_dev(fwd, skb->len);
4416                 if (unlikely(err))
4417                         goto err;
4418
4419                 skb->dev = fwd;
4420                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4421                 generic_xdp_tx(skb, xdp_prog);
4422                 return 0;
4423         }
4424
4425         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4426 err:
4427         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4428         return err;
4429 }
4430
4431 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4432 {
4433         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4434
4435         if (unlikely(flags))
4436                 return XDP_ABORTED;
4437
4438         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4439          * by map_idr) is used for ifindex based XDP redirect.
4440          */
4441         ri->tgt_index = ifindex;
4442         ri->map_id = INT_MAX;
4443         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4444
4445         return XDP_REDIRECT;
4446 }
4447
4448 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4449         .func           = bpf_xdp_redirect,
4450         .gpl_only       = false,
4451         .ret_type       = RET_INTEGER,
4452         .arg1_type      = ARG_ANYTHING,
4453         .arg2_type      = ARG_ANYTHING,
4454 };
4455
4456 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key,
4457            u64, flags)
4458 {
4459         return map->ops->map_redirect(map, key, flags);
4460 }
4461
4462 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4463         .func           = bpf_xdp_redirect_map,
4464         .gpl_only       = false,
4465         .ret_type       = RET_INTEGER,
4466         .arg1_type      = ARG_CONST_MAP_PTR,
4467         .arg2_type      = ARG_ANYTHING,
4468         .arg3_type      = ARG_ANYTHING,
4469 };
4470
4471 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4472                                   unsigned long off, unsigned long len)
4473 {
4474         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4475
4476         if (unlikely(!ptr))
4477                 return len;
4478         if (ptr != dst_buff)
4479                 memcpy(dst_buff, ptr, len);
4480
4481         return 0;
4482 }
4483
4484 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4485            u64, flags, void *, meta, u64, meta_size)
4486 {
4487         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4488
4489         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4490                 return -EINVAL;
4491         if (unlikely(!skb || skb_size > skb->len))
4492                 return -EFAULT;
4493
4494         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4495                                 bpf_skb_copy);
4496 }
4497
4498 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4499         .func           = bpf_skb_event_output,
4500         .gpl_only       = true,
4501         .ret_type       = RET_INTEGER,
4502         .arg1_type      = ARG_PTR_TO_CTX,
4503         .arg2_type      = ARG_CONST_MAP_PTR,
4504         .arg3_type      = ARG_ANYTHING,
4505         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4506         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4507 };
4508
4509 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4510
4511 const struct bpf_func_proto bpf_skb_output_proto = {
4512         .func           = bpf_skb_event_output,
4513         .gpl_only       = true,
4514         .ret_type       = RET_INTEGER,
4515         .arg1_type      = ARG_PTR_TO_BTF_ID,
4516         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4517         .arg2_type      = ARG_CONST_MAP_PTR,
4518         .arg3_type      = ARG_ANYTHING,
4519         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4520         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4521 };
4522
4523 static unsigned short bpf_tunnel_key_af(u64 flags)
4524 {
4525         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4526 }
4527
4528 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4529            u32, size, u64, flags)
4530 {
4531         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4532         u8 compat[sizeof(struct bpf_tunnel_key)];
4533         void *to_orig = to;
4534         int err;
4535
4536         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4537                                          BPF_F_TUNINFO_FLAGS)))) {
4538                 err = -EINVAL;
4539                 goto err_clear;
4540         }
4541         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4542                 err = -EPROTO;
4543                 goto err_clear;
4544         }
4545         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4546                 err = -EINVAL;
4547                 switch (size) {
4548                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4549                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4550                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4551                         goto set_compat;
4552                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4553                         /* Fixup deprecated structure layouts here, so we have
4554                          * a common path later on.
4555                          */
4556                         if (ip_tunnel_info_af(info) != AF_INET)
4557                                 goto err_clear;
4558 set_compat:
4559                         to = (struct bpf_tunnel_key *)compat;
4560                         break;
4561                 default:
4562                         goto err_clear;
4563                 }
4564         }
4565
4566         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4567         to->tunnel_tos = info->key.tos;
4568         to->tunnel_ttl = info->key.ttl;
4569         if (flags & BPF_F_TUNINFO_FLAGS)
4570                 to->tunnel_flags = info->key.tun_flags;
4571         else
4572                 to->tunnel_ext = 0;
4573
4574         if (flags & BPF_F_TUNINFO_IPV6) {
4575                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4576                        sizeof(to->remote_ipv6));
4577                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4578                        sizeof(to->local_ipv6));
4579                 to->tunnel_label = be32_to_cpu(info->key.label);
4580         } else {
4581                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4582                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4583                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4584                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4585                 to->tunnel_label = 0;
4586         }
4587
4588         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4589                 memcpy(to_orig, to, size);
4590
4591         return 0;
4592 err_clear:
4593         memset(to_orig, 0, size);
4594         return err;
4595 }
4596
4597 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4598         .func           = bpf_skb_get_tunnel_key,
4599         .gpl_only       = false,
4600         .ret_type       = RET_INTEGER,
4601         .arg1_type      = ARG_PTR_TO_CTX,
4602         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4603         .arg3_type      = ARG_CONST_SIZE,
4604         .arg4_type      = ARG_ANYTHING,
4605 };
4606
4607 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4608 {
4609         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4610         int err;
4611
4612         if (unlikely(!info ||
4613                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4614                 err = -ENOENT;
4615                 goto err_clear;
4616         }
4617         if (unlikely(size < info->options_len)) {
4618                 err = -ENOMEM;
4619                 goto err_clear;
4620         }
4621
4622         ip_tunnel_info_opts_get(to, info);
4623         if (size > info->options_len)
4624                 memset(to + info->options_len, 0, size - info->options_len);
4625
4626         return info->options_len;
4627 err_clear:
4628         memset(to, 0, size);
4629         return err;
4630 }
4631
4632 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4633         .func           = bpf_skb_get_tunnel_opt,
4634         .gpl_only       = false,
4635         .ret_type       = RET_INTEGER,
4636         .arg1_type      = ARG_PTR_TO_CTX,
4637         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4638         .arg3_type      = ARG_CONST_SIZE,
4639 };
4640
4641 static struct metadata_dst __percpu *md_dst;
4642
4643 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4644            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4645 {
4646         struct metadata_dst *md = this_cpu_ptr(md_dst);
4647         u8 compat[sizeof(struct bpf_tunnel_key)];
4648         struct ip_tunnel_info *info;
4649
4650         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4651                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER |
4652                                BPF_F_NO_TUNNEL_KEY)))
4653                 return -EINVAL;
4654         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4655                 switch (size) {
4656                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4657                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4658                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4659                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4660                         /* Fixup deprecated structure layouts here, so we have
4661                          * a common path later on.
4662                          */
4663                         memcpy(compat, from, size);
4664                         memset(compat + size, 0, sizeof(compat) - size);
4665                         from = (const struct bpf_tunnel_key *) compat;
4666                         break;
4667                 default:
4668                         return -EINVAL;
4669                 }
4670         }
4671         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4672                      from->tunnel_ext))
4673                 return -EINVAL;
4674
4675         skb_dst_drop(skb);
4676         dst_hold((struct dst_entry *) md);
4677         skb_dst_set(skb, (struct dst_entry *) md);
4678
4679         info = &md->u.tun_info;
4680         memset(info, 0, sizeof(*info));
4681         info->mode = IP_TUNNEL_INFO_TX;
4682
4683         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4684         if (flags & BPF_F_DONT_FRAGMENT)
4685                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4686         if (flags & BPF_F_ZERO_CSUM_TX)
4687                 info->key.tun_flags &= ~TUNNEL_CSUM;
4688         if (flags & BPF_F_SEQ_NUMBER)
4689                 info->key.tun_flags |= TUNNEL_SEQ;
4690         if (flags & BPF_F_NO_TUNNEL_KEY)
4691                 info->key.tun_flags &= ~TUNNEL_KEY;
4692
4693         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4694         info->key.tos = from->tunnel_tos;
4695         info->key.ttl = from->tunnel_ttl;
4696
4697         if (flags & BPF_F_TUNINFO_IPV6) {
4698                 info->mode |= IP_TUNNEL_INFO_IPV6;
4699                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4700                        sizeof(from->remote_ipv6));
4701                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4702                        sizeof(from->local_ipv6));
4703                 info->key.label = cpu_to_be32(from->tunnel_label) &
4704                                   IPV6_FLOWLABEL_MASK;
4705         } else {
4706                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4707                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4708                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4709         }
4710
4711         return 0;
4712 }
4713
4714 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4715         .func           = bpf_skb_set_tunnel_key,
4716         .gpl_only       = false,
4717         .ret_type       = RET_INTEGER,
4718         .arg1_type      = ARG_PTR_TO_CTX,
4719         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4720         .arg3_type      = ARG_CONST_SIZE,
4721         .arg4_type      = ARG_ANYTHING,
4722 };
4723
4724 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4725            const u8 *, from, u32, size)
4726 {
4727         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4728         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4729
4730         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4731                 return -EINVAL;
4732         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4733                 return -ENOMEM;
4734
4735         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4736
4737         return 0;
4738 }
4739
4740 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4741         .func           = bpf_skb_set_tunnel_opt,
4742         .gpl_only       = false,
4743         .ret_type       = RET_INTEGER,
4744         .arg1_type      = ARG_PTR_TO_CTX,
4745         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4746         .arg3_type      = ARG_CONST_SIZE,
4747 };
4748
4749 static const struct bpf_func_proto *
4750 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4751 {
4752         if (!md_dst) {
4753                 struct metadata_dst __percpu *tmp;
4754
4755                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4756                                                 METADATA_IP_TUNNEL,
4757                                                 GFP_KERNEL);
4758                 if (!tmp)
4759                         return NULL;
4760                 if (cmpxchg(&md_dst, NULL, tmp))
4761                         metadata_dst_free_percpu(tmp);
4762         }
4763
4764         switch (which) {
4765         case BPF_FUNC_skb_set_tunnel_key:
4766                 return &bpf_skb_set_tunnel_key_proto;
4767         case BPF_FUNC_skb_set_tunnel_opt:
4768                 return &bpf_skb_set_tunnel_opt_proto;
4769         default:
4770                 return NULL;
4771         }
4772 }
4773
4774 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4775            u32, idx)
4776 {
4777         struct bpf_array *array = container_of(map, struct bpf_array, map);
4778         struct cgroup *cgrp;
4779         struct sock *sk;
4780
4781         sk = skb_to_full_sk(skb);
4782         if (!sk || !sk_fullsock(sk))
4783                 return -ENOENT;
4784         if (unlikely(idx >= array->map.max_entries))
4785                 return -E2BIG;
4786
4787         cgrp = READ_ONCE(array->ptrs[idx]);
4788         if (unlikely(!cgrp))
4789                 return -EAGAIN;
4790
4791         return sk_under_cgroup_hierarchy(sk, cgrp);
4792 }
4793
4794 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4795         .func           = bpf_skb_under_cgroup,
4796         .gpl_only       = false,
4797         .ret_type       = RET_INTEGER,
4798         .arg1_type      = ARG_PTR_TO_CTX,
4799         .arg2_type      = ARG_CONST_MAP_PTR,
4800         .arg3_type      = ARG_ANYTHING,
4801 };
4802
4803 #ifdef CONFIG_SOCK_CGROUP_DATA
4804 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4805 {
4806         struct cgroup *cgrp;
4807
4808         sk = sk_to_full_sk(sk);
4809         if (!sk || !sk_fullsock(sk))
4810                 return 0;
4811
4812         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4813         return cgroup_id(cgrp);
4814 }
4815
4816 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4817 {
4818         return __bpf_sk_cgroup_id(skb->sk);
4819 }
4820
4821 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4822         .func           = bpf_skb_cgroup_id,
4823         .gpl_only       = false,
4824         .ret_type       = RET_INTEGER,
4825         .arg1_type      = ARG_PTR_TO_CTX,
4826 };
4827
4828 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4829                                               int ancestor_level)
4830 {
4831         struct cgroup *ancestor;
4832         struct cgroup *cgrp;
4833
4834         sk = sk_to_full_sk(sk);
4835         if (!sk || !sk_fullsock(sk))
4836                 return 0;
4837
4838         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4839         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4840         if (!ancestor)
4841                 return 0;
4842
4843         return cgroup_id(ancestor);
4844 }
4845
4846 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4847            ancestor_level)
4848 {
4849         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4850 }
4851
4852 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4853         .func           = bpf_skb_ancestor_cgroup_id,
4854         .gpl_only       = false,
4855         .ret_type       = RET_INTEGER,
4856         .arg1_type      = ARG_PTR_TO_CTX,
4857         .arg2_type      = ARG_ANYTHING,
4858 };
4859
4860 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4861 {
4862         return __bpf_sk_cgroup_id(sk);
4863 }
4864
4865 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4866         .func           = bpf_sk_cgroup_id,
4867         .gpl_only       = false,
4868         .ret_type       = RET_INTEGER,
4869         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4870 };
4871
4872 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4873 {
4874         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4875 }
4876
4877 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4878         .func           = bpf_sk_ancestor_cgroup_id,
4879         .gpl_only       = false,
4880         .ret_type       = RET_INTEGER,
4881         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4882         .arg2_type      = ARG_ANYTHING,
4883 };
4884 #endif
4885
4886 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4887                                   unsigned long off, unsigned long len)
4888 {
4889         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4890
4891         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4892         return 0;
4893 }
4894
4895 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4896            u64, flags, void *, meta, u64, meta_size)
4897 {
4898         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4899
4900         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4901                 return -EINVAL;
4902
4903         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4904                 return -EFAULT;
4905
4906         return bpf_event_output(map, flags, meta, meta_size, xdp,
4907                                 xdp_size, bpf_xdp_copy);
4908 }
4909
4910 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4911         .func           = bpf_xdp_event_output,
4912         .gpl_only       = true,
4913         .ret_type       = RET_INTEGER,
4914         .arg1_type      = ARG_PTR_TO_CTX,
4915         .arg2_type      = ARG_CONST_MAP_PTR,
4916         .arg3_type      = ARG_ANYTHING,
4917         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4918         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4919 };
4920
4921 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4922
4923 const struct bpf_func_proto bpf_xdp_output_proto = {
4924         .func           = bpf_xdp_event_output,
4925         .gpl_only       = true,
4926         .ret_type       = RET_INTEGER,
4927         .arg1_type      = ARG_PTR_TO_BTF_ID,
4928         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4929         .arg2_type      = ARG_CONST_MAP_PTR,
4930         .arg3_type      = ARG_ANYTHING,
4931         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4932         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4933 };
4934
4935 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4936 {
4937         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4938 }
4939
4940 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4941         .func           = bpf_get_socket_cookie,
4942         .gpl_only       = false,
4943         .ret_type       = RET_INTEGER,
4944         .arg1_type      = ARG_PTR_TO_CTX,
4945 };
4946
4947 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4948 {
4949         return __sock_gen_cookie(ctx->sk);
4950 }
4951
4952 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4953         .func           = bpf_get_socket_cookie_sock_addr,
4954         .gpl_only       = false,
4955         .ret_type       = RET_INTEGER,
4956         .arg1_type      = ARG_PTR_TO_CTX,
4957 };
4958
4959 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4960 {
4961         return __sock_gen_cookie(ctx);
4962 }
4963
4964 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4965         .func           = bpf_get_socket_cookie_sock,
4966         .gpl_only       = false,
4967         .ret_type       = RET_INTEGER,
4968         .arg1_type      = ARG_PTR_TO_CTX,
4969 };
4970
4971 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4972 {
4973         return sk ? sock_gen_cookie(sk) : 0;
4974 }
4975
4976 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4977         .func           = bpf_get_socket_ptr_cookie,
4978         .gpl_only       = false,
4979         .ret_type       = RET_INTEGER,
4980         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4981 };
4982
4983 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4984 {
4985         return __sock_gen_cookie(ctx->sk);
4986 }
4987
4988 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4989         .func           = bpf_get_socket_cookie_sock_ops,
4990         .gpl_only       = false,
4991         .ret_type       = RET_INTEGER,
4992         .arg1_type      = ARG_PTR_TO_CTX,
4993 };
4994
4995 static u64 __bpf_get_netns_cookie(struct sock *sk)
4996 {
4997         const struct net *net = sk ? sock_net(sk) : &init_net;
4998
4999         return net->net_cookie;
5000 }
5001
5002 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
5003 {
5004         return __bpf_get_netns_cookie(ctx);
5005 }
5006
5007 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
5008         .func           = bpf_get_netns_cookie_sock,
5009         .gpl_only       = false,
5010         .ret_type       = RET_INTEGER,
5011         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5012 };
5013
5014 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
5015 {
5016         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5017 }
5018
5019 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
5020         .func           = bpf_get_netns_cookie_sock_addr,
5021         .gpl_only       = false,
5022         .ret_type       = RET_INTEGER,
5023         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5024 };
5025
5026 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
5027 {
5028         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5029 }
5030
5031 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
5032         .func           = bpf_get_netns_cookie_sock_ops,
5033         .gpl_only       = false,
5034         .ret_type       = RET_INTEGER,
5035         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5036 };
5037
5038 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
5039 {
5040         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
5041 }
5042
5043 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
5044         .func           = bpf_get_netns_cookie_sk_msg,
5045         .gpl_only       = false,
5046         .ret_type       = RET_INTEGER,
5047         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5048 };
5049
5050 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5051 {
5052         struct sock *sk = sk_to_full_sk(skb->sk);
5053         kuid_t kuid;
5054
5055         if (!sk || !sk_fullsock(sk))
5056                 return overflowuid;
5057         kuid = sock_net_uid(sock_net(sk), sk);
5058         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5059 }
5060
5061 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5062         .func           = bpf_get_socket_uid,
5063         .gpl_only       = false,
5064         .ret_type       = RET_INTEGER,
5065         .arg1_type      = ARG_PTR_TO_CTX,
5066 };
5067
5068 static int sol_socket_sockopt(struct sock *sk, int optname,
5069                               char *optval, int *optlen,
5070                               bool getopt)
5071 {
5072         switch (optname) {
5073         case SO_REUSEADDR:
5074         case SO_SNDBUF:
5075         case SO_RCVBUF:
5076         case SO_KEEPALIVE:
5077         case SO_PRIORITY:
5078         case SO_REUSEPORT:
5079         case SO_RCVLOWAT:
5080         case SO_MARK:
5081         case SO_MAX_PACING_RATE:
5082         case SO_BINDTOIFINDEX:
5083         case SO_TXREHASH:
5084                 if (*optlen != sizeof(int))
5085                         return -EINVAL;
5086                 break;
5087         case SO_BINDTODEVICE:
5088                 break;
5089         default:
5090                 return -EINVAL;
5091         }
5092
5093         if (getopt) {
5094                 if (optname == SO_BINDTODEVICE)
5095                         return -EINVAL;
5096                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5097                                      KERNEL_SOCKPTR(optval),
5098                                      KERNEL_SOCKPTR(optlen));
5099         }
5100
5101         return sk_setsockopt(sk, SOL_SOCKET, optname,
5102                              KERNEL_SOCKPTR(optval), *optlen);
5103 }
5104
5105 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5106                                   char *optval, int optlen)
5107 {
5108         struct tcp_sock *tp = tcp_sk(sk);
5109         unsigned long timeout;
5110         int val;
5111
5112         if (optlen != sizeof(int))
5113                 return -EINVAL;
5114
5115         val = *(int *)optval;
5116
5117         /* Only some options are supported */
5118         switch (optname) {
5119         case TCP_BPF_IW:
5120                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5121                         return -EINVAL;
5122                 tcp_snd_cwnd_set(tp, val);
5123                 break;
5124         case TCP_BPF_SNDCWND_CLAMP:
5125                 if (val <= 0)
5126                         return -EINVAL;
5127                 tp->snd_cwnd_clamp = val;
5128                 tp->snd_ssthresh = val;
5129                 break;
5130         case TCP_BPF_DELACK_MAX:
5131                 timeout = usecs_to_jiffies(val);
5132                 if (timeout > TCP_DELACK_MAX ||
5133                     timeout < TCP_TIMEOUT_MIN)
5134                         return -EINVAL;
5135                 inet_csk(sk)->icsk_delack_max = timeout;
5136                 break;
5137         case TCP_BPF_RTO_MIN:
5138                 timeout = usecs_to_jiffies(val);
5139                 if (timeout > TCP_RTO_MIN ||
5140                     timeout < TCP_TIMEOUT_MIN)
5141                         return -EINVAL;
5142                 inet_csk(sk)->icsk_rto_min = timeout;
5143                 break;
5144         default:
5145                 return -EINVAL;
5146         }
5147
5148         return 0;
5149 }
5150
5151 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval,
5152                                       int *optlen, bool getopt)
5153 {
5154         struct tcp_sock *tp;
5155         int ret;
5156
5157         if (*optlen < 2)
5158                 return -EINVAL;
5159
5160         if (getopt) {
5161                 if (!inet_csk(sk)->icsk_ca_ops)
5162                         return -EINVAL;
5163                 /* BPF expects NULL-terminated tcp-cc string */
5164                 optval[--(*optlen)] = '\0';
5165                 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION,
5166                                          KERNEL_SOCKPTR(optval),
5167                                          KERNEL_SOCKPTR(optlen));
5168         }
5169
5170         /* "cdg" is the only cc that alloc a ptr
5171          * in inet_csk_ca area.  The bpf-tcp-cc may
5172          * overwrite this ptr after switching to cdg.
5173          */
5174         if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen))
5175                 return -ENOTSUPP;
5176
5177         /* It stops this looping
5178          *
5179          * .init => bpf_setsockopt(tcp_cc) => .init =>
5180          * bpf_setsockopt(tcp_cc)" => .init => ....
5181          *
5182          * The second bpf_setsockopt(tcp_cc) is not allowed
5183          * in order to break the loop when both .init
5184          * are the same bpf prog.
5185          *
5186          * This applies even the second bpf_setsockopt(tcp_cc)
5187          * does not cause a loop.  This limits only the first
5188          * '.init' can call bpf_setsockopt(TCP_CONGESTION) to
5189          * pick a fallback cc (eg. peer does not support ECN)
5190          * and the second '.init' cannot fallback to
5191          * another.
5192          */
5193         tp = tcp_sk(sk);
5194         if (tp->bpf_chg_cc_inprogress)
5195                 return -EBUSY;
5196
5197         tp->bpf_chg_cc_inprogress = 1;
5198         ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION,
5199                                 KERNEL_SOCKPTR(optval), *optlen);
5200         tp->bpf_chg_cc_inprogress = 0;
5201         return ret;
5202 }
5203
5204 static int sol_tcp_sockopt(struct sock *sk, int optname,
5205                            char *optval, int *optlen,
5206                            bool getopt)
5207 {
5208         if (sk->sk_protocol != IPPROTO_TCP)
5209                 return -EINVAL;
5210
5211         switch (optname) {
5212         case TCP_NODELAY:
5213         case TCP_MAXSEG:
5214         case TCP_KEEPIDLE:
5215         case TCP_KEEPINTVL:
5216         case TCP_KEEPCNT:
5217         case TCP_SYNCNT:
5218         case TCP_WINDOW_CLAMP:
5219         case TCP_THIN_LINEAR_TIMEOUTS:
5220         case TCP_USER_TIMEOUT:
5221         case TCP_NOTSENT_LOWAT:
5222         case TCP_SAVE_SYN:
5223                 if (*optlen != sizeof(int))
5224                         return -EINVAL;
5225                 break;
5226         case TCP_CONGESTION:
5227                 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt);
5228         case TCP_SAVED_SYN:
5229                 if (*optlen < 1)
5230                         return -EINVAL;
5231                 break;
5232         default:
5233                 if (getopt)
5234                         return -EINVAL;
5235                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5236         }
5237
5238         if (getopt) {
5239                 if (optname == TCP_SAVED_SYN) {
5240                         struct tcp_sock *tp = tcp_sk(sk);
5241
5242                         if (!tp->saved_syn ||
5243                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5244                                 return -EINVAL;
5245                         memcpy(optval, tp->saved_syn->data, *optlen);
5246                         /* It cannot free tp->saved_syn here because it
5247                          * does not know if the user space still needs it.
5248                          */
5249                         return 0;
5250                 }
5251
5252                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5253                                          KERNEL_SOCKPTR(optval),
5254                                          KERNEL_SOCKPTR(optlen));
5255         }
5256
5257         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5258                                  KERNEL_SOCKPTR(optval), *optlen);
5259 }
5260
5261 static int sol_ip_sockopt(struct sock *sk, int optname,
5262                           char *optval, int *optlen,
5263                           bool getopt)
5264 {
5265         if (sk->sk_family != AF_INET)
5266                 return -EINVAL;
5267
5268         switch (optname) {
5269         case IP_TOS:
5270                 if (*optlen != sizeof(int))
5271                         return -EINVAL;
5272                 break;
5273         default:
5274                 return -EINVAL;
5275         }
5276
5277         if (getopt)
5278                 return do_ip_getsockopt(sk, SOL_IP, optname,
5279                                         KERNEL_SOCKPTR(optval),
5280                                         KERNEL_SOCKPTR(optlen));
5281
5282         return do_ip_setsockopt(sk, SOL_IP, optname,
5283                                 KERNEL_SOCKPTR(optval), *optlen);
5284 }
5285
5286 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5287                             char *optval, int *optlen,
5288                             bool getopt)
5289 {
5290         if (sk->sk_family != AF_INET6)
5291                 return -EINVAL;
5292
5293         switch (optname) {
5294         case IPV6_TCLASS:
5295         case IPV6_AUTOFLOWLABEL:
5296                 if (*optlen != sizeof(int))
5297                         return -EINVAL;
5298                 break;
5299         default:
5300                 return -EINVAL;
5301         }
5302
5303         if (getopt)
5304                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5305                                                       KERNEL_SOCKPTR(optval),
5306                                                       KERNEL_SOCKPTR(optlen));
5307
5308         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5309                                               KERNEL_SOCKPTR(optval), *optlen);
5310 }
5311
5312 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5313                             char *optval, int optlen)
5314 {
5315         if (!sk_fullsock(sk))
5316                 return -EINVAL;
5317
5318         if (level == SOL_SOCKET)
5319                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5320         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5321                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5322         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5323                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5324         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5325                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5326
5327         return -EINVAL;
5328 }
5329
5330 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5331                            char *optval, int optlen)
5332 {
5333         if (sk_fullsock(sk))
5334                 sock_owned_by_me(sk);
5335         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5336 }
5337
5338 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5339                             char *optval, int optlen)
5340 {
5341         int err, saved_optlen = optlen;
5342
5343         if (!sk_fullsock(sk)) {
5344                 err = -EINVAL;
5345                 goto done;
5346         }
5347
5348         if (level == SOL_SOCKET)
5349                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5350         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5351                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5352         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5353                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5354         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5355                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5356         else
5357                 err = -EINVAL;
5358
5359 done:
5360         if (err)
5361                 optlen = 0;
5362         if (optlen < saved_optlen)
5363                 memset(optval + optlen, 0, saved_optlen - optlen);
5364         return err;
5365 }
5366
5367 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5368                            char *optval, int optlen)
5369 {
5370         if (sk_fullsock(sk))
5371                 sock_owned_by_me(sk);
5372         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5373 }
5374
5375 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5376            int, optname, char *, optval, int, optlen)
5377 {
5378         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5379 }
5380
5381 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5382         .func           = bpf_sk_setsockopt,
5383         .gpl_only       = false,
5384         .ret_type       = RET_INTEGER,
5385         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5386         .arg2_type      = ARG_ANYTHING,
5387         .arg3_type      = ARG_ANYTHING,
5388         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5389         .arg5_type      = ARG_CONST_SIZE,
5390 };
5391
5392 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5393            int, optname, char *, optval, int, optlen)
5394 {
5395         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5396 }
5397
5398 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5399         .func           = bpf_sk_getsockopt,
5400         .gpl_only       = false,
5401         .ret_type       = RET_INTEGER,
5402         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5403         .arg2_type      = ARG_ANYTHING,
5404         .arg3_type      = ARG_ANYTHING,
5405         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5406         .arg5_type      = ARG_CONST_SIZE,
5407 };
5408
5409 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5410            int, optname, char *, optval, int, optlen)
5411 {
5412         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5413 }
5414
5415 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5416         .func           = bpf_unlocked_sk_setsockopt,
5417         .gpl_only       = false,
5418         .ret_type       = RET_INTEGER,
5419         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5420         .arg2_type      = ARG_ANYTHING,
5421         .arg3_type      = ARG_ANYTHING,
5422         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5423         .arg5_type      = ARG_CONST_SIZE,
5424 };
5425
5426 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5427            int, optname, char *, optval, int, optlen)
5428 {
5429         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5430 }
5431
5432 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5433         .func           = bpf_unlocked_sk_getsockopt,
5434         .gpl_only       = false,
5435         .ret_type       = RET_INTEGER,
5436         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5437         .arg2_type      = ARG_ANYTHING,
5438         .arg3_type      = ARG_ANYTHING,
5439         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5440         .arg5_type      = ARG_CONST_SIZE,
5441 };
5442
5443 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5444            int, level, int, optname, char *, optval, int, optlen)
5445 {
5446         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5447 }
5448
5449 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5450         .func           = bpf_sock_addr_setsockopt,
5451         .gpl_only       = false,
5452         .ret_type       = RET_INTEGER,
5453         .arg1_type      = ARG_PTR_TO_CTX,
5454         .arg2_type      = ARG_ANYTHING,
5455         .arg3_type      = ARG_ANYTHING,
5456         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5457         .arg5_type      = ARG_CONST_SIZE,
5458 };
5459
5460 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5461            int, level, int, optname, char *, optval, int, optlen)
5462 {
5463         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5464 }
5465
5466 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5467         .func           = bpf_sock_addr_getsockopt,
5468         .gpl_only       = false,
5469         .ret_type       = RET_INTEGER,
5470         .arg1_type      = ARG_PTR_TO_CTX,
5471         .arg2_type      = ARG_ANYTHING,
5472         .arg3_type      = ARG_ANYTHING,
5473         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5474         .arg5_type      = ARG_CONST_SIZE,
5475 };
5476
5477 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5478            int, level, int, optname, char *, optval, int, optlen)
5479 {
5480         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5481 }
5482
5483 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5484         .func           = bpf_sock_ops_setsockopt,
5485         .gpl_only       = false,
5486         .ret_type       = RET_INTEGER,
5487         .arg1_type      = ARG_PTR_TO_CTX,
5488         .arg2_type      = ARG_ANYTHING,
5489         .arg3_type      = ARG_ANYTHING,
5490         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5491         .arg5_type      = ARG_CONST_SIZE,
5492 };
5493
5494 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5495                                 int optname, const u8 **start)
5496 {
5497         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5498         const u8 *hdr_start;
5499         int ret;
5500
5501         if (syn_skb) {
5502                 /* sk is a request_sock here */
5503
5504                 if (optname == TCP_BPF_SYN) {
5505                         hdr_start = syn_skb->data;
5506                         ret = tcp_hdrlen(syn_skb);
5507                 } else if (optname == TCP_BPF_SYN_IP) {
5508                         hdr_start = skb_network_header(syn_skb);
5509                         ret = skb_network_header_len(syn_skb) +
5510                                 tcp_hdrlen(syn_skb);
5511                 } else {
5512                         /* optname == TCP_BPF_SYN_MAC */
5513                         hdr_start = skb_mac_header(syn_skb);
5514                         ret = skb_mac_header_len(syn_skb) +
5515                                 skb_network_header_len(syn_skb) +
5516                                 tcp_hdrlen(syn_skb);
5517                 }
5518         } else {
5519                 struct sock *sk = bpf_sock->sk;
5520                 struct saved_syn *saved_syn;
5521
5522                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5523                         /* synack retransmit. bpf_sock->syn_skb will
5524                          * not be available.  It has to resort to
5525                          * saved_syn (if it is saved).
5526                          */
5527                         saved_syn = inet_reqsk(sk)->saved_syn;
5528                 else
5529                         saved_syn = tcp_sk(sk)->saved_syn;
5530
5531                 if (!saved_syn)
5532                         return -ENOENT;
5533
5534                 if (optname == TCP_BPF_SYN) {
5535                         hdr_start = saved_syn->data +
5536                                 saved_syn->mac_hdrlen +
5537                                 saved_syn->network_hdrlen;
5538                         ret = saved_syn->tcp_hdrlen;
5539                 } else if (optname == TCP_BPF_SYN_IP) {
5540                         hdr_start = saved_syn->data +
5541                                 saved_syn->mac_hdrlen;
5542                         ret = saved_syn->network_hdrlen +
5543                                 saved_syn->tcp_hdrlen;
5544                 } else {
5545                         /* optname == TCP_BPF_SYN_MAC */
5546
5547                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5548                         if (!saved_syn->mac_hdrlen)
5549                                 return -ENOENT;
5550
5551                         hdr_start = saved_syn->data;
5552                         ret = saved_syn->mac_hdrlen +
5553                                 saved_syn->network_hdrlen +
5554                                 saved_syn->tcp_hdrlen;
5555                 }
5556         }
5557
5558         *start = hdr_start;
5559         return ret;
5560 }
5561
5562 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5563            int, level, int, optname, char *, optval, int, optlen)
5564 {
5565         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5566             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5567                 int ret, copy_len = 0;
5568                 const u8 *start;
5569
5570                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5571                 if (ret > 0) {
5572                         copy_len = ret;
5573                         if (optlen < copy_len) {
5574                                 copy_len = optlen;
5575                                 ret = -ENOSPC;
5576                         }
5577
5578                         memcpy(optval, start, copy_len);
5579                 }
5580
5581                 /* Zero out unused buffer at the end */
5582                 memset(optval + copy_len, 0, optlen - copy_len);
5583
5584                 return ret;
5585         }
5586
5587         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5588 }
5589
5590 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5591         .func           = bpf_sock_ops_getsockopt,
5592         .gpl_only       = false,
5593         .ret_type       = RET_INTEGER,
5594         .arg1_type      = ARG_PTR_TO_CTX,
5595         .arg2_type      = ARG_ANYTHING,
5596         .arg3_type      = ARG_ANYTHING,
5597         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5598         .arg5_type      = ARG_CONST_SIZE,
5599 };
5600
5601 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5602            int, argval)
5603 {
5604         struct sock *sk = bpf_sock->sk;
5605         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5606
5607         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5608                 return -EINVAL;
5609
5610         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5611
5612         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5613 }
5614
5615 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5616         .func           = bpf_sock_ops_cb_flags_set,
5617         .gpl_only       = false,
5618         .ret_type       = RET_INTEGER,
5619         .arg1_type      = ARG_PTR_TO_CTX,
5620         .arg2_type      = ARG_ANYTHING,
5621 };
5622
5623 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5624 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5625
5626 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5627            int, addr_len)
5628 {
5629 #ifdef CONFIG_INET
5630         struct sock *sk = ctx->sk;
5631         u32 flags = BIND_FROM_BPF;
5632         int err;
5633
5634         err = -EINVAL;
5635         if (addr_len < offsetofend(struct sockaddr, sa_family))
5636                 return err;
5637         if (addr->sa_family == AF_INET) {
5638                 if (addr_len < sizeof(struct sockaddr_in))
5639                         return err;
5640                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5641                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5642                 return __inet_bind(sk, addr, addr_len, flags);
5643 #if IS_ENABLED(CONFIG_IPV6)
5644         } else if (addr->sa_family == AF_INET6) {
5645                 if (addr_len < SIN6_LEN_RFC2133)
5646                         return err;
5647                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5648                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5649                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5650                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5651                  */
5652                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5653 #endif /* CONFIG_IPV6 */
5654         }
5655 #endif /* CONFIG_INET */
5656
5657         return -EAFNOSUPPORT;
5658 }
5659
5660 static const struct bpf_func_proto bpf_bind_proto = {
5661         .func           = bpf_bind,
5662         .gpl_only       = false,
5663         .ret_type       = RET_INTEGER,
5664         .arg1_type      = ARG_PTR_TO_CTX,
5665         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5666         .arg3_type      = ARG_CONST_SIZE,
5667 };
5668
5669 #ifdef CONFIG_XFRM
5670
5671 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \
5672     (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES))
5673
5674 struct metadata_dst __percpu *xfrm_bpf_md_dst;
5675 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst);
5676
5677 #endif
5678
5679 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5680            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5681 {
5682         const struct sec_path *sp = skb_sec_path(skb);
5683         const struct xfrm_state *x;
5684
5685         if (!sp || unlikely(index >= sp->len || flags))
5686                 goto err_clear;
5687
5688         x = sp->xvec[index];
5689
5690         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5691                 goto err_clear;
5692
5693         to->reqid = x->props.reqid;
5694         to->spi = x->id.spi;
5695         to->family = x->props.family;
5696         to->ext = 0;
5697
5698         if (to->family == AF_INET6) {
5699                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5700                        sizeof(to->remote_ipv6));
5701         } else {
5702                 to->remote_ipv4 = x->props.saddr.a4;
5703                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5704         }
5705
5706         return 0;
5707 err_clear:
5708         memset(to, 0, size);
5709         return -EINVAL;
5710 }
5711
5712 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5713         .func           = bpf_skb_get_xfrm_state,
5714         .gpl_only       = false,
5715         .ret_type       = RET_INTEGER,
5716         .arg1_type      = ARG_PTR_TO_CTX,
5717         .arg2_type      = ARG_ANYTHING,
5718         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5719         .arg4_type      = ARG_CONST_SIZE,
5720         .arg5_type      = ARG_ANYTHING,
5721 };
5722 #endif
5723
5724 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5725 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5726                                   const struct neighbour *neigh,
5727                                   const struct net_device *dev, u32 mtu)
5728 {
5729         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5730         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5731         params->h_vlan_TCI = 0;
5732         params->h_vlan_proto = 0;
5733         if (mtu)
5734                 params->mtu_result = mtu; /* union with tot_len */
5735
5736         return 0;
5737 }
5738 #endif
5739
5740 #if IS_ENABLED(CONFIG_INET)
5741 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5742                                u32 flags, bool check_mtu)
5743 {
5744         struct fib_nh_common *nhc;
5745         struct in_device *in_dev;
5746         struct neighbour *neigh;
5747         struct net_device *dev;
5748         struct fib_result res;
5749         struct flowi4 fl4;
5750         u32 mtu = 0;
5751         int err;
5752
5753         dev = dev_get_by_index_rcu(net, params->ifindex);
5754         if (unlikely(!dev))
5755                 return -ENODEV;
5756
5757         /* verify forwarding is enabled on this interface */
5758         in_dev = __in_dev_get_rcu(dev);
5759         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5760                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5761
5762         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5763                 fl4.flowi4_iif = 1;
5764                 fl4.flowi4_oif = params->ifindex;
5765         } else {
5766                 fl4.flowi4_iif = params->ifindex;
5767                 fl4.flowi4_oif = 0;
5768         }
5769         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5770         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5771         fl4.flowi4_flags = 0;
5772
5773         fl4.flowi4_proto = params->l4_protocol;
5774         fl4.daddr = params->ipv4_dst;
5775         fl4.saddr = params->ipv4_src;
5776         fl4.fl4_sport = params->sport;
5777         fl4.fl4_dport = params->dport;
5778         fl4.flowi4_multipath_hash = 0;
5779
5780         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5781                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5782                 struct fib_table *tb;
5783
5784                 tb = fib_get_table(net, tbid);
5785                 if (unlikely(!tb))
5786                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5787
5788                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5789         } else {
5790                 fl4.flowi4_mark = 0;
5791                 fl4.flowi4_secid = 0;
5792                 fl4.flowi4_tun_key.tun_id = 0;
5793                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5794
5795                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5796         }
5797
5798         if (err) {
5799                 /* map fib lookup errors to RTN_ type */
5800                 if (err == -EINVAL)
5801                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5802                 if (err == -EHOSTUNREACH)
5803                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5804                 if (err == -EACCES)
5805                         return BPF_FIB_LKUP_RET_PROHIBIT;
5806
5807                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5808         }
5809
5810         if (res.type != RTN_UNICAST)
5811                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5812
5813         if (fib_info_num_path(res.fi) > 1)
5814                 fib_select_path(net, &res, &fl4, NULL);
5815
5816         if (check_mtu) {
5817                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5818                 if (params->tot_len > mtu) {
5819                         params->mtu_result = mtu; /* union with tot_len */
5820                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5821                 }
5822         }
5823
5824         nhc = res.nhc;
5825
5826         /* do not handle lwt encaps right now */
5827         if (nhc->nhc_lwtstate)
5828                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5829
5830         dev = nhc->nhc_dev;
5831
5832         params->rt_metric = res.fi->fib_priority;
5833         params->ifindex = dev->ifindex;
5834
5835         /* xdp and cls_bpf programs are run in RCU-bh so
5836          * rcu_read_lock_bh is not needed here
5837          */
5838         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5839                 if (nhc->nhc_gw_family)
5840                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5841
5842                 neigh = __ipv4_neigh_lookup_noref(dev,
5843                                                  (__force u32)params->ipv4_dst);
5844         } else {
5845                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5846
5847                 params->family = AF_INET6;
5848                 *dst = nhc->nhc_gw.ipv6;
5849                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5850         }
5851
5852         if (!neigh)
5853                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5854
5855         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5856 }
5857 #endif
5858
5859 #if IS_ENABLED(CONFIG_IPV6)
5860 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5861                                u32 flags, bool check_mtu)
5862 {
5863         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5864         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5865         struct fib6_result res = {};
5866         struct neighbour *neigh;
5867         struct net_device *dev;
5868         struct inet6_dev *idev;
5869         struct flowi6 fl6;
5870         int strict = 0;
5871         int oif, err;
5872         u32 mtu = 0;
5873
5874         /* link local addresses are never forwarded */
5875         if (rt6_need_strict(dst) || rt6_need_strict(src))
5876                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5877
5878         dev = dev_get_by_index_rcu(net, params->ifindex);
5879         if (unlikely(!dev))
5880                 return -ENODEV;
5881
5882         idev = __in6_dev_get_safely(dev);
5883         if (unlikely(!idev || !idev->cnf.forwarding))
5884                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5885
5886         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5887                 fl6.flowi6_iif = 1;
5888                 oif = fl6.flowi6_oif = params->ifindex;
5889         } else {
5890                 oif = fl6.flowi6_iif = params->ifindex;
5891                 fl6.flowi6_oif = 0;
5892                 strict = RT6_LOOKUP_F_HAS_SADDR;
5893         }
5894         fl6.flowlabel = params->flowinfo;
5895         fl6.flowi6_scope = 0;
5896         fl6.flowi6_flags = 0;
5897         fl6.mp_hash = 0;
5898
5899         fl6.flowi6_proto = params->l4_protocol;
5900         fl6.daddr = *dst;
5901         fl6.saddr = *src;
5902         fl6.fl6_sport = params->sport;
5903         fl6.fl6_dport = params->dport;
5904
5905         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5906                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5907                 struct fib6_table *tb;
5908
5909                 tb = ipv6_stub->fib6_get_table(net, tbid);
5910                 if (unlikely(!tb))
5911                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5912
5913                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5914                                                    strict);
5915         } else {
5916                 fl6.flowi6_mark = 0;
5917                 fl6.flowi6_secid = 0;
5918                 fl6.flowi6_tun_key.tun_id = 0;
5919                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5920
5921                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5922         }
5923
5924         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5925                      res.f6i == net->ipv6.fib6_null_entry))
5926                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5927
5928         switch (res.fib6_type) {
5929         /* only unicast is forwarded */
5930         case RTN_UNICAST:
5931                 break;
5932         case RTN_BLACKHOLE:
5933                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5934         case RTN_UNREACHABLE:
5935                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5936         case RTN_PROHIBIT:
5937                 return BPF_FIB_LKUP_RET_PROHIBIT;
5938         default:
5939                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5940         }
5941
5942         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5943                                     fl6.flowi6_oif != 0, NULL, strict);
5944
5945         if (check_mtu) {
5946                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5947                 if (params->tot_len > mtu) {
5948                         params->mtu_result = mtu; /* union with tot_len */
5949                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5950                 }
5951         }
5952
5953         if (res.nh->fib_nh_lws)
5954                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5955
5956         if (res.nh->fib_nh_gw_family)
5957                 *dst = res.nh->fib_nh_gw6;
5958
5959         dev = res.nh->fib_nh_dev;
5960         params->rt_metric = res.f6i->fib6_metric;
5961         params->ifindex = dev->ifindex;
5962
5963         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5964          * not needed here.
5965          */
5966         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5967         if (!neigh)
5968                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5969
5970         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5971 }
5972 #endif
5973
5974 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5975            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5976 {
5977         if (plen < sizeof(*params))
5978                 return -EINVAL;
5979
5980         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5981                 return -EINVAL;
5982
5983         switch (params->family) {
5984 #if IS_ENABLED(CONFIG_INET)
5985         case AF_INET:
5986                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5987                                            flags, true);
5988 #endif
5989 #if IS_ENABLED(CONFIG_IPV6)
5990         case AF_INET6:
5991                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5992                                            flags, true);
5993 #endif
5994         }
5995         return -EAFNOSUPPORT;
5996 }
5997
5998 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5999         .func           = bpf_xdp_fib_lookup,
6000         .gpl_only       = true,
6001         .ret_type       = RET_INTEGER,
6002         .arg1_type      = ARG_PTR_TO_CTX,
6003         .arg2_type      = ARG_PTR_TO_MEM,
6004         .arg3_type      = ARG_CONST_SIZE,
6005         .arg4_type      = ARG_ANYTHING,
6006 };
6007
6008 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
6009            struct bpf_fib_lookup *, params, int, plen, u32, flags)
6010 {
6011         struct net *net = dev_net(skb->dev);
6012         int rc = -EAFNOSUPPORT;
6013         bool check_mtu = false;
6014
6015         if (plen < sizeof(*params))
6016                 return -EINVAL;
6017
6018         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
6019                 return -EINVAL;
6020
6021         if (params->tot_len)
6022                 check_mtu = true;
6023
6024         switch (params->family) {
6025 #if IS_ENABLED(CONFIG_INET)
6026         case AF_INET:
6027                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
6028                 break;
6029 #endif
6030 #if IS_ENABLED(CONFIG_IPV6)
6031         case AF_INET6:
6032                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
6033                 break;
6034 #endif
6035         }
6036
6037         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
6038                 struct net_device *dev;
6039
6040                 /* When tot_len isn't provided by user, check skb
6041                  * against MTU of FIB lookup resulting net_device
6042                  */
6043                 dev = dev_get_by_index_rcu(net, params->ifindex);
6044                 if (!is_skb_forwardable(dev, skb))
6045                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
6046
6047                 params->mtu_result = dev->mtu; /* union with tot_len */
6048         }
6049
6050         return rc;
6051 }
6052
6053 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
6054         .func           = bpf_skb_fib_lookup,
6055         .gpl_only       = true,
6056         .ret_type       = RET_INTEGER,
6057         .arg1_type      = ARG_PTR_TO_CTX,
6058         .arg2_type      = ARG_PTR_TO_MEM,
6059         .arg3_type      = ARG_CONST_SIZE,
6060         .arg4_type      = ARG_ANYTHING,
6061 };
6062
6063 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
6064                                             u32 ifindex)
6065 {
6066         struct net *netns = dev_net(dev_curr);
6067
6068         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
6069         if (ifindex == 0)
6070                 return dev_curr;
6071
6072         return dev_get_by_index_rcu(netns, ifindex);
6073 }
6074
6075 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
6076            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6077 {
6078         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6079         struct net_device *dev = skb->dev;
6080         int skb_len, dev_len;
6081         int mtu;
6082
6083         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
6084                 return -EINVAL;
6085
6086         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
6087                 return -EINVAL;
6088
6089         dev = __dev_via_ifindex(dev, ifindex);
6090         if (unlikely(!dev))
6091                 return -ENODEV;
6092
6093         mtu = READ_ONCE(dev->mtu);
6094
6095         dev_len = mtu + dev->hard_header_len;
6096
6097         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6098         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6099
6100         skb_len += len_diff; /* minus result pass check */
6101         if (skb_len <= dev_len) {
6102                 ret = BPF_MTU_CHK_RET_SUCCESS;
6103                 goto out;
6104         }
6105         /* At this point, skb->len exceed MTU, but as it include length of all
6106          * segments, it can still be below MTU.  The SKB can possibly get
6107          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6108          * must choose if segs are to be MTU checked.
6109          */
6110         if (skb_is_gso(skb)) {
6111                 ret = BPF_MTU_CHK_RET_SUCCESS;
6112
6113                 if (flags & BPF_MTU_CHK_SEGS &&
6114                     !skb_gso_validate_network_len(skb, mtu))
6115                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6116         }
6117 out:
6118         /* BPF verifier guarantees valid pointer */
6119         *mtu_len = mtu;
6120
6121         return ret;
6122 }
6123
6124 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6125            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6126 {
6127         struct net_device *dev = xdp->rxq->dev;
6128         int xdp_len = xdp->data_end - xdp->data;
6129         int ret = BPF_MTU_CHK_RET_SUCCESS;
6130         int mtu, dev_len;
6131
6132         /* XDP variant doesn't support multi-buffer segment check (yet) */
6133         if (unlikely(flags))
6134                 return -EINVAL;
6135
6136         dev = __dev_via_ifindex(dev, ifindex);
6137         if (unlikely(!dev))
6138                 return -ENODEV;
6139
6140         mtu = READ_ONCE(dev->mtu);
6141
6142         /* Add L2-header as dev MTU is L3 size */
6143         dev_len = mtu + dev->hard_header_len;
6144
6145         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6146         if (*mtu_len)
6147                 xdp_len = *mtu_len + dev->hard_header_len;
6148
6149         xdp_len += len_diff; /* minus result pass check */
6150         if (xdp_len > dev_len)
6151                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6152
6153         /* BPF verifier guarantees valid pointer */
6154         *mtu_len = mtu;
6155
6156         return ret;
6157 }
6158
6159 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6160         .func           = bpf_skb_check_mtu,
6161         .gpl_only       = true,
6162         .ret_type       = RET_INTEGER,
6163         .arg1_type      = ARG_PTR_TO_CTX,
6164         .arg2_type      = ARG_ANYTHING,
6165         .arg3_type      = ARG_PTR_TO_INT,
6166         .arg4_type      = ARG_ANYTHING,
6167         .arg5_type      = ARG_ANYTHING,
6168 };
6169
6170 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6171         .func           = bpf_xdp_check_mtu,
6172         .gpl_only       = true,
6173         .ret_type       = RET_INTEGER,
6174         .arg1_type      = ARG_PTR_TO_CTX,
6175         .arg2_type      = ARG_ANYTHING,
6176         .arg3_type      = ARG_PTR_TO_INT,
6177         .arg4_type      = ARG_ANYTHING,
6178         .arg5_type      = ARG_ANYTHING,
6179 };
6180
6181 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6182 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6183 {
6184         int err;
6185         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6186
6187         if (!seg6_validate_srh(srh, len, false))
6188                 return -EINVAL;
6189
6190         switch (type) {
6191         case BPF_LWT_ENCAP_SEG6_INLINE:
6192                 if (skb->protocol != htons(ETH_P_IPV6))
6193                         return -EBADMSG;
6194
6195                 err = seg6_do_srh_inline(skb, srh);
6196                 break;
6197         case BPF_LWT_ENCAP_SEG6:
6198                 skb_reset_inner_headers(skb);
6199                 skb->encapsulation = 1;
6200                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6201                 break;
6202         default:
6203                 return -EINVAL;
6204         }
6205
6206         bpf_compute_data_pointers(skb);
6207         if (err)
6208                 return err;
6209
6210         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6211
6212         return seg6_lookup_nexthop(skb, NULL, 0);
6213 }
6214 #endif /* CONFIG_IPV6_SEG6_BPF */
6215
6216 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6217 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6218                              bool ingress)
6219 {
6220         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6221 }
6222 #endif
6223
6224 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6225            u32, len)
6226 {
6227         switch (type) {
6228 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6229         case BPF_LWT_ENCAP_SEG6:
6230         case BPF_LWT_ENCAP_SEG6_INLINE:
6231                 return bpf_push_seg6_encap(skb, type, hdr, len);
6232 #endif
6233 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6234         case BPF_LWT_ENCAP_IP:
6235                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6236 #endif
6237         default:
6238                 return -EINVAL;
6239         }
6240 }
6241
6242 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6243            void *, hdr, u32, len)
6244 {
6245         switch (type) {
6246 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6247         case BPF_LWT_ENCAP_IP:
6248                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6249 #endif
6250         default:
6251                 return -EINVAL;
6252         }
6253 }
6254
6255 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6256         .func           = bpf_lwt_in_push_encap,
6257         .gpl_only       = false,
6258         .ret_type       = RET_INTEGER,
6259         .arg1_type      = ARG_PTR_TO_CTX,
6260         .arg2_type      = ARG_ANYTHING,
6261         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6262         .arg4_type      = ARG_CONST_SIZE
6263 };
6264
6265 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6266         .func           = bpf_lwt_xmit_push_encap,
6267         .gpl_only       = false,
6268         .ret_type       = RET_INTEGER,
6269         .arg1_type      = ARG_PTR_TO_CTX,
6270         .arg2_type      = ARG_ANYTHING,
6271         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6272         .arg4_type      = ARG_CONST_SIZE
6273 };
6274
6275 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6276 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6277            const void *, from, u32, len)
6278 {
6279         struct seg6_bpf_srh_state *srh_state =
6280                 this_cpu_ptr(&seg6_bpf_srh_states);
6281         struct ipv6_sr_hdr *srh = srh_state->srh;
6282         void *srh_tlvs, *srh_end, *ptr;
6283         int srhoff = 0;
6284
6285         if (srh == NULL)
6286                 return -EINVAL;
6287
6288         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6289         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6290
6291         ptr = skb->data + offset;
6292         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6293                 srh_state->valid = false;
6294         else if (ptr < (void *)&srh->flags ||
6295                  ptr + len > (void *)&srh->segments)
6296                 return -EFAULT;
6297
6298         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6299                 return -EFAULT;
6300         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6301                 return -EINVAL;
6302         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6303
6304         memcpy(skb->data + offset, from, len);
6305         return 0;
6306 }
6307
6308 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6309         .func           = bpf_lwt_seg6_store_bytes,
6310         .gpl_only       = false,
6311         .ret_type       = RET_INTEGER,
6312         .arg1_type      = ARG_PTR_TO_CTX,
6313         .arg2_type      = ARG_ANYTHING,
6314         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6315         .arg4_type      = ARG_CONST_SIZE
6316 };
6317
6318 static void bpf_update_srh_state(struct sk_buff *skb)
6319 {
6320         struct seg6_bpf_srh_state *srh_state =
6321                 this_cpu_ptr(&seg6_bpf_srh_states);
6322         int srhoff = 0;
6323
6324         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6325                 srh_state->srh = NULL;
6326         } else {
6327                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6328                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6329                 srh_state->valid = true;
6330         }
6331 }
6332
6333 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6334            u32, action, void *, param, u32, param_len)
6335 {
6336         struct seg6_bpf_srh_state *srh_state =
6337                 this_cpu_ptr(&seg6_bpf_srh_states);
6338         int hdroff = 0;
6339         int err;
6340
6341         switch (action) {
6342         case SEG6_LOCAL_ACTION_END_X:
6343                 if (!seg6_bpf_has_valid_srh(skb))
6344                         return -EBADMSG;
6345                 if (param_len != sizeof(struct in6_addr))
6346                         return -EINVAL;
6347                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6348         case SEG6_LOCAL_ACTION_END_T:
6349                 if (!seg6_bpf_has_valid_srh(skb))
6350                         return -EBADMSG;
6351                 if (param_len != sizeof(int))
6352                         return -EINVAL;
6353                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6354         case SEG6_LOCAL_ACTION_END_DT6:
6355                 if (!seg6_bpf_has_valid_srh(skb))
6356                         return -EBADMSG;
6357                 if (param_len != sizeof(int))
6358                         return -EINVAL;
6359
6360                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6361                         return -EBADMSG;
6362                 if (!pskb_pull(skb, hdroff))
6363                         return -EBADMSG;
6364
6365                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6366                 skb_reset_network_header(skb);
6367                 skb_reset_transport_header(skb);
6368                 skb->encapsulation = 0;
6369
6370                 bpf_compute_data_pointers(skb);
6371                 bpf_update_srh_state(skb);
6372                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6373         case SEG6_LOCAL_ACTION_END_B6:
6374                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6375                         return -EBADMSG;
6376                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6377                                           param, param_len);
6378                 if (!err)
6379                         bpf_update_srh_state(skb);
6380
6381                 return err;
6382         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6383                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6384                         return -EBADMSG;
6385                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6386                                           param, param_len);
6387                 if (!err)
6388                         bpf_update_srh_state(skb);
6389
6390                 return err;
6391         default:
6392                 return -EINVAL;
6393         }
6394 }
6395
6396 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6397         .func           = bpf_lwt_seg6_action,
6398         .gpl_only       = false,
6399         .ret_type       = RET_INTEGER,
6400         .arg1_type      = ARG_PTR_TO_CTX,
6401         .arg2_type      = ARG_ANYTHING,
6402         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6403         .arg4_type      = ARG_CONST_SIZE
6404 };
6405
6406 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6407            s32, len)
6408 {
6409         struct seg6_bpf_srh_state *srh_state =
6410                 this_cpu_ptr(&seg6_bpf_srh_states);
6411         struct ipv6_sr_hdr *srh = srh_state->srh;
6412         void *srh_end, *srh_tlvs, *ptr;
6413         struct ipv6hdr *hdr;
6414         int srhoff = 0;
6415         int ret;
6416
6417         if (unlikely(srh == NULL))
6418                 return -EINVAL;
6419
6420         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6421                         ((srh->first_segment + 1) << 4));
6422         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6423                         srh_state->hdrlen);
6424         ptr = skb->data + offset;
6425
6426         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6427                 return -EFAULT;
6428         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6429                 return -EFAULT;
6430
6431         if (len > 0) {
6432                 ret = skb_cow_head(skb, len);
6433                 if (unlikely(ret < 0))
6434                         return ret;
6435
6436                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6437         } else {
6438                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6439         }
6440
6441         bpf_compute_data_pointers(skb);
6442         if (unlikely(ret < 0))
6443                 return ret;
6444
6445         hdr = (struct ipv6hdr *)skb->data;
6446         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6447
6448         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6449                 return -EINVAL;
6450         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6451         srh_state->hdrlen += len;
6452         srh_state->valid = false;
6453         return 0;
6454 }
6455
6456 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6457         .func           = bpf_lwt_seg6_adjust_srh,
6458         .gpl_only       = false,
6459         .ret_type       = RET_INTEGER,
6460         .arg1_type      = ARG_PTR_TO_CTX,
6461         .arg2_type      = ARG_ANYTHING,
6462         .arg3_type      = ARG_ANYTHING,
6463 };
6464 #endif /* CONFIG_IPV6_SEG6_BPF */
6465
6466 #ifdef CONFIG_INET
6467 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6468                               int dif, int sdif, u8 family, u8 proto)
6469 {
6470         struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo;
6471         bool refcounted = false;
6472         struct sock *sk = NULL;
6473
6474         if (family == AF_INET) {
6475                 __be32 src4 = tuple->ipv4.saddr;
6476                 __be32 dst4 = tuple->ipv4.daddr;
6477
6478                 if (proto == IPPROTO_TCP)
6479                         sk = __inet_lookup(net, hinfo, NULL, 0,
6480                                            src4, tuple->ipv4.sport,
6481                                            dst4, tuple->ipv4.dport,
6482                                            dif, sdif, &refcounted);
6483                 else
6484                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6485                                                dst4, tuple->ipv4.dport,
6486                                                dif, sdif, net->ipv4.udp_table, NULL);
6487 #if IS_ENABLED(CONFIG_IPV6)
6488         } else {
6489                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6490                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6491
6492                 if (proto == IPPROTO_TCP)
6493                         sk = __inet6_lookup(net, hinfo, NULL, 0,
6494                                             src6, tuple->ipv6.sport,
6495                                             dst6, ntohs(tuple->ipv6.dport),
6496                                             dif, sdif, &refcounted);
6497                 else if (likely(ipv6_bpf_stub))
6498                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6499                                                             src6, tuple->ipv6.sport,
6500                                                             dst6, tuple->ipv6.dport,
6501                                                             dif, sdif,
6502                                                             net->ipv4.udp_table, NULL);
6503 #endif
6504         }
6505
6506         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6507                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6508                 sk = NULL;
6509         }
6510         return sk;
6511 }
6512
6513 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6514  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6515  */
6516 static struct sock *
6517 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6518                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6519                  u64 flags)
6520 {
6521         struct sock *sk = NULL;
6522         struct net *net;
6523         u8 family;
6524         int sdif;
6525
6526         if (len == sizeof(tuple->ipv4))
6527                 family = AF_INET;
6528         else if (len == sizeof(tuple->ipv6))
6529                 family = AF_INET6;
6530         else
6531                 return NULL;
6532
6533         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6534                 goto out;
6535
6536         if (family == AF_INET)
6537                 sdif = inet_sdif(skb);
6538         else
6539                 sdif = inet6_sdif(skb);
6540
6541         if ((s32)netns_id < 0) {
6542                 net = caller_net;
6543                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6544         } else {
6545                 net = get_net_ns_by_id(caller_net, netns_id);
6546                 if (unlikely(!net))
6547                         goto out;
6548                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6549                 put_net(net);
6550         }
6551
6552 out:
6553         return sk;
6554 }
6555
6556 static struct sock *
6557 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6558                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6559                 u64 flags)
6560 {
6561         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6562                                            ifindex, proto, netns_id, flags);
6563
6564         if (sk) {
6565                 struct sock *sk2 = sk_to_full_sk(sk);
6566
6567                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6568                  * sock refcnt is decremented to prevent a request_sock leak.
6569                  */
6570                 if (!sk_fullsock(sk2))
6571                         sk2 = NULL;
6572                 if (sk2 != sk) {
6573                         sock_gen_put(sk);
6574                         /* Ensure there is no need to bump sk2 refcnt */
6575                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6576                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6577                                 return NULL;
6578                         }
6579                         sk = sk2;
6580                 }
6581         }
6582
6583         return sk;
6584 }
6585
6586 static struct sock *
6587 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6588                u8 proto, u64 netns_id, u64 flags)
6589 {
6590         struct net *caller_net;
6591         int ifindex;
6592
6593         if (skb->dev) {
6594                 caller_net = dev_net(skb->dev);
6595                 ifindex = skb->dev->ifindex;
6596         } else {
6597                 caller_net = sock_net(skb->sk);
6598                 ifindex = 0;
6599         }
6600
6601         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6602                                 netns_id, flags);
6603 }
6604
6605 static struct sock *
6606 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6607               u8 proto, u64 netns_id, u64 flags)
6608 {
6609         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6610                                          flags);
6611
6612         if (sk) {
6613                 struct sock *sk2 = sk_to_full_sk(sk);
6614
6615                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6616                  * sock refcnt is decremented to prevent a request_sock leak.
6617                  */
6618                 if (!sk_fullsock(sk2))
6619                         sk2 = NULL;
6620                 if (sk2 != sk) {
6621                         sock_gen_put(sk);
6622                         /* Ensure there is no need to bump sk2 refcnt */
6623                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6624                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6625                                 return NULL;
6626                         }
6627                         sk = sk2;
6628                 }
6629         }
6630
6631         return sk;
6632 }
6633
6634 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6635            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6636 {
6637         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6638                                              netns_id, flags);
6639 }
6640
6641 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6642         .func           = bpf_skc_lookup_tcp,
6643         .gpl_only       = false,
6644         .pkt_access     = true,
6645         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6646         .arg1_type      = ARG_PTR_TO_CTX,
6647         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6648         .arg3_type      = ARG_CONST_SIZE,
6649         .arg4_type      = ARG_ANYTHING,
6650         .arg5_type      = ARG_ANYTHING,
6651 };
6652
6653 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6654            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6655 {
6656         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6657                                             netns_id, flags);
6658 }
6659
6660 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6661         .func           = bpf_sk_lookup_tcp,
6662         .gpl_only       = false,
6663         .pkt_access     = true,
6664         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6665         .arg1_type      = ARG_PTR_TO_CTX,
6666         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6667         .arg3_type      = ARG_CONST_SIZE,
6668         .arg4_type      = ARG_ANYTHING,
6669         .arg5_type      = ARG_ANYTHING,
6670 };
6671
6672 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6673            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6674 {
6675         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6676                                             netns_id, flags);
6677 }
6678
6679 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6680         .func           = bpf_sk_lookup_udp,
6681         .gpl_only       = false,
6682         .pkt_access     = true,
6683         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6684         .arg1_type      = ARG_PTR_TO_CTX,
6685         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6686         .arg3_type      = ARG_CONST_SIZE,
6687         .arg4_type      = ARG_ANYTHING,
6688         .arg5_type      = ARG_ANYTHING,
6689 };
6690
6691 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6692 {
6693         if (sk && sk_is_refcounted(sk))
6694                 sock_gen_put(sk);
6695         return 0;
6696 }
6697
6698 static const struct bpf_func_proto bpf_sk_release_proto = {
6699         .func           = bpf_sk_release,
6700         .gpl_only       = false,
6701         .ret_type       = RET_INTEGER,
6702         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6703 };
6704
6705 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6706            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6707 {
6708         struct net *caller_net = dev_net(ctx->rxq->dev);
6709         int ifindex = ctx->rxq->dev->ifindex;
6710
6711         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6712                                               ifindex, IPPROTO_UDP, netns_id,
6713                                               flags);
6714 }
6715
6716 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6717         .func           = bpf_xdp_sk_lookup_udp,
6718         .gpl_only       = false,
6719         .pkt_access     = true,
6720         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6721         .arg1_type      = ARG_PTR_TO_CTX,
6722         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6723         .arg3_type      = ARG_CONST_SIZE,
6724         .arg4_type      = ARG_ANYTHING,
6725         .arg5_type      = ARG_ANYTHING,
6726 };
6727
6728 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6729            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6730 {
6731         struct net *caller_net = dev_net(ctx->rxq->dev);
6732         int ifindex = ctx->rxq->dev->ifindex;
6733
6734         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6735                                                ifindex, IPPROTO_TCP, netns_id,
6736                                                flags);
6737 }
6738
6739 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6740         .func           = bpf_xdp_skc_lookup_tcp,
6741         .gpl_only       = false,
6742         .pkt_access     = true,
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_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6752            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6753 {
6754         struct net *caller_net = dev_net(ctx->rxq->dev);
6755         int ifindex = ctx->rxq->dev->ifindex;
6756
6757         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6758                                               ifindex, IPPROTO_TCP, netns_id,
6759                                               flags);
6760 }
6761
6762 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6763         .func           = bpf_xdp_sk_lookup_tcp,
6764         .gpl_only       = false,
6765         .pkt_access     = true,
6766         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6767         .arg1_type      = ARG_PTR_TO_CTX,
6768         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6769         .arg3_type      = ARG_CONST_SIZE,
6770         .arg4_type      = ARG_ANYTHING,
6771         .arg5_type      = ARG_ANYTHING,
6772 };
6773
6774 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6775            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6776 {
6777         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6778                                                sock_net(ctx->sk), 0,
6779                                                IPPROTO_TCP, netns_id, flags);
6780 }
6781
6782 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6783         .func           = bpf_sock_addr_skc_lookup_tcp,
6784         .gpl_only       = false,
6785         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6786         .arg1_type      = ARG_PTR_TO_CTX,
6787         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6788         .arg3_type      = ARG_CONST_SIZE,
6789         .arg4_type      = ARG_ANYTHING,
6790         .arg5_type      = ARG_ANYTHING,
6791 };
6792
6793 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6794            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6795 {
6796         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6797                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6798                                               netns_id, flags);
6799 }
6800
6801 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6802         .func           = bpf_sock_addr_sk_lookup_tcp,
6803         .gpl_only       = false,
6804         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6805         .arg1_type      = ARG_PTR_TO_CTX,
6806         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6807         .arg3_type      = ARG_CONST_SIZE,
6808         .arg4_type      = ARG_ANYTHING,
6809         .arg5_type      = ARG_ANYTHING,
6810 };
6811
6812 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6813            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6814 {
6815         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6816                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6817                                               netns_id, flags);
6818 }
6819
6820 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6821         .func           = bpf_sock_addr_sk_lookup_udp,
6822         .gpl_only       = false,
6823         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6824         .arg1_type      = ARG_PTR_TO_CTX,
6825         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6826         .arg3_type      = ARG_CONST_SIZE,
6827         .arg4_type      = ARG_ANYTHING,
6828         .arg5_type      = ARG_ANYTHING,
6829 };
6830
6831 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6832                                   struct bpf_insn_access_aux *info)
6833 {
6834         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6835                                           icsk_retransmits))
6836                 return false;
6837
6838         if (off % size != 0)
6839                 return false;
6840
6841         switch (off) {
6842         case offsetof(struct bpf_tcp_sock, bytes_received):
6843         case offsetof(struct bpf_tcp_sock, bytes_acked):
6844                 return size == sizeof(__u64);
6845         default:
6846                 return size == sizeof(__u32);
6847         }
6848 }
6849
6850 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6851                                     const struct bpf_insn *si,
6852                                     struct bpf_insn *insn_buf,
6853                                     struct bpf_prog *prog, u32 *target_size)
6854 {
6855         struct bpf_insn *insn = insn_buf;
6856
6857 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6858         do {                                                            \
6859                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6860                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6861                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6862                                       si->dst_reg, si->src_reg,         \
6863                                       offsetof(struct tcp_sock, FIELD)); \
6864         } while (0)
6865
6866 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6867         do {                                                            \
6868                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6869                                           FIELD) >                      \
6870                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6871                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6872                                         struct inet_connection_sock,    \
6873                                         FIELD),                         \
6874                                       si->dst_reg, si->src_reg,         \
6875                                       offsetof(                         \
6876                                         struct inet_connection_sock,    \
6877                                         FIELD));                        \
6878         } while (0)
6879
6880         switch (si->off) {
6881         case offsetof(struct bpf_tcp_sock, rtt_min):
6882                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6883                              sizeof(struct minmax));
6884                 BUILD_BUG_ON(sizeof(struct minmax) <
6885                              sizeof(struct minmax_sample));
6886
6887                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6888                                       offsetof(struct tcp_sock, rtt_min) +
6889                                       offsetof(struct minmax_sample, v));
6890                 break;
6891         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6892                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6893                 break;
6894         case offsetof(struct bpf_tcp_sock, srtt_us):
6895                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6896                 break;
6897         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6898                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6899                 break;
6900         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6901                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6902                 break;
6903         case offsetof(struct bpf_tcp_sock, snd_nxt):
6904                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6905                 break;
6906         case offsetof(struct bpf_tcp_sock, snd_una):
6907                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6908                 break;
6909         case offsetof(struct bpf_tcp_sock, mss_cache):
6910                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6911                 break;
6912         case offsetof(struct bpf_tcp_sock, ecn_flags):
6913                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6914                 break;
6915         case offsetof(struct bpf_tcp_sock, rate_delivered):
6916                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6917                 break;
6918         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6919                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6920                 break;
6921         case offsetof(struct bpf_tcp_sock, packets_out):
6922                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6923                 break;
6924         case offsetof(struct bpf_tcp_sock, retrans_out):
6925                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6926                 break;
6927         case offsetof(struct bpf_tcp_sock, total_retrans):
6928                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6929                 break;
6930         case offsetof(struct bpf_tcp_sock, segs_in):
6931                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6932                 break;
6933         case offsetof(struct bpf_tcp_sock, data_segs_in):
6934                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6935                 break;
6936         case offsetof(struct bpf_tcp_sock, segs_out):
6937                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6938                 break;
6939         case offsetof(struct bpf_tcp_sock, data_segs_out):
6940                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6941                 break;
6942         case offsetof(struct bpf_tcp_sock, lost_out):
6943                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6944                 break;
6945         case offsetof(struct bpf_tcp_sock, sacked_out):
6946                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6947                 break;
6948         case offsetof(struct bpf_tcp_sock, bytes_received):
6949                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6950                 break;
6951         case offsetof(struct bpf_tcp_sock, bytes_acked):
6952                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6953                 break;
6954         case offsetof(struct bpf_tcp_sock, dsack_dups):
6955                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6956                 break;
6957         case offsetof(struct bpf_tcp_sock, delivered):
6958                 BPF_TCP_SOCK_GET_COMMON(delivered);
6959                 break;
6960         case offsetof(struct bpf_tcp_sock, delivered_ce):
6961                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6962                 break;
6963         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6964                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6965                 break;
6966         }
6967
6968         return insn - insn_buf;
6969 }
6970
6971 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6972 {
6973         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6974                 return (unsigned long)sk;
6975
6976         return (unsigned long)NULL;
6977 }
6978
6979 const struct bpf_func_proto bpf_tcp_sock_proto = {
6980         .func           = bpf_tcp_sock,
6981         .gpl_only       = false,
6982         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6983         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6984 };
6985
6986 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6987 {
6988         sk = sk_to_full_sk(sk);
6989
6990         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6991                 return (unsigned long)sk;
6992
6993         return (unsigned long)NULL;
6994 }
6995
6996 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6997         .func           = bpf_get_listener_sock,
6998         .gpl_only       = false,
6999         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
7000         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
7001 };
7002
7003 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
7004 {
7005         unsigned int iphdr_len;
7006
7007         switch (skb_protocol(skb, true)) {
7008         case cpu_to_be16(ETH_P_IP):
7009                 iphdr_len = sizeof(struct iphdr);
7010                 break;
7011         case cpu_to_be16(ETH_P_IPV6):
7012                 iphdr_len = sizeof(struct ipv6hdr);
7013                 break;
7014         default:
7015                 return 0;
7016         }
7017
7018         if (skb_headlen(skb) < iphdr_len)
7019                 return 0;
7020
7021         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
7022                 return 0;
7023
7024         return INET_ECN_set_ce(skb);
7025 }
7026
7027 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7028                                   struct bpf_insn_access_aux *info)
7029 {
7030         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
7031                 return false;
7032
7033         if (off % size != 0)
7034                 return false;
7035
7036         switch (off) {
7037         default:
7038                 return size == sizeof(__u32);
7039         }
7040 }
7041
7042 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
7043                                     const struct bpf_insn *si,
7044                                     struct bpf_insn *insn_buf,
7045                                     struct bpf_prog *prog, u32 *target_size)
7046 {
7047         struct bpf_insn *insn = insn_buf;
7048
7049 #define BPF_XDP_SOCK_GET(FIELD)                                         \
7050         do {                                                            \
7051                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
7052                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
7053                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
7054                                       si->dst_reg, si->src_reg,         \
7055                                       offsetof(struct xdp_sock, FIELD)); \
7056         } while (0)
7057
7058         switch (si->off) {
7059         case offsetof(struct bpf_xdp_sock, queue_id):
7060                 BPF_XDP_SOCK_GET(queue_id);
7061                 break;
7062         }
7063
7064         return insn - insn_buf;
7065 }
7066
7067 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
7068         .func           = bpf_skb_ecn_set_ce,
7069         .gpl_only       = false,
7070         .ret_type       = RET_INTEGER,
7071         .arg1_type      = ARG_PTR_TO_CTX,
7072 };
7073
7074 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7075            struct tcphdr *, th, u32, th_len)
7076 {
7077 #ifdef CONFIG_SYN_COOKIES
7078         u32 cookie;
7079         int ret;
7080
7081         if (unlikely(!sk || th_len < sizeof(*th)))
7082                 return -EINVAL;
7083
7084         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
7085         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7086                 return -EINVAL;
7087
7088         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7089                 return -EINVAL;
7090
7091         if (!th->ack || th->rst || th->syn)
7092                 return -ENOENT;
7093
7094         if (unlikely(iph_len < sizeof(struct iphdr)))
7095                 return -EINVAL;
7096
7097         if (tcp_synq_no_recent_overflow(sk))
7098                 return -ENOENT;
7099
7100         cookie = ntohl(th->ack_seq) - 1;
7101
7102         /* Both struct iphdr and struct ipv6hdr have the version field at the
7103          * same offset so we can cast to the shorter header (struct iphdr).
7104          */
7105         switch (((struct iphdr *)iph)->version) {
7106         case 4:
7107                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7108                         return -EINVAL;
7109
7110                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7111                 break;
7112
7113 #if IS_BUILTIN(CONFIG_IPV6)
7114         case 6:
7115                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7116                         return -EINVAL;
7117
7118                 if (sk->sk_family != AF_INET6)
7119                         return -EINVAL;
7120
7121                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7122                 break;
7123 #endif /* CONFIG_IPV6 */
7124
7125         default:
7126                 return -EPROTONOSUPPORT;
7127         }
7128
7129         if (ret > 0)
7130                 return 0;
7131
7132         return -ENOENT;
7133 #else
7134         return -ENOTSUPP;
7135 #endif
7136 }
7137
7138 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7139         .func           = bpf_tcp_check_syncookie,
7140         .gpl_only       = true,
7141         .pkt_access     = true,
7142         .ret_type       = RET_INTEGER,
7143         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7144         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7145         .arg3_type      = ARG_CONST_SIZE,
7146         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7147         .arg5_type      = ARG_CONST_SIZE,
7148 };
7149
7150 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7151            struct tcphdr *, th, u32, th_len)
7152 {
7153 #ifdef CONFIG_SYN_COOKIES
7154         u32 cookie;
7155         u16 mss;
7156
7157         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7158                 return -EINVAL;
7159
7160         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7161                 return -EINVAL;
7162
7163         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7164                 return -ENOENT;
7165
7166         if (!th->syn || th->ack || th->fin || th->rst)
7167                 return -EINVAL;
7168
7169         if (unlikely(iph_len < sizeof(struct iphdr)))
7170                 return -EINVAL;
7171
7172         /* Both struct iphdr and struct ipv6hdr have the version field at the
7173          * same offset so we can cast to the shorter header (struct iphdr).
7174          */
7175         switch (((struct iphdr *)iph)->version) {
7176         case 4:
7177                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7178                         return -EINVAL;
7179
7180                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7181                 break;
7182
7183 #if IS_BUILTIN(CONFIG_IPV6)
7184         case 6:
7185                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7186                         return -EINVAL;
7187
7188                 if (sk->sk_family != AF_INET6)
7189                         return -EINVAL;
7190
7191                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7192                 break;
7193 #endif /* CONFIG_IPV6 */
7194
7195         default:
7196                 return -EPROTONOSUPPORT;
7197         }
7198         if (mss == 0)
7199                 return -ENOENT;
7200
7201         return cookie | ((u64)mss << 32);
7202 #else
7203         return -EOPNOTSUPP;
7204 #endif /* CONFIG_SYN_COOKIES */
7205 }
7206
7207 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7208         .func           = bpf_tcp_gen_syncookie,
7209         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7210         .pkt_access     = true,
7211         .ret_type       = RET_INTEGER,
7212         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7213         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7214         .arg3_type      = ARG_CONST_SIZE,
7215         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7216         .arg5_type      = ARG_CONST_SIZE,
7217 };
7218
7219 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7220 {
7221         if (!sk || flags != 0)
7222                 return -EINVAL;
7223         if (!skb_at_tc_ingress(skb))
7224                 return -EOPNOTSUPP;
7225         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7226                 return -ENETUNREACH;
7227         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7228                 return -ESOCKTNOSUPPORT;
7229         if (sk_is_refcounted(sk) &&
7230             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7231                 return -ENOENT;
7232
7233         skb_orphan(skb);
7234         skb->sk = sk;
7235         skb->destructor = sock_pfree;
7236
7237         return 0;
7238 }
7239
7240 static const struct bpf_func_proto bpf_sk_assign_proto = {
7241         .func           = bpf_sk_assign,
7242         .gpl_only       = false,
7243         .ret_type       = RET_INTEGER,
7244         .arg1_type      = ARG_PTR_TO_CTX,
7245         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7246         .arg3_type      = ARG_ANYTHING,
7247 };
7248
7249 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7250                                     u8 search_kind, const u8 *magic,
7251                                     u8 magic_len, bool *eol)
7252 {
7253         u8 kind, kind_len;
7254
7255         *eol = false;
7256
7257         while (op < opend) {
7258                 kind = op[0];
7259
7260                 if (kind == TCPOPT_EOL) {
7261                         *eol = true;
7262                         return ERR_PTR(-ENOMSG);
7263                 } else if (kind == TCPOPT_NOP) {
7264                         op++;
7265                         continue;
7266                 }
7267
7268                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7269                         /* Something is wrong in the received header.
7270                          * Follow the TCP stack's tcp_parse_options()
7271                          * and just bail here.
7272                          */
7273                         return ERR_PTR(-EFAULT);
7274
7275                 kind_len = op[1];
7276                 if (search_kind == kind) {
7277                         if (!magic_len)
7278                                 return op;
7279
7280                         if (magic_len > kind_len - 2)
7281                                 return ERR_PTR(-ENOMSG);
7282
7283                         if (!memcmp(&op[2], magic, magic_len))
7284                                 return op;
7285                 }
7286
7287                 op += kind_len;
7288         }
7289
7290         return ERR_PTR(-ENOMSG);
7291 }
7292
7293 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7294            void *, search_res, u32, len, u64, flags)
7295 {
7296         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7297         const u8 *op, *opend, *magic, *search = search_res;
7298         u8 search_kind, search_len, copy_len, magic_len;
7299         int ret;
7300
7301         /* 2 byte is the minimal option len except TCPOPT_NOP and
7302          * TCPOPT_EOL which are useless for the bpf prog to learn
7303          * and this helper disallow loading them also.
7304          */
7305         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7306                 return -EINVAL;
7307
7308         search_kind = search[0];
7309         search_len = search[1];
7310
7311         if (search_len > len || search_kind == TCPOPT_NOP ||
7312             search_kind == TCPOPT_EOL)
7313                 return -EINVAL;
7314
7315         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7316                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7317                 if (search_len != 4 && search_len != 6)
7318                         return -EINVAL;
7319                 magic = &search[2];
7320                 magic_len = search_len - 2;
7321         } else {
7322                 if (search_len)
7323                         return -EINVAL;
7324                 magic = NULL;
7325                 magic_len = 0;
7326         }
7327
7328         if (load_syn) {
7329                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7330                 if (ret < 0)
7331                         return ret;
7332
7333                 opend = op + ret;
7334                 op += sizeof(struct tcphdr);
7335         } else {
7336                 if (!bpf_sock->skb ||
7337                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7338                         /* This bpf_sock->op cannot call this helper */
7339                         return -EPERM;
7340
7341                 opend = bpf_sock->skb_data_end;
7342                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7343         }
7344
7345         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7346                                 &eol);
7347         if (IS_ERR(op))
7348                 return PTR_ERR(op);
7349
7350         copy_len = op[1];
7351         ret = copy_len;
7352         if (copy_len > len) {
7353                 ret = -ENOSPC;
7354                 copy_len = len;
7355         }
7356
7357         memcpy(search_res, op, copy_len);
7358         return ret;
7359 }
7360
7361 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7362         .func           = bpf_sock_ops_load_hdr_opt,
7363         .gpl_only       = false,
7364         .ret_type       = RET_INTEGER,
7365         .arg1_type      = ARG_PTR_TO_CTX,
7366         .arg2_type      = ARG_PTR_TO_MEM,
7367         .arg3_type      = ARG_CONST_SIZE,
7368         .arg4_type      = ARG_ANYTHING,
7369 };
7370
7371 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7372            const void *, from, u32, len, u64, flags)
7373 {
7374         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7375         const u8 *op, *new_op, *magic = NULL;
7376         struct sk_buff *skb;
7377         bool eol;
7378
7379         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7380                 return -EPERM;
7381
7382         if (len < 2 || flags)
7383                 return -EINVAL;
7384
7385         new_op = from;
7386         new_kind = new_op[0];
7387         new_kind_len = new_op[1];
7388
7389         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7390             new_kind == TCPOPT_EOL)
7391                 return -EINVAL;
7392
7393         if (new_kind_len > bpf_sock->remaining_opt_len)
7394                 return -ENOSPC;
7395
7396         /* 253 is another experimental kind */
7397         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7398                 if (new_kind_len < 4)
7399                         return -EINVAL;
7400                 /* Match for the 2 byte magic also.
7401                  * RFC 6994: the magic could be 2 or 4 bytes.
7402                  * Hence, matching by 2 byte only is on the
7403                  * conservative side but it is the right
7404                  * thing to do for the 'search-for-duplication'
7405                  * purpose.
7406                  */
7407                 magic = &new_op[2];
7408                 magic_len = 2;
7409         }
7410
7411         /* Check for duplication */
7412         skb = bpf_sock->skb;
7413         op = skb->data + sizeof(struct tcphdr);
7414         opend = bpf_sock->skb_data_end;
7415
7416         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7417                                 &eol);
7418         if (!IS_ERR(op))
7419                 return -EEXIST;
7420
7421         if (PTR_ERR(op) != -ENOMSG)
7422                 return PTR_ERR(op);
7423
7424         if (eol)
7425                 /* The option has been ended.  Treat it as no more
7426                  * header option can be written.
7427                  */
7428                 return -ENOSPC;
7429
7430         /* No duplication found.  Store the header option. */
7431         memcpy(opend, from, new_kind_len);
7432
7433         bpf_sock->remaining_opt_len -= new_kind_len;
7434         bpf_sock->skb_data_end += new_kind_len;
7435
7436         return 0;
7437 }
7438
7439 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7440         .func           = bpf_sock_ops_store_hdr_opt,
7441         .gpl_only       = false,
7442         .ret_type       = RET_INTEGER,
7443         .arg1_type      = ARG_PTR_TO_CTX,
7444         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7445         .arg3_type      = ARG_CONST_SIZE,
7446         .arg4_type      = ARG_ANYTHING,
7447 };
7448
7449 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7450            u32, len, u64, flags)
7451 {
7452         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7453                 return -EPERM;
7454
7455         if (flags || len < 2)
7456                 return -EINVAL;
7457
7458         if (len > bpf_sock->remaining_opt_len)
7459                 return -ENOSPC;
7460
7461         bpf_sock->remaining_opt_len -= len;
7462
7463         return 0;
7464 }
7465
7466 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7467         .func           = bpf_sock_ops_reserve_hdr_opt,
7468         .gpl_only       = false,
7469         .ret_type       = RET_INTEGER,
7470         .arg1_type      = ARG_PTR_TO_CTX,
7471         .arg2_type      = ARG_ANYTHING,
7472         .arg3_type      = ARG_ANYTHING,
7473 };
7474
7475 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7476            u64, tstamp, u32, tstamp_type)
7477 {
7478         /* skb_clear_delivery_time() is done for inet protocol */
7479         if (skb->protocol != htons(ETH_P_IP) &&
7480             skb->protocol != htons(ETH_P_IPV6))
7481                 return -EOPNOTSUPP;
7482
7483         switch (tstamp_type) {
7484         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7485                 if (!tstamp)
7486                         return -EINVAL;
7487                 skb->tstamp = tstamp;
7488                 skb->mono_delivery_time = 1;
7489                 break;
7490         case BPF_SKB_TSTAMP_UNSPEC:
7491                 if (tstamp)
7492                         return -EINVAL;
7493                 skb->tstamp = 0;
7494                 skb->mono_delivery_time = 0;
7495                 break;
7496         default:
7497                 return -EINVAL;
7498         }
7499
7500         return 0;
7501 }
7502
7503 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7504         .func           = bpf_skb_set_tstamp,
7505         .gpl_only       = false,
7506         .ret_type       = RET_INTEGER,
7507         .arg1_type      = ARG_PTR_TO_CTX,
7508         .arg2_type      = ARG_ANYTHING,
7509         .arg3_type      = ARG_ANYTHING,
7510 };
7511
7512 #ifdef CONFIG_SYN_COOKIES
7513 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7514            struct tcphdr *, th, u32, th_len)
7515 {
7516         u32 cookie;
7517         u16 mss;
7518
7519         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7520                 return -EINVAL;
7521
7522         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7523         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7524
7525         return cookie | ((u64)mss << 32);
7526 }
7527
7528 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7529         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7530         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7531         .pkt_access     = true,
7532         .ret_type       = RET_INTEGER,
7533         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7534         .arg1_size      = sizeof(struct iphdr),
7535         .arg2_type      = ARG_PTR_TO_MEM,
7536         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7537 };
7538
7539 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7540            struct tcphdr *, th, u32, th_len)
7541 {
7542 #if IS_BUILTIN(CONFIG_IPV6)
7543         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7544                 sizeof(struct ipv6hdr);
7545         u32 cookie;
7546         u16 mss;
7547
7548         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7549                 return -EINVAL;
7550
7551         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7552         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7553
7554         return cookie | ((u64)mss << 32);
7555 #else
7556         return -EPROTONOSUPPORT;
7557 #endif
7558 }
7559
7560 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7561         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7562         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7563         .pkt_access     = true,
7564         .ret_type       = RET_INTEGER,
7565         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7566         .arg1_size      = sizeof(struct ipv6hdr),
7567         .arg2_type      = ARG_PTR_TO_MEM,
7568         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
7569 };
7570
7571 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7572            struct tcphdr *, th)
7573 {
7574         u32 cookie = ntohl(th->ack_seq) - 1;
7575
7576         if (__cookie_v4_check(iph, th, cookie) > 0)
7577                 return 0;
7578
7579         return -EACCES;
7580 }
7581
7582 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7583         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7584         .gpl_only       = true, /* __cookie_v4_check is GPL */
7585         .pkt_access     = true,
7586         .ret_type       = RET_INTEGER,
7587         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7588         .arg1_size      = sizeof(struct iphdr),
7589         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7590         .arg2_size      = sizeof(struct tcphdr),
7591 };
7592
7593 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7594            struct tcphdr *, th)
7595 {
7596 #if IS_BUILTIN(CONFIG_IPV6)
7597         u32 cookie = ntohl(th->ack_seq) - 1;
7598
7599         if (__cookie_v6_check(iph, th, cookie) > 0)
7600                 return 0;
7601
7602         return -EACCES;
7603 #else
7604         return -EPROTONOSUPPORT;
7605 #endif
7606 }
7607
7608 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7609         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7610         .gpl_only       = true, /* __cookie_v6_check is GPL */
7611         .pkt_access     = true,
7612         .ret_type       = RET_INTEGER,
7613         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7614         .arg1_size      = sizeof(struct ipv6hdr),
7615         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7616         .arg2_size      = sizeof(struct tcphdr),
7617 };
7618 #endif /* CONFIG_SYN_COOKIES */
7619
7620 #endif /* CONFIG_INET */
7621
7622 bool bpf_helper_changes_pkt_data(void *func)
7623 {
7624         if (func == bpf_skb_vlan_push ||
7625             func == bpf_skb_vlan_pop ||
7626             func == bpf_skb_store_bytes ||
7627             func == bpf_skb_change_proto ||
7628             func == bpf_skb_change_head ||
7629             func == sk_skb_change_head ||
7630             func == bpf_skb_change_tail ||
7631             func == sk_skb_change_tail ||
7632             func == bpf_skb_adjust_room ||
7633             func == sk_skb_adjust_room ||
7634             func == bpf_skb_pull_data ||
7635             func == sk_skb_pull_data ||
7636             func == bpf_clone_redirect ||
7637             func == bpf_l3_csum_replace ||
7638             func == bpf_l4_csum_replace ||
7639             func == bpf_xdp_adjust_head ||
7640             func == bpf_xdp_adjust_meta ||
7641             func == bpf_msg_pull_data ||
7642             func == bpf_msg_push_data ||
7643             func == bpf_msg_pop_data ||
7644             func == bpf_xdp_adjust_tail ||
7645 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7646             func == bpf_lwt_seg6_store_bytes ||
7647             func == bpf_lwt_seg6_adjust_srh ||
7648             func == bpf_lwt_seg6_action ||
7649 #endif
7650 #ifdef CONFIG_INET
7651             func == bpf_sock_ops_store_hdr_opt ||
7652 #endif
7653             func == bpf_lwt_in_push_encap ||
7654             func == bpf_lwt_xmit_push_encap)
7655                 return true;
7656
7657         return false;
7658 }
7659
7660 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7661 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7662
7663 static const struct bpf_func_proto *
7664 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7665 {
7666         const struct bpf_func_proto *func_proto;
7667
7668         func_proto = cgroup_common_func_proto(func_id, prog);
7669         if (func_proto)
7670                 return func_proto;
7671
7672         func_proto = cgroup_current_func_proto(func_id, prog);
7673         if (func_proto)
7674                 return func_proto;
7675
7676         switch (func_id) {
7677         case BPF_FUNC_get_socket_cookie:
7678                 return &bpf_get_socket_cookie_sock_proto;
7679         case BPF_FUNC_get_netns_cookie:
7680                 return &bpf_get_netns_cookie_sock_proto;
7681         case BPF_FUNC_perf_event_output:
7682                 return &bpf_event_output_data_proto;
7683         case BPF_FUNC_sk_storage_get:
7684                 return &bpf_sk_storage_get_cg_sock_proto;
7685         case BPF_FUNC_ktime_get_coarse_ns:
7686                 return &bpf_ktime_get_coarse_ns_proto;
7687         default:
7688                 return bpf_base_func_proto(func_id);
7689         }
7690 }
7691
7692 static const struct bpf_func_proto *
7693 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7694 {
7695         const struct bpf_func_proto *func_proto;
7696
7697         func_proto = cgroup_common_func_proto(func_id, prog);
7698         if (func_proto)
7699                 return func_proto;
7700
7701         func_proto = cgroup_current_func_proto(func_id, prog);
7702         if (func_proto)
7703                 return func_proto;
7704
7705         switch (func_id) {
7706         case BPF_FUNC_bind:
7707                 switch (prog->expected_attach_type) {
7708                 case BPF_CGROUP_INET4_CONNECT:
7709                 case BPF_CGROUP_INET6_CONNECT:
7710                         return &bpf_bind_proto;
7711                 default:
7712                         return NULL;
7713                 }
7714         case BPF_FUNC_get_socket_cookie:
7715                 return &bpf_get_socket_cookie_sock_addr_proto;
7716         case BPF_FUNC_get_netns_cookie:
7717                 return &bpf_get_netns_cookie_sock_addr_proto;
7718         case BPF_FUNC_perf_event_output:
7719                 return &bpf_event_output_data_proto;
7720 #ifdef CONFIG_INET
7721         case BPF_FUNC_sk_lookup_tcp:
7722                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7723         case BPF_FUNC_sk_lookup_udp:
7724                 return &bpf_sock_addr_sk_lookup_udp_proto;
7725         case BPF_FUNC_sk_release:
7726                 return &bpf_sk_release_proto;
7727         case BPF_FUNC_skc_lookup_tcp:
7728                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7729 #endif /* CONFIG_INET */
7730         case BPF_FUNC_sk_storage_get:
7731                 return &bpf_sk_storage_get_proto;
7732         case BPF_FUNC_sk_storage_delete:
7733                 return &bpf_sk_storage_delete_proto;
7734         case BPF_FUNC_setsockopt:
7735                 switch (prog->expected_attach_type) {
7736                 case BPF_CGROUP_INET4_BIND:
7737                 case BPF_CGROUP_INET6_BIND:
7738                 case BPF_CGROUP_INET4_CONNECT:
7739                 case BPF_CGROUP_INET6_CONNECT:
7740                 case BPF_CGROUP_UDP4_RECVMSG:
7741                 case BPF_CGROUP_UDP6_RECVMSG:
7742                 case BPF_CGROUP_UDP4_SENDMSG:
7743                 case BPF_CGROUP_UDP6_SENDMSG:
7744                 case BPF_CGROUP_INET4_GETPEERNAME:
7745                 case BPF_CGROUP_INET6_GETPEERNAME:
7746                 case BPF_CGROUP_INET4_GETSOCKNAME:
7747                 case BPF_CGROUP_INET6_GETSOCKNAME:
7748                         return &bpf_sock_addr_setsockopt_proto;
7749                 default:
7750                         return NULL;
7751                 }
7752         case BPF_FUNC_getsockopt:
7753                 switch (prog->expected_attach_type) {
7754                 case BPF_CGROUP_INET4_BIND:
7755                 case BPF_CGROUP_INET6_BIND:
7756                 case BPF_CGROUP_INET4_CONNECT:
7757                 case BPF_CGROUP_INET6_CONNECT:
7758                 case BPF_CGROUP_UDP4_RECVMSG:
7759                 case BPF_CGROUP_UDP6_RECVMSG:
7760                 case BPF_CGROUP_UDP4_SENDMSG:
7761                 case BPF_CGROUP_UDP6_SENDMSG:
7762                 case BPF_CGROUP_INET4_GETPEERNAME:
7763                 case BPF_CGROUP_INET6_GETPEERNAME:
7764                 case BPF_CGROUP_INET4_GETSOCKNAME:
7765                 case BPF_CGROUP_INET6_GETSOCKNAME:
7766                         return &bpf_sock_addr_getsockopt_proto;
7767                 default:
7768                         return NULL;
7769                 }
7770         default:
7771                 return bpf_sk_base_func_proto(func_id);
7772         }
7773 }
7774
7775 static const struct bpf_func_proto *
7776 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7777 {
7778         switch (func_id) {
7779         case BPF_FUNC_skb_load_bytes:
7780                 return &bpf_skb_load_bytes_proto;
7781         case BPF_FUNC_skb_load_bytes_relative:
7782                 return &bpf_skb_load_bytes_relative_proto;
7783         case BPF_FUNC_get_socket_cookie:
7784                 return &bpf_get_socket_cookie_proto;
7785         case BPF_FUNC_get_socket_uid:
7786                 return &bpf_get_socket_uid_proto;
7787         case BPF_FUNC_perf_event_output:
7788                 return &bpf_skb_event_output_proto;
7789         default:
7790                 return bpf_sk_base_func_proto(func_id);
7791         }
7792 }
7793
7794 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7795 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7796
7797 static const struct bpf_func_proto *
7798 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7799 {
7800         const struct bpf_func_proto *func_proto;
7801
7802         func_proto = cgroup_common_func_proto(func_id, prog);
7803         if (func_proto)
7804                 return func_proto;
7805
7806         switch (func_id) {
7807         case BPF_FUNC_sk_fullsock:
7808                 return &bpf_sk_fullsock_proto;
7809         case BPF_FUNC_sk_storage_get:
7810                 return &bpf_sk_storage_get_proto;
7811         case BPF_FUNC_sk_storage_delete:
7812                 return &bpf_sk_storage_delete_proto;
7813         case BPF_FUNC_perf_event_output:
7814                 return &bpf_skb_event_output_proto;
7815 #ifdef CONFIG_SOCK_CGROUP_DATA
7816         case BPF_FUNC_skb_cgroup_id:
7817                 return &bpf_skb_cgroup_id_proto;
7818         case BPF_FUNC_skb_ancestor_cgroup_id:
7819                 return &bpf_skb_ancestor_cgroup_id_proto;
7820         case BPF_FUNC_sk_cgroup_id:
7821                 return &bpf_sk_cgroup_id_proto;
7822         case BPF_FUNC_sk_ancestor_cgroup_id:
7823                 return &bpf_sk_ancestor_cgroup_id_proto;
7824 #endif
7825 #ifdef CONFIG_INET
7826         case BPF_FUNC_sk_lookup_tcp:
7827                 return &bpf_sk_lookup_tcp_proto;
7828         case BPF_FUNC_sk_lookup_udp:
7829                 return &bpf_sk_lookup_udp_proto;
7830         case BPF_FUNC_sk_release:
7831                 return &bpf_sk_release_proto;
7832         case BPF_FUNC_skc_lookup_tcp:
7833                 return &bpf_skc_lookup_tcp_proto;
7834         case BPF_FUNC_tcp_sock:
7835                 return &bpf_tcp_sock_proto;
7836         case BPF_FUNC_get_listener_sock:
7837                 return &bpf_get_listener_sock_proto;
7838         case BPF_FUNC_skb_ecn_set_ce:
7839                 return &bpf_skb_ecn_set_ce_proto;
7840 #endif
7841         default:
7842                 return sk_filter_func_proto(func_id, prog);
7843         }
7844 }
7845
7846 static const struct bpf_func_proto *
7847 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7848 {
7849         switch (func_id) {
7850         case BPF_FUNC_skb_store_bytes:
7851                 return &bpf_skb_store_bytes_proto;
7852         case BPF_FUNC_skb_load_bytes:
7853                 return &bpf_skb_load_bytes_proto;
7854         case BPF_FUNC_skb_load_bytes_relative:
7855                 return &bpf_skb_load_bytes_relative_proto;
7856         case BPF_FUNC_skb_pull_data:
7857                 return &bpf_skb_pull_data_proto;
7858         case BPF_FUNC_csum_diff:
7859                 return &bpf_csum_diff_proto;
7860         case BPF_FUNC_csum_update:
7861                 return &bpf_csum_update_proto;
7862         case BPF_FUNC_csum_level:
7863                 return &bpf_csum_level_proto;
7864         case BPF_FUNC_l3_csum_replace:
7865                 return &bpf_l3_csum_replace_proto;
7866         case BPF_FUNC_l4_csum_replace:
7867                 return &bpf_l4_csum_replace_proto;
7868         case BPF_FUNC_clone_redirect:
7869                 return &bpf_clone_redirect_proto;
7870         case BPF_FUNC_get_cgroup_classid:
7871                 return &bpf_get_cgroup_classid_proto;
7872         case BPF_FUNC_skb_vlan_push:
7873                 return &bpf_skb_vlan_push_proto;
7874         case BPF_FUNC_skb_vlan_pop:
7875                 return &bpf_skb_vlan_pop_proto;
7876         case BPF_FUNC_skb_change_proto:
7877                 return &bpf_skb_change_proto_proto;
7878         case BPF_FUNC_skb_change_type:
7879                 return &bpf_skb_change_type_proto;
7880         case BPF_FUNC_skb_adjust_room:
7881                 return &bpf_skb_adjust_room_proto;
7882         case BPF_FUNC_skb_change_tail:
7883                 return &bpf_skb_change_tail_proto;
7884         case BPF_FUNC_skb_change_head:
7885                 return &bpf_skb_change_head_proto;
7886         case BPF_FUNC_skb_get_tunnel_key:
7887                 return &bpf_skb_get_tunnel_key_proto;
7888         case BPF_FUNC_skb_set_tunnel_key:
7889                 return bpf_get_skb_set_tunnel_proto(func_id);
7890         case BPF_FUNC_skb_get_tunnel_opt:
7891                 return &bpf_skb_get_tunnel_opt_proto;
7892         case BPF_FUNC_skb_set_tunnel_opt:
7893                 return bpf_get_skb_set_tunnel_proto(func_id);
7894         case BPF_FUNC_redirect:
7895                 return &bpf_redirect_proto;
7896         case BPF_FUNC_redirect_neigh:
7897                 return &bpf_redirect_neigh_proto;
7898         case BPF_FUNC_redirect_peer:
7899                 return &bpf_redirect_peer_proto;
7900         case BPF_FUNC_get_route_realm:
7901                 return &bpf_get_route_realm_proto;
7902         case BPF_FUNC_get_hash_recalc:
7903                 return &bpf_get_hash_recalc_proto;
7904         case BPF_FUNC_set_hash_invalid:
7905                 return &bpf_set_hash_invalid_proto;
7906         case BPF_FUNC_set_hash:
7907                 return &bpf_set_hash_proto;
7908         case BPF_FUNC_perf_event_output:
7909                 return &bpf_skb_event_output_proto;
7910         case BPF_FUNC_get_smp_processor_id:
7911                 return &bpf_get_smp_processor_id_proto;
7912         case BPF_FUNC_skb_under_cgroup:
7913                 return &bpf_skb_under_cgroup_proto;
7914         case BPF_FUNC_get_socket_cookie:
7915                 return &bpf_get_socket_cookie_proto;
7916         case BPF_FUNC_get_socket_uid:
7917                 return &bpf_get_socket_uid_proto;
7918         case BPF_FUNC_fib_lookup:
7919                 return &bpf_skb_fib_lookup_proto;
7920         case BPF_FUNC_check_mtu:
7921                 return &bpf_skb_check_mtu_proto;
7922         case BPF_FUNC_sk_fullsock:
7923                 return &bpf_sk_fullsock_proto;
7924         case BPF_FUNC_sk_storage_get:
7925                 return &bpf_sk_storage_get_proto;
7926         case BPF_FUNC_sk_storage_delete:
7927                 return &bpf_sk_storage_delete_proto;
7928 #ifdef CONFIG_XFRM
7929         case BPF_FUNC_skb_get_xfrm_state:
7930                 return &bpf_skb_get_xfrm_state_proto;
7931 #endif
7932 #ifdef CONFIG_CGROUP_NET_CLASSID
7933         case BPF_FUNC_skb_cgroup_classid:
7934                 return &bpf_skb_cgroup_classid_proto;
7935 #endif
7936 #ifdef CONFIG_SOCK_CGROUP_DATA
7937         case BPF_FUNC_skb_cgroup_id:
7938                 return &bpf_skb_cgroup_id_proto;
7939         case BPF_FUNC_skb_ancestor_cgroup_id:
7940                 return &bpf_skb_ancestor_cgroup_id_proto;
7941 #endif
7942 #ifdef CONFIG_INET
7943         case BPF_FUNC_sk_lookup_tcp:
7944                 return &bpf_sk_lookup_tcp_proto;
7945         case BPF_FUNC_sk_lookup_udp:
7946                 return &bpf_sk_lookup_udp_proto;
7947         case BPF_FUNC_sk_release:
7948                 return &bpf_sk_release_proto;
7949         case BPF_FUNC_tcp_sock:
7950                 return &bpf_tcp_sock_proto;
7951         case BPF_FUNC_get_listener_sock:
7952                 return &bpf_get_listener_sock_proto;
7953         case BPF_FUNC_skc_lookup_tcp:
7954                 return &bpf_skc_lookup_tcp_proto;
7955         case BPF_FUNC_tcp_check_syncookie:
7956                 return &bpf_tcp_check_syncookie_proto;
7957         case BPF_FUNC_skb_ecn_set_ce:
7958                 return &bpf_skb_ecn_set_ce_proto;
7959         case BPF_FUNC_tcp_gen_syncookie:
7960                 return &bpf_tcp_gen_syncookie_proto;
7961         case BPF_FUNC_sk_assign:
7962                 return &bpf_sk_assign_proto;
7963         case BPF_FUNC_skb_set_tstamp:
7964                 return &bpf_skb_set_tstamp_proto;
7965 #ifdef CONFIG_SYN_COOKIES
7966         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7967                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7968         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7969                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7970         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7971                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7972         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7973                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7974 #endif
7975 #endif
7976         default:
7977                 return bpf_sk_base_func_proto(func_id);
7978         }
7979 }
7980
7981 static const struct bpf_func_proto *
7982 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7983 {
7984         switch (func_id) {
7985         case BPF_FUNC_perf_event_output:
7986                 return &bpf_xdp_event_output_proto;
7987         case BPF_FUNC_get_smp_processor_id:
7988                 return &bpf_get_smp_processor_id_proto;
7989         case BPF_FUNC_csum_diff:
7990                 return &bpf_csum_diff_proto;
7991         case BPF_FUNC_xdp_adjust_head:
7992                 return &bpf_xdp_adjust_head_proto;
7993         case BPF_FUNC_xdp_adjust_meta:
7994                 return &bpf_xdp_adjust_meta_proto;
7995         case BPF_FUNC_redirect:
7996                 return &bpf_xdp_redirect_proto;
7997         case BPF_FUNC_redirect_map:
7998                 return &bpf_xdp_redirect_map_proto;
7999         case BPF_FUNC_xdp_adjust_tail:
8000                 return &bpf_xdp_adjust_tail_proto;
8001         case BPF_FUNC_xdp_get_buff_len:
8002                 return &bpf_xdp_get_buff_len_proto;
8003         case BPF_FUNC_xdp_load_bytes:
8004                 return &bpf_xdp_load_bytes_proto;
8005         case BPF_FUNC_xdp_store_bytes:
8006                 return &bpf_xdp_store_bytes_proto;
8007         case BPF_FUNC_fib_lookup:
8008                 return &bpf_xdp_fib_lookup_proto;
8009         case BPF_FUNC_check_mtu:
8010                 return &bpf_xdp_check_mtu_proto;
8011 #ifdef CONFIG_INET
8012         case BPF_FUNC_sk_lookup_udp:
8013                 return &bpf_xdp_sk_lookup_udp_proto;
8014         case BPF_FUNC_sk_lookup_tcp:
8015                 return &bpf_xdp_sk_lookup_tcp_proto;
8016         case BPF_FUNC_sk_release:
8017                 return &bpf_sk_release_proto;
8018         case BPF_FUNC_skc_lookup_tcp:
8019                 return &bpf_xdp_skc_lookup_tcp_proto;
8020         case BPF_FUNC_tcp_check_syncookie:
8021                 return &bpf_tcp_check_syncookie_proto;
8022         case BPF_FUNC_tcp_gen_syncookie:
8023                 return &bpf_tcp_gen_syncookie_proto;
8024 #ifdef CONFIG_SYN_COOKIES
8025         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
8026                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
8027         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
8028                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
8029         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
8030                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
8031         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
8032                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
8033 #endif
8034 #endif
8035         default:
8036                 return bpf_sk_base_func_proto(func_id);
8037         }
8038
8039 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)
8040         /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The
8041          * kfuncs are defined in two different modules, and we want to be able
8042          * to use them interchangably with the same BTF type ID. Because modules
8043          * can't de-duplicate BTF IDs between each other, we need the type to be
8044          * referenced in the vmlinux BTF or the verifier will get confused about
8045          * the different types. So we add this dummy type reference which will
8046          * be included in vmlinux BTF, allowing both modules to refer to the
8047          * same type ID.
8048          */
8049         BTF_TYPE_EMIT(struct nf_conn___init);
8050 #endif
8051 }
8052
8053 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
8054 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
8055
8056 static const struct bpf_func_proto *
8057 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8058 {
8059         const struct bpf_func_proto *func_proto;
8060
8061         func_proto = cgroup_common_func_proto(func_id, prog);
8062         if (func_proto)
8063                 return func_proto;
8064
8065         switch (func_id) {
8066         case BPF_FUNC_setsockopt:
8067                 return &bpf_sock_ops_setsockopt_proto;
8068         case BPF_FUNC_getsockopt:
8069                 return &bpf_sock_ops_getsockopt_proto;
8070         case BPF_FUNC_sock_ops_cb_flags_set:
8071                 return &bpf_sock_ops_cb_flags_set_proto;
8072         case BPF_FUNC_sock_map_update:
8073                 return &bpf_sock_map_update_proto;
8074         case BPF_FUNC_sock_hash_update:
8075                 return &bpf_sock_hash_update_proto;
8076         case BPF_FUNC_get_socket_cookie:
8077                 return &bpf_get_socket_cookie_sock_ops_proto;
8078         case BPF_FUNC_perf_event_output:
8079                 return &bpf_event_output_data_proto;
8080         case BPF_FUNC_sk_storage_get:
8081                 return &bpf_sk_storage_get_proto;
8082         case BPF_FUNC_sk_storage_delete:
8083                 return &bpf_sk_storage_delete_proto;
8084         case BPF_FUNC_get_netns_cookie:
8085                 return &bpf_get_netns_cookie_sock_ops_proto;
8086 #ifdef CONFIG_INET
8087         case BPF_FUNC_load_hdr_opt:
8088                 return &bpf_sock_ops_load_hdr_opt_proto;
8089         case BPF_FUNC_store_hdr_opt:
8090                 return &bpf_sock_ops_store_hdr_opt_proto;
8091         case BPF_FUNC_reserve_hdr_opt:
8092                 return &bpf_sock_ops_reserve_hdr_opt_proto;
8093         case BPF_FUNC_tcp_sock:
8094                 return &bpf_tcp_sock_proto;
8095 #endif /* CONFIG_INET */
8096         default:
8097                 return bpf_sk_base_func_proto(func_id);
8098         }
8099 }
8100
8101 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
8102 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
8103
8104 static const struct bpf_func_proto *
8105 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8106 {
8107         switch (func_id) {
8108         case BPF_FUNC_msg_redirect_map:
8109                 return &bpf_msg_redirect_map_proto;
8110         case BPF_FUNC_msg_redirect_hash:
8111                 return &bpf_msg_redirect_hash_proto;
8112         case BPF_FUNC_msg_apply_bytes:
8113                 return &bpf_msg_apply_bytes_proto;
8114         case BPF_FUNC_msg_cork_bytes:
8115                 return &bpf_msg_cork_bytes_proto;
8116         case BPF_FUNC_msg_pull_data:
8117                 return &bpf_msg_pull_data_proto;
8118         case BPF_FUNC_msg_push_data:
8119                 return &bpf_msg_push_data_proto;
8120         case BPF_FUNC_msg_pop_data:
8121                 return &bpf_msg_pop_data_proto;
8122         case BPF_FUNC_perf_event_output:
8123                 return &bpf_event_output_data_proto;
8124         case BPF_FUNC_get_current_uid_gid:
8125                 return &bpf_get_current_uid_gid_proto;
8126         case BPF_FUNC_get_current_pid_tgid:
8127                 return &bpf_get_current_pid_tgid_proto;
8128         case BPF_FUNC_sk_storage_get:
8129                 return &bpf_sk_storage_get_proto;
8130         case BPF_FUNC_sk_storage_delete:
8131                 return &bpf_sk_storage_delete_proto;
8132         case BPF_FUNC_get_netns_cookie:
8133                 return &bpf_get_netns_cookie_sk_msg_proto;
8134 #ifdef CONFIG_CGROUPS
8135         case BPF_FUNC_get_current_cgroup_id:
8136                 return &bpf_get_current_cgroup_id_proto;
8137         case BPF_FUNC_get_current_ancestor_cgroup_id:
8138                 return &bpf_get_current_ancestor_cgroup_id_proto;
8139 #endif
8140 #ifdef CONFIG_CGROUP_NET_CLASSID
8141         case BPF_FUNC_get_cgroup_classid:
8142                 return &bpf_get_cgroup_classid_curr_proto;
8143 #endif
8144         default:
8145                 return bpf_sk_base_func_proto(func_id);
8146         }
8147 }
8148
8149 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8150 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8151
8152 static const struct bpf_func_proto *
8153 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8154 {
8155         switch (func_id) {
8156         case BPF_FUNC_skb_store_bytes:
8157                 return &bpf_skb_store_bytes_proto;
8158         case BPF_FUNC_skb_load_bytes:
8159                 return &bpf_skb_load_bytes_proto;
8160         case BPF_FUNC_skb_pull_data:
8161                 return &sk_skb_pull_data_proto;
8162         case BPF_FUNC_skb_change_tail:
8163                 return &sk_skb_change_tail_proto;
8164         case BPF_FUNC_skb_change_head:
8165                 return &sk_skb_change_head_proto;
8166         case BPF_FUNC_skb_adjust_room:
8167                 return &sk_skb_adjust_room_proto;
8168         case BPF_FUNC_get_socket_cookie:
8169                 return &bpf_get_socket_cookie_proto;
8170         case BPF_FUNC_get_socket_uid:
8171                 return &bpf_get_socket_uid_proto;
8172         case BPF_FUNC_sk_redirect_map:
8173                 return &bpf_sk_redirect_map_proto;
8174         case BPF_FUNC_sk_redirect_hash:
8175                 return &bpf_sk_redirect_hash_proto;
8176         case BPF_FUNC_perf_event_output:
8177                 return &bpf_skb_event_output_proto;
8178 #ifdef CONFIG_INET
8179         case BPF_FUNC_sk_lookup_tcp:
8180                 return &bpf_sk_lookup_tcp_proto;
8181         case BPF_FUNC_sk_lookup_udp:
8182                 return &bpf_sk_lookup_udp_proto;
8183         case BPF_FUNC_sk_release:
8184                 return &bpf_sk_release_proto;
8185         case BPF_FUNC_skc_lookup_tcp:
8186                 return &bpf_skc_lookup_tcp_proto;
8187 #endif
8188         default:
8189                 return bpf_sk_base_func_proto(func_id);
8190         }
8191 }
8192
8193 static const struct bpf_func_proto *
8194 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8195 {
8196         switch (func_id) {
8197         case BPF_FUNC_skb_load_bytes:
8198                 return &bpf_flow_dissector_load_bytes_proto;
8199         default:
8200                 return bpf_sk_base_func_proto(func_id);
8201         }
8202 }
8203
8204 static const struct bpf_func_proto *
8205 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8206 {
8207         switch (func_id) {
8208         case BPF_FUNC_skb_load_bytes:
8209                 return &bpf_skb_load_bytes_proto;
8210         case BPF_FUNC_skb_pull_data:
8211                 return &bpf_skb_pull_data_proto;
8212         case BPF_FUNC_csum_diff:
8213                 return &bpf_csum_diff_proto;
8214         case BPF_FUNC_get_cgroup_classid:
8215                 return &bpf_get_cgroup_classid_proto;
8216         case BPF_FUNC_get_route_realm:
8217                 return &bpf_get_route_realm_proto;
8218         case BPF_FUNC_get_hash_recalc:
8219                 return &bpf_get_hash_recalc_proto;
8220         case BPF_FUNC_perf_event_output:
8221                 return &bpf_skb_event_output_proto;
8222         case BPF_FUNC_get_smp_processor_id:
8223                 return &bpf_get_smp_processor_id_proto;
8224         case BPF_FUNC_skb_under_cgroup:
8225                 return &bpf_skb_under_cgroup_proto;
8226         default:
8227                 return bpf_sk_base_func_proto(func_id);
8228         }
8229 }
8230
8231 static const struct bpf_func_proto *
8232 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8233 {
8234         switch (func_id) {
8235         case BPF_FUNC_lwt_push_encap:
8236                 return &bpf_lwt_in_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_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8244 {
8245         switch (func_id) {
8246         case BPF_FUNC_skb_get_tunnel_key:
8247                 return &bpf_skb_get_tunnel_key_proto;
8248         case BPF_FUNC_skb_set_tunnel_key:
8249                 return bpf_get_skb_set_tunnel_proto(func_id);
8250         case BPF_FUNC_skb_get_tunnel_opt:
8251                 return &bpf_skb_get_tunnel_opt_proto;
8252         case BPF_FUNC_skb_set_tunnel_opt:
8253                 return bpf_get_skb_set_tunnel_proto(func_id);
8254         case BPF_FUNC_redirect:
8255                 return &bpf_redirect_proto;
8256         case BPF_FUNC_clone_redirect:
8257                 return &bpf_clone_redirect_proto;
8258         case BPF_FUNC_skb_change_tail:
8259                 return &bpf_skb_change_tail_proto;
8260         case BPF_FUNC_skb_change_head:
8261                 return &bpf_skb_change_head_proto;
8262         case BPF_FUNC_skb_store_bytes:
8263                 return &bpf_skb_store_bytes_proto;
8264         case BPF_FUNC_csum_update:
8265                 return &bpf_csum_update_proto;
8266         case BPF_FUNC_csum_level:
8267                 return &bpf_csum_level_proto;
8268         case BPF_FUNC_l3_csum_replace:
8269                 return &bpf_l3_csum_replace_proto;
8270         case BPF_FUNC_l4_csum_replace:
8271                 return &bpf_l4_csum_replace_proto;
8272         case BPF_FUNC_set_hash_invalid:
8273                 return &bpf_set_hash_invalid_proto;
8274         case BPF_FUNC_lwt_push_encap:
8275                 return &bpf_lwt_xmit_push_encap_proto;
8276         default:
8277                 return lwt_out_func_proto(func_id, prog);
8278         }
8279 }
8280
8281 static const struct bpf_func_proto *
8282 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8283 {
8284         switch (func_id) {
8285 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8286         case BPF_FUNC_lwt_seg6_store_bytes:
8287                 return &bpf_lwt_seg6_store_bytes_proto;
8288         case BPF_FUNC_lwt_seg6_action:
8289                 return &bpf_lwt_seg6_action_proto;
8290         case BPF_FUNC_lwt_seg6_adjust_srh:
8291                 return &bpf_lwt_seg6_adjust_srh_proto;
8292 #endif
8293         default:
8294                 return lwt_out_func_proto(func_id, prog);
8295         }
8296 }
8297
8298 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8299                                     const struct bpf_prog *prog,
8300                                     struct bpf_insn_access_aux *info)
8301 {
8302         const int size_default = sizeof(__u32);
8303
8304         if (off < 0 || off >= sizeof(struct __sk_buff))
8305                 return false;
8306
8307         /* The verifier guarantees that size > 0. */
8308         if (off % size != 0)
8309                 return false;
8310
8311         switch (off) {
8312         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8313                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8314                         return false;
8315                 break;
8316         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8317         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8318         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8319         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8320         case bpf_ctx_range(struct __sk_buff, data):
8321         case bpf_ctx_range(struct __sk_buff, data_meta):
8322         case bpf_ctx_range(struct __sk_buff, data_end):
8323                 if (size != size_default)
8324                         return false;
8325                 break;
8326         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8327                 return false;
8328         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8329                 if (type == BPF_WRITE || size != sizeof(__u64))
8330                         return false;
8331                 break;
8332         case bpf_ctx_range(struct __sk_buff, tstamp):
8333                 if (size != sizeof(__u64))
8334                         return false;
8335                 break;
8336         case offsetof(struct __sk_buff, sk):
8337                 if (type == BPF_WRITE || size != sizeof(__u64))
8338                         return false;
8339                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8340                 break;
8341         case offsetof(struct __sk_buff, tstamp_type):
8342                 return false;
8343         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8344                 /* Explicitly prohibit access to padding in __sk_buff. */
8345                 return false;
8346         default:
8347                 /* Only narrow read access allowed for now. */
8348                 if (type == BPF_WRITE) {
8349                         if (size != size_default)
8350                                 return false;
8351                 } else {
8352                         bpf_ctx_record_field_size(info, size_default);
8353                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8354                                 return false;
8355                 }
8356         }
8357
8358         return true;
8359 }
8360
8361 static bool sk_filter_is_valid_access(int off, int size,
8362                                       enum bpf_access_type type,
8363                                       const struct bpf_prog *prog,
8364                                       struct bpf_insn_access_aux *info)
8365 {
8366         switch (off) {
8367         case bpf_ctx_range(struct __sk_buff, tc_classid):
8368         case bpf_ctx_range(struct __sk_buff, data):
8369         case bpf_ctx_range(struct __sk_buff, data_meta):
8370         case bpf_ctx_range(struct __sk_buff, data_end):
8371         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8372         case bpf_ctx_range(struct __sk_buff, tstamp):
8373         case bpf_ctx_range(struct __sk_buff, wire_len):
8374         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8375                 return false;
8376         }
8377
8378         if (type == BPF_WRITE) {
8379                 switch (off) {
8380                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8381                         break;
8382                 default:
8383                         return false;
8384                 }
8385         }
8386
8387         return bpf_skb_is_valid_access(off, size, type, prog, info);
8388 }
8389
8390 static bool cg_skb_is_valid_access(int off, int size,
8391                                    enum bpf_access_type type,
8392                                    const struct bpf_prog *prog,
8393                                    struct bpf_insn_access_aux *info)
8394 {
8395         switch (off) {
8396         case bpf_ctx_range(struct __sk_buff, tc_classid):
8397         case bpf_ctx_range(struct __sk_buff, data_meta):
8398         case bpf_ctx_range(struct __sk_buff, wire_len):
8399                 return false;
8400         case bpf_ctx_range(struct __sk_buff, data):
8401         case bpf_ctx_range(struct __sk_buff, data_end):
8402                 if (!bpf_capable())
8403                         return false;
8404                 break;
8405         }
8406
8407         if (type == BPF_WRITE) {
8408                 switch (off) {
8409                 case bpf_ctx_range(struct __sk_buff, mark):
8410                 case bpf_ctx_range(struct __sk_buff, priority):
8411                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8412                         break;
8413                 case bpf_ctx_range(struct __sk_buff, tstamp):
8414                         if (!bpf_capable())
8415                                 return false;
8416                         break;
8417                 default:
8418                         return false;
8419                 }
8420         }
8421
8422         switch (off) {
8423         case bpf_ctx_range(struct __sk_buff, data):
8424                 info->reg_type = PTR_TO_PACKET;
8425                 break;
8426         case bpf_ctx_range(struct __sk_buff, data_end):
8427                 info->reg_type = PTR_TO_PACKET_END;
8428                 break;
8429         }
8430
8431         return bpf_skb_is_valid_access(off, size, type, prog, info);
8432 }
8433
8434 static bool lwt_is_valid_access(int off, int size,
8435                                 enum bpf_access_type type,
8436                                 const struct bpf_prog *prog,
8437                                 struct bpf_insn_access_aux *info)
8438 {
8439         switch (off) {
8440         case bpf_ctx_range(struct __sk_buff, tc_classid):
8441         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8442         case bpf_ctx_range(struct __sk_buff, data_meta):
8443         case bpf_ctx_range(struct __sk_buff, tstamp):
8444         case bpf_ctx_range(struct __sk_buff, wire_len):
8445         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8446                 return false;
8447         }
8448
8449         if (type == BPF_WRITE) {
8450                 switch (off) {
8451                 case bpf_ctx_range(struct __sk_buff, mark):
8452                 case bpf_ctx_range(struct __sk_buff, priority):
8453                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8454                         break;
8455                 default:
8456                         return false;
8457                 }
8458         }
8459
8460         switch (off) {
8461         case bpf_ctx_range(struct __sk_buff, data):
8462                 info->reg_type = PTR_TO_PACKET;
8463                 break;
8464         case bpf_ctx_range(struct __sk_buff, data_end):
8465                 info->reg_type = PTR_TO_PACKET_END;
8466                 break;
8467         }
8468
8469         return bpf_skb_is_valid_access(off, size, type, prog, info);
8470 }
8471
8472 /* Attach type specific accesses */
8473 static bool __sock_filter_check_attach_type(int off,
8474                                             enum bpf_access_type access_type,
8475                                             enum bpf_attach_type attach_type)
8476 {
8477         switch (off) {
8478         case offsetof(struct bpf_sock, bound_dev_if):
8479         case offsetof(struct bpf_sock, mark):
8480         case offsetof(struct bpf_sock, priority):
8481                 switch (attach_type) {
8482                 case BPF_CGROUP_INET_SOCK_CREATE:
8483                 case BPF_CGROUP_INET_SOCK_RELEASE:
8484                         goto full_access;
8485                 default:
8486                         return false;
8487                 }
8488         case bpf_ctx_range(struct bpf_sock, src_ip4):
8489                 switch (attach_type) {
8490                 case BPF_CGROUP_INET4_POST_BIND:
8491                         goto read_only;
8492                 default:
8493                         return false;
8494                 }
8495         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8496                 switch (attach_type) {
8497                 case BPF_CGROUP_INET6_POST_BIND:
8498                         goto read_only;
8499                 default:
8500                         return false;
8501                 }
8502         case bpf_ctx_range(struct bpf_sock, src_port):
8503                 switch (attach_type) {
8504                 case BPF_CGROUP_INET4_POST_BIND:
8505                 case BPF_CGROUP_INET6_POST_BIND:
8506                         goto read_only;
8507                 default:
8508                         return false;
8509                 }
8510         }
8511 read_only:
8512         return access_type == BPF_READ;
8513 full_access:
8514         return true;
8515 }
8516
8517 bool bpf_sock_common_is_valid_access(int off, int size,
8518                                      enum bpf_access_type type,
8519                                      struct bpf_insn_access_aux *info)
8520 {
8521         switch (off) {
8522         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8523                 return false;
8524         default:
8525                 return bpf_sock_is_valid_access(off, size, type, info);
8526         }
8527 }
8528
8529 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8530                               struct bpf_insn_access_aux *info)
8531 {
8532         const int size_default = sizeof(__u32);
8533         int field_size;
8534
8535         if (off < 0 || off >= sizeof(struct bpf_sock))
8536                 return false;
8537         if (off % size != 0)
8538                 return false;
8539
8540         switch (off) {
8541         case offsetof(struct bpf_sock, state):
8542         case offsetof(struct bpf_sock, family):
8543         case offsetof(struct bpf_sock, type):
8544         case offsetof(struct bpf_sock, protocol):
8545         case offsetof(struct bpf_sock, src_port):
8546         case offsetof(struct bpf_sock, rx_queue_mapping):
8547         case bpf_ctx_range(struct bpf_sock, src_ip4):
8548         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8549         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8550         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8551                 bpf_ctx_record_field_size(info, size_default);
8552                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8553         case bpf_ctx_range(struct bpf_sock, dst_port):
8554                 field_size = size == size_default ?
8555                         size_default : sizeof_field(struct bpf_sock, dst_port);
8556                 bpf_ctx_record_field_size(info, field_size);
8557                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8558         case offsetofend(struct bpf_sock, dst_port) ...
8559              offsetof(struct bpf_sock, dst_ip4) - 1:
8560                 return false;
8561         }
8562
8563         return size == size_default;
8564 }
8565
8566 static bool sock_filter_is_valid_access(int off, int size,
8567                                         enum bpf_access_type type,
8568                                         const struct bpf_prog *prog,
8569                                         struct bpf_insn_access_aux *info)
8570 {
8571         if (!bpf_sock_is_valid_access(off, size, type, info))
8572                 return false;
8573         return __sock_filter_check_attach_type(off, type,
8574                                                prog->expected_attach_type);
8575 }
8576
8577 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8578                              const struct bpf_prog *prog)
8579 {
8580         /* Neither direct read nor direct write requires any preliminary
8581          * action.
8582          */
8583         return 0;
8584 }
8585
8586 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8587                                 const struct bpf_prog *prog, int drop_verdict)
8588 {
8589         struct bpf_insn *insn = insn_buf;
8590
8591         if (!direct_write)
8592                 return 0;
8593
8594         /* if (!skb->cloned)
8595          *       goto start;
8596          *
8597          * (Fast-path, otherwise approximation that we might be
8598          *  a clone, do the rest in helper.)
8599          */
8600         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8601         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8602         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8603
8604         /* ret = bpf_skb_pull_data(skb, 0); */
8605         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8606         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8607         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8608                                BPF_FUNC_skb_pull_data);
8609         /* if (!ret)
8610          *      goto restore;
8611          * return TC_ACT_SHOT;
8612          */
8613         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8614         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8615         *insn++ = BPF_EXIT_INSN();
8616
8617         /* restore: */
8618         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8619         /* start: */
8620         *insn++ = prog->insnsi[0];
8621
8622         return insn - insn_buf;
8623 }
8624
8625 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8626                           struct bpf_insn *insn_buf)
8627 {
8628         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8629         struct bpf_insn *insn = insn_buf;
8630
8631         if (!indirect) {
8632                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8633         } else {
8634                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8635                 if (orig->imm)
8636                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8637         }
8638         /* We're guaranteed here that CTX is in R6. */
8639         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8640
8641         switch (BPF_SIZE(orig->code)) {
8642         case BPF_B:
8643                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8644                 break;
8645         case BPF_H:
8646                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8647                 break;
8648         case BPF_W:
8649                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8650                 break;
8651         }
8652
8653         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8654         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8655         *insn++ = BPF_EXIT_INSN();
8656
8657         return insn - insn_buf;
8658 }
8659
8660 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8661                                const struct bpf_prog *prog)
8662 {
8663         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8664 }
8665
8666 static bool tc_cls_act_is_valid_access(int off, int size,
8667                                        enum bpf_access_type type,
8668                                        const struct bpf_prog *prog,
8669                                        struct bpf_insn_access_aux *info)
8670 {
8671         if (type == BPF_WRITE) {
8672                 switch (off) {
8673                 case bpf_ctx_range(struct __sk_buff, mark):
8674                 case bpf_ctx_range(struct __sk_buff, tc_index):
8675                 case bpf_ctx_range(struct __sk_buff, priority):
8676                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8677                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8678                 case bpf_ctx_range(struct __sk_buff, tstamp):
8679                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8680                         break;
8681                 default:
8682                         return false;
8683                 }
8684         }
8685
8686         switch (off) {
8687         case bpf_ctx_range(struct __sk_buff, data):
8688                 info->reg_type = PTR_TO_PACKET;
8689                 break;
8690         case bpf_ctx_range(struct __sk_buff, data_meta):
8691                 info->reg_type = PTR_TO_PACKET_META;
8692                 break;
8693         case bpf_ctx_range(struct __sk_buff, data_end):
8694                 info->reg_type = PTR_TO_PACKET_END;
8695                 break;
8696         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8697                 return false;
8698         case offsetof(struct __sk_buff, tstamp_type):
8699                 /* The convert_ctx_access() on reading and writing
8700                  * __sk_buff->tstamp depends on whether the bpf prog
8701                  * has used __sk_buff->tstamp_type or not.
8702                  * Thus, we need to set prog->tstamp_type_access
8703                  * earlier during is_valid_access() here.
8704                  */
8705                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8706                 return size == sizeof(__u8);
8707         }
8708
8709         return bpf_skb_is_valid_access(off, size, type, prog, info);
8710 }
8711
8712 DEFINE_MUTEX(nf_conn_btf_access_lock);
8713 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock);
8714
8715 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log,
8716                               const struct bpf_reg_state *reg,
8717                               int off, int size, enum bpf_access_type atype,
8718                               u32 *next_btf_id, enum bpf_type_flag *flag);
8719 EXPORT_SYMBOL_GPL(nfct_btf_struct_access);
8720
8721 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log,
8722                                         const struct bpf_reg_state *reg,
8723                                         int off, int size, enum bpf_access_type atype,
8724                                         u32 *next_btf_id, enum bpf_type_flag *flag)
8725 {
8726         int ret = -EACCES;
8727
8728         if (atype == BPF_READ)
8729                 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8730
8731         mutex_lock(&nf_conn_btf_access_lock);
8732         if (nfct_btf_struct_access)
8733                 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8734         mutex_unlock(&nf_conn_btf_access_lock);
8735
8736         return ret;
8737 }
8738
8739 static bool __is_valid_xdp_access(int off, int size)
8740 {
8741         if (off < 0 || off >= sizeof(struct xdp_md))
8742                 return false;
8743         if (off % size != 0)
8744                 return false;
8745         if (size != sizeof(__u32))
8746                 return false;
8747
8748         return true;
8749 }
8750
8751 static bool xdp_is_valid_access(int off, int size,
8752                                 enum bpf_access_type type,
8753                                 const struct bpf_prog *prog,
8754                                 struct bpf_insn_access_aux *info)
8755 {
8756         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8757                 switch (off) {
8758                 case offsetof(struct xdp_md, egress_ifindex):
8759                         return false;
8760                 }
8761         }
8762
8763         if (type == BPF_WRITE) {
8764                 if (bpf_prog_is_offloaded(prog->aux)) {
8765                         switch (off) {
8766                         case offsetof(struct xdp_md, rx_queue_index):
8767                                 return __is_valid_xdp_access(off, size);
8768                         }
8769                 }
8770                 return false;
8771         }
8772
8773         switch (off) {
8774         case offsetof(struct xdp_md, data):
8775                 info->reg_type = PTR_TO_PACKET;
8776                 break;
8777         case offsetof(struct xdp_md, data_meta):
8778                 info->reg_type = PTR_TO_PACKET_META;
8779                 break;
8780         case offsetof(struct xdp_md, data_end):
8781                 info->reg_type = PTR_TO_PACKET_END;
8782                 break;
8783         }
8784
8785         return __is_valid_xdp_access(off, size);
8786 }
8787
8788 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8789 {
8790         const u32 act_max = XDP_REDIRECT;
8791
8792         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8793                      act > act_max ? "Illegal" : "Driver unsupported",
8794                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8795 }
8796 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8797
8798 static int xdp_btf_struct_access(struct bpf_verifier_log *log,
8799                                  const struct bpf_reg_state *reg,
8800                                  int off, int size, enum bpf_access_type atype,
8801                                  u32 *next_btf_id, enum bpf_type_flag *flag)
8802 {
8803         int ret = -EACCES;
8804
8805         if (atype == BPF_READ)
8806                 return btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8807
8808         mutex_lock(&nf_conn_btf_access_lock);
8809         if (nfct_btf_struct_access)
8810                 ret = nfct_btf_struct_access(log, reg, off, size, atype, next_btf_id, flag);
8811         mutex_unlock(&nf_conn_btf_access_lock);
8812
8813         return ret;
8814 }
8815
8816 static bool sock_addr_is_valid_access(int off, int size,
8817                                       enum bpf_access_type type,
8818                                       const struct bpf_prog *prog,
8819                                       struct bpf_insn_access_aux *info)
8820 {
8821         const int size_default = sizeof(__u32);
8822
8823         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8824                 return false;
8825         if (off % size != 0)
8826                 return false;
8827
8828         /* Disallow access to IPv6 fields from IPv4 contex and vise
8829          * versa.
8830          */
8831         switch (off) {
8832         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8833                 switch (prog->expected_attach_type) {
8834                 case BPF_CGROUP_INET4_BIND:
8835                 case BPF_CGROUP_INET4_CONNECT:
8836                 case BPF_CGROUP_INET4_GETPEERNAME:
8837                 case BPF_CGROUP_INET4_GETSOCKNAME:
8838                 case BPF_CGROUP_UDP4_SENDMSG:
8839                 case BPF_CGROUP_UDP4_RECVMSG:
8840                         break;
8841                 default:
8842                         return false;
8843                 }
8844                 break;
8845         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8846                 switch (prog->expected_attach_type) {
8847                 case BPF_CGROUP_INET6_BIND:
8848                 case BPF_CGROUP_INET6_CONNECT:
8849                 case BPF_CGROUP_INET6_GETPEERNAME:
8850                 case BPF_CGROUP_INET6_GETSOCKNAME:
8851                 case BPF_CGROUP_UDP6_SENDMSG:
8852                 case BPF_CGROUP_UDP6_RECVMSG:
8853                         break;
8854                 default:
8855                         return false;
8856                 }
8857                 break;
8858         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8859                 switch (prog->expected_attach_type) {
8860                 case BPF_CGROUP_UDP4_SENDMSG:
8861                         break;
8862                 default:
8863                         return false;
8864                 }
8865                 break;
8866         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8867                                 msg_src_ip6[3]):
8868                 switch (prog->expected_attach_type) {
8869                 case BPF_CGROUP_UDP6_SENDMSG:
8870                         break;
8871                 default:
8872                         return false;
8873                 }
8874                 break;
8875         }
8876
8877         switch (off) {
8878         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8879         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8880         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8881         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8882                                 msg_src_ip6[3]):
8883         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8884                 if (type == BPF_READ) {
8885                         bpf_ctx_record_field_size(info, size_default);
8886
8887                         if (bpf_ctx_wide_access_ok(off, size,
8888                                                    struct bpf_sock_addr,
8889                                                    user_ip6))
8890                                 return true;
8891
8892                         if (bpf_ctx_wide_access_ok(off, size,
8893                                                    struct bpf_sock_addr,
8894                                                    msg_src_ip6))
8895                                 return true;
8896
8897                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8898                                 return false;
8899                 } else {
8900                         if (bpf_ctx_wide_access_ok(off, size,
8901                                                    struct bpf_sock_addr,
8902                                                    user_ip6))
8903                                 return true;
8904
8905                         if (bpf_ctx_wide_access_ok(off, size,
8906                                                    struct bpf_sock_addr,
8907                                                    msg_src_ip6))
8908                                 return true;
8909
8910                         if (size != size_default)
8911                                 return false;
8912                 }
8913                 break;
8914         case offsetof(struct bpf_sock_addr, sk):
8915                 if (type != BPF_READ)
8916                         return false;
8917                 if (size != sizeof(__u64))
8918                         return false;
8919                 info->reg_type = PTR_TO_SOCKET;
8920                 break;
8921         default:
8922                 if (type == BPF_READ) {
8923                         if (size != size_default)
8924                                 return false;
8925                 } else {
8926                         return false;
8927                 }
8928         }
8929
8930         return true;
8931 }
8932
8933 static bool sock_ops_is_valid_access(int off, int size,
8934                                      enum bpf_access_type type,
8935                                      const struct bpf_prog *prog,
8936                                      struct bpf_insn_access_aux *info)
8937 {
8938         const int size_default = sizeof(__u32);
8939
8940         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8941                 return false;
8942
8943         /* The verifier guarantees that size > 0. */
8944         if (off % size != 0)
8945                 return false;
8946
8947         if (type == BPF_WRITE) {
8948                 switch (off) {
8949                 case offsetof(struct bpf_sock_ops, reply):
8950                 case offsetof(struct bpf_sock_ops, sk_txhash):
8951                         if (size != size_default)
8952                                 return false;
8953                         break;
8954                 default:
8955                         return false;
8956                 }
8957         } else {
8958                 switch (off) {
8959                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8960                                         bytes_acked):
8961                         if (size != sizeof(__u64))
8962                                 return false;
8963                         break;
8964                 case offsetof(struct bpf_sock_ops, sk):
8965                         if (size != sizeof(__u64))
8966                                 return false;
8967                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8968                         break;
8969                 case offsetof(struct bpf_sock_ops, skb_data):
8970                         if (size != sizeof(__u64))
8971                                 return false;
8972                         info->reg_type = PTR_TO_PACKET;
8973                         break;
8974                 case offsetof(struct bpf_sock_ops, skb_data_end):
8975                         if (size != sizeof(__u64))
8976                                 return false;
8977                         info->reg_type = PTR_TO_PACKET_END;
8978                         break;
8979                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8980                         bpf_ctx_record_field_size(info, size_default);
8981                         return bpf_ctx_narrow_access_ok(off, size,
8982                                                         size_default);
8983                 case offsetof(struct bpf_sock_ops, skb_hwtstamp):
8984                         if (size != sizeof(__u64))
8985                                 return false;
8986                         break;
8987                 default:
8988                         if (size != size_default)
8989                                 return false;
8990                         break;
8991                 }
8992         }
8993
8994         return true;
8995 }
8996
8997 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8998                            const struct bpf_prog *prog)
8999 {
9000         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
9001 }
9002
9003 static bool sk_skb_is_valid_access(int off, int size,
9004                                    enum bpf_access_type type,
9005                                    const struct bpf_prog *prog,
9006                                    struct bpf_insn_access_aux *info)
9007 {
9008         switch (off) {
9009         case bpf_ctx_range(struct __sk_buff, tc_classid):
9010         case bpf_ctx_range(struct __sk_buff, data_meta):
9011         case bpf_ctx_range(struct __sk_buff, tstamp):
9012         case bpf_ctx_range(struct __sk_buff, wire_len):
9013         case bpf_ctx_range(struct __sk_buff, hwtstamp):
9014                 return false;
9015         }
9016
9017         if (type == BPF_WRITE) {
9018                 switch (off) {
9019                 case bpf_ctx_range(struct __sk_buff, tc_index):
9020                 case bpf_ctx_range(struct __sk_buff, priority):
9021                         break;
9022                 default:
9023                         return false;
9024                 }
9025         }
9026
9027         switch (off) {
9028         case bpf_ctx_range(struct __sk_buff, mark):
9029                 return false;
9030         case bpf_ctx_range(struct __sk_buff, data):
9031                 info->reg_type = PTR_TO_PACKET;
9032                 break;
9033         case bpf_ctx_range(struct __sk_buff, data_end):
9034                 info->reg_type = PTR_TO_PACKET_END;
9035                 break;
9036         }
9037
9038         return bpf_skb_is_valid_access(off, size, type, prog, info);
9039 }
9040
9041 static bool sk_msg_is_valid_access(int off, int size,
9042                                    enum bpf_access_type type,
9043                                    const struct bpf_prog *prog,
9044                                    struct bpf_insn_access_aux *info)
9045 {
9046         if (type == BPF_WRITE)
9047                 return false;
9048
9049         if (off % size != 0)
9050                 return false;
9051
9052         switch (off) {
9053         case offsetof(struct sk_msg_md, data):
9054                 info->reg_type = PTR_TO_PACKET;
9055                 if (size != sizeof(__u64))
9056                         return false;
9057                 break;
9058         case offsetof(struct sk_msg_md, data_end):
9059                 info->reg_type = PTR_TO_PACKET_END;
9060                 if (size != sizeof(__u64))
9061                         return false;
9062                 break;
9063         case offsetof(struct sk_msg_md, sk):
9064                 if (size != sizeof(__u64))
9065                         return false;
9066                 info->reg_type = PTR_TO_SOCKET;
9067                 break;
9068         case bpf_ctx_range(struct sk_msg_md, family):
9069         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
9070         case bpf_ctx_range(struct sk_msg_md, local_ip4):
9071         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
9072         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
9073         case bpf_ctx_range(struct sk_msg_md, remote_port):
9074         case bpf_ctx_range(struct sk_msg_md, local_port):
9075         case bpf_ctx_range(struct sk_msg_md, size):
9076                 if (size != sizeof(__u32))
9077                         return false;
9078                 break;
9079         default:
9080                 return false;
9081         }
9082         return true;
9083 }
9084
9085 static bool flow_dissector_is_valid_access(int off, int size,
9086                                            enum bpf_access_type type,
9087                                            const struct bpf_prog *prog,
9088                                            struct bpf_insn_access_aux *info)
9089 {
9090         const int size_default = sizeof(__u32);
9091
9092         if (off < 0 || off >= sizeof(struct __sk_buff))
9093                 return false;
9094
9095         if (type == BPF_WRITE)
9096                 return false;
9097
9098         switch (off) {
9099         case bpf_ctx_range(struct __sk_buff, data):
9100                 if (size != size_default)
9101                         return false;
9102                 info->reg_type = PTR_TO_PACKET;
9103                 return true;
9104         case bpf_ctx_range(struct __sk_buff, data_end):
9105                 if (size != size_default)
9106                         return false;
9107                 info->reg_type = PTR_TO_PACKET_END;
9108                 return true;
9109         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
9110                 if (size != sizeof(__u64))
9111                         return false;
9112                 info->reg_type = PTR_TO_FLOW_KEYS;
9113                 return true;
9114         default:
9115                 return false;
9116         }
9117 }
9118
9119 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
9120                                              const struct bpf_insn *si,
9121                                              struct bpf_insn *insn_buf,
9122                                              struct bpf_prog *prog,
9123                                              u32 *target_size)
9124
9125 {
9126         struct bpf_insn *insn = insn_buf;
9127
9128         switch (si->off) {
9129         case offsetof(struct __sk_buff, data):
9130                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
9131                                       si->dst_reg, si->src_reg,
9132                                       offsetof(struct bpf_flow_dissector, data));
9133                 break;
9134
9135         case offsetof(struct __sk_buff, data_end):
9136                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
9137                                       si->dst_reg, si->src_reg,
9138                                       offsetof(struct bpf_flow_dissector, data_end));
9139                 break;
9140
9141         case offsetof(struct __sk_buff, flow_keys):
9142                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
9143                                       si->dst_reg, si->src_reg,
9144                                       offsetof(struct bpf_flow_dissector, flow_keys));
9145                 break;
9146         }
9147
9148         return insn - insn_buf;
9149 }
9150
9151 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
9152                                                      struct bpf_insn *insn)
9153 {
9154         __u8 value_reg = si->dst_reg;
9155         __u8 skb_reg = si->src_reg;
9156         /* AX is needed because src_reg and dst_reg could be the same */
9157         __u8 tmp_reg = BPF_REG_AX;
9158
9159         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9160                               PKT_VLAN_PRESENT_OFFSET);
9161         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9162                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9163         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9164         *insn++ = BPF_JMP_A(1);
9165         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9166
9167         return insn;
9168 }
9169
9170 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg,
9171                                                   struct bpf_insn *insn)
9172 {
9173         /* si->dst_reg = skb_shinfo(SKB); */
9174 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9175         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9176                               BPF_REG_AX, skb_reg,
9177                               offsetof(struct sk_buff, end));
9178         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9179                               dst_reg, skb_reg,
9180                               offsetof(struct sk_buff, head));
9181         *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX);
9182 #else
9183         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9184                               dst_reg, skb_reg,
9185                               offsetof(struct sk_buff, end));
9186 #endif
9187
9188         return insn;
9189 }
9190
9191 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9192                                                 const struct bpf_insn *si,
9193                                                 struct bpf_insn *insn)
9194 {
9195         __u8 value_reg = si->dst_reg;
9196         __u8 skb_reg = si->src_reg;
9197
9198 #ifdef CONFIG_NET_CLS_ACT
9199         /* If the tstamp_type is read,
9200          * the bpf prog is aware the tstamp could have delivery time.
9201          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9202          */
9203         if (!prog->tstamp_type_access) {
9204                 /* AX is needed because src_reg and dst_reg could be the same */
9205                 __u8 tmp_reg = BPF_REG_AX;
9206
9207                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9208                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9209                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9210                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9211                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9212                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9213                  * read 0 as the (rcv) timestamp.
9214                  */
9215                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9216                 *insn++ = BPF_JMP_A(1);
9217         }
9218 #endif
9219
9220         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9221                               offsetof(struct sk_buff, tstamp));
9222         return insn;
9223 }
9224
9225 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9226                                                  const struct bpf_insn *si,
9227                                                  struct bpf_insn *insn)
9228 {
9229         __u8 value_reg = si->src_reg;
9230         __u8 skb_reg = si->dst_reg;
9231
9232 #ifdef CONFIG_NET_CLS_ACT
9233         /* If the tstamp_type is read,
9234          * the bpf prog is aware the tstamp could have delivery time.
9235          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9236          * Otherwise, writing at ingress will have to clear the
9237          * mono_delivery_time bit also.
9238          */
9239         if (!prog->tstamp_type_access) {
9240                 __u8 tmp_reg = BPF_REG_AX;
9241
9242                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9243                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9244                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9245                 /* goto <store> */
9246                 *insn++ = BPF_JMP_A(2);
9247                 /* <clear>: mono_delivery_time */
9248                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9249                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9250         }
9251 #endif
9252
9253         /* <store>: skb->tstamp = tstamp */
9254         *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9255                               offsetof(struct sk_buff, tstamp));
9256         return insn;
9257 }
9258
9259 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9260                                   const struct bpf_insn *si,
9261                                   struct bpf_insn *insn_buf,
9262                                   struct bpf_prog *prog, u32 *target_size)
9263 {
9264         struct bpf_insn *insn = insn_buf;
9265         int off;
9266
9267         switch (si->off) {
9268         case offsetof(struct __sk_buff, len):
9269                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9270                                       bpf_target_off(struct sk_buff, len, 4,
9271                                                      target_size));
9272                 break;
9273
9274         case offsetof(struct __sk_buff, protocol):
9275                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9276                                       bpf_target_off(struct sk_buff, protocol, 2,
9277                                                      target_size));
9278                 break;
9279
9280         case offsetof(struct __sk_buff, vlan_proto):
9281                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9282                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9283                                                      target_size));
9284                 break;
9285
9286         case offsetof(struct __sk_buff, priority):
9287                 if (type == BPF_WRITE)
9288                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9289                                               bpf_target_off(struct sk_buff, priority, 4,
9290                                                              target_size));
9291                 else
9292                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9293                                               bpf_target_off(struct sk_buff, priority, 4,
9294                                                              target_size));
9295                 break;
9296
9297         case offsetof(struct __sk_buff, ingress_ifindex):
9298                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9299                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9300                                                      target_size));
9301                 break;
9302
9303         case offsetof(struct __sk_buff, ifindex):
9304                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9305                                       si->dst_reg, si->src_reg,
9306                                       offsetof(struct sk_buff, dev));
9307                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9308                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9309                                       bpf_target_off(struct net_device, ifindex, 4,
9310                                                      target_size));
9311                 break;
9312
9313         case offsetof(struct __sk_buff, hash):
9314                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9315                                       bpf_target_off(struct sk_buff, hash, 4,
9316                                                      target_size));
9317                 break;
9318
9319         case offsetof(struct __sk_buff, mark):
9320                 if (type == BPF_WRITE)
9321                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9322                                               bpf_target_off(struct sk_buff, mark, 4,
9323                                                              target_size));
9324                 else
9325                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9326                                               bpf_target_off(struct sk_buff, mark, 4,
9327                                                              target_size));
9328                 break;
9329
9330         case offsetof(struct __sk_buff, pkt_type):
9331                 *target_size = 1;
9332                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9333                                       PKT_TYPE_OFFSET);
9334                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9335 #ifdef __BIG_ENDIAN_BITFIELD
9336                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9337 #endif
9338                 break;
9339
9340         case offsetof(struct __sk_buff, queue_mapping):
9341                 if (type == BPF_WRITE) {
9342                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9343                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9344                                               bpf_target_off(struct sk_buff,
9345                                                              queue_mapping,
9346                                                              2, target_size));
9347                 } else {
9348                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9349                                               bpf_target_off(struct sk_buff,
9350                                                              queue_mapping,
9351                                                              2, target_size));
9352                 }
9353                 break;
9354
9355         case offsetof(struct __sk_buff, vlan_present):
9356                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9357                                       bpf_target_off(struct sk_buff,
9358                                                      vlan_all, 4, target_size));
9359                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9360                 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1);
9361                 break;
9362
9363         case offsetof(struct __sk_buff, vlan_tci):
9364                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9365                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9366                                                      target_size));
9367                 break;
9368
9369         case offsetof(struct __sk_buff, cb[0]) ...
9370              offsetofend(struct __sk_buff, cb[4]) - 1:
9371                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9372                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9373                               offsetof(struct qdisc_skb_cb, data)) %
9374                              sizeof(__u64));
9375
9376                 prog->cb_access = 1;
9377                 off  = si->off;
9378                 off -= offsetof(struct __sk_buff, cb[0]);
9379                 off += offsetof(struct sk_buff, cb);
9380                 off += offsetof(struct qdisc_skb_cb, data);
9381                 if (type == BPF_WRITE)
9382                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9383                                               si->src_reg, off);
9384                 else
9385                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9386                                               si->src_reg, off);
9387                 break;
9388
9389         case offsetof(struct __sk_buff, tc_classid):
9390                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9391
9392                 off  = si->off;
9393                 off -= offsetof(struct __sk_buff, tc_classid);
9394                 off += offsetof(struct sk_buff, cb);
9395                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9396                 *target_size = 2;
9397                 if (type == BPF_WRITE)
9398                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9399                                               si->src_reg, off);
9400                 else
9401                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9402                                               si->src_reg, off);
9403                 break;
9404
9405         case offsetof(struct __sk_buff, data):
9406                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9407                                       si->dst_reg, si->src_reg,
9408                                       offsetof(struct sk_buff, data));
9409                 break;
9410
9411         case offsetof(struct __sk_buff, data_meta):
9412                 off  = si->off;
9413                 off -= offsetof(struct __sk_buff, data_meta);
9414                 off += offsetof(struct sk_buff, cb);
9415                 off += offsetof(struct bpf_skb_data_end, data_meta);
9416                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9417                                       si->src_reg, off);
9418                 break;
9419
9420         case offsetof(struct __sk_buff, data_end):
9421                 off  = si->off;
9422                 off -= offsetof(struct __sk_buff, data_end);
9423                 off += offsetof(struct sk_buff, cb);
9424                 off += offsetof(struct bpf_skb_data_end, data_end);
9425                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9426                                       si->src_reg, off);
9427                 break;
9428
9429         case offsetof(struct __sk_buff, tc_index):
9430 #ifdef CONFIG_NET_SCHED
9431                 if (type == BPF_WRITE)
9432                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9433                                               bpf_target_off(struct sk_buff, tc_index, 2,
9434                                                              target_size));
9435                 else
9436                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9437                                               bpf_target_off(struct sk_buff, tc_index, 2,
9438                                                              target_size));
9439 #else
9440                 *target_size = 2;
9441                 if (type == BPF_WRITE)
9442                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9443                 else
9444                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9445 #endif
9446                 break;
9447
9448         case offsetof(struct __sk_buff, napi_id):
9449 #if defined(CONFIG_NET_RX_BUSY_POLL)
9450                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9451                                       bpf_target_off(struct sk_buff, napi_id, 4,
9452                                                      target_size));
9453                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9454                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9455 #else
9456                 *target_size = 4;
9457                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9458 #endif
9459                 break;
9460         case offsetof(struct __sk_buff, family):
9461                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9462
9463                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9464                                       si->dst_reg, si->src_reg,
9465                                       offsetof(struct sk_buff, sk));
9466                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9467                                       bpf_target_off(struct sock_common,
9468                                                      skc_family,
9469                                                      2, target_size));
9470                 break;
9471         case offsetof(struct __sk_buff, remote_ip4):
9472                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9473
9474                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9475                                       si->dst_reg, si->src_reg,
9476                                       offsetof(struct sk_buff, sk));
9477                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9478                                       bpf_target_off(struct sock_common,
9479                                                      skc_daddr,
9480                                                      4, target_size));
9481                 break;
9482         case offsetof(struct __sk_buff, local_ip4):
9483                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9484                                           skc_rcv_saddr) != 4);
9485
9486                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9487                                       si->dst_reg, si->src_reg,
9488                                       offsetof(struct sk_buff, sk));
9489                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9490                                       bpf_target_off(struct sock_common,
9491                                                      skc_rcv_saddr,
9492                                                      4, target_size));
9493                 break;
9494         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9495              offsetof(struct __sk_buff, remote_ip6[3]):
9496 #if IS_ENABLED(CONFIG_IPV6)
9497                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9498                                           skc_v6_daddr.s6_addr32[0]) != 4);
9499
9500                 off = si->off;
9501                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
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_W, si->dst_reg, si->dst_reg,
9507                                       offsetof(struct sock_common,
9508                                                skc_v6_daddr.s6_addr32[0]) +
9509                                       off);
9510 #else
9511                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9512 #endif
9513                 break;
9514         case offsetof(struct __sk_buff, local_ip6[0]) ...
9515              offsetof(struct __sk_buff, local_ip6[3]):
9516 #if IS_ENABLED(CONFIG_IPV6)
9517                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9518                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9519
9520                 off = si->off;
9521                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9522
9523                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9524                                       si->dst_reg, si->src_reg,
9525                                       offsetof(struct sk_buff, sk));
9526                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9527                                       offsetof(struct sock_common,
9528                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9529                                       off);
9530 #else
9531                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9532 #endif
9533                 break;
9534
9535         case offsetof(struct __sk_buff, remote_port):
9536                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9537
9538                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9539                                       si->dst_reg, si->src_reg,
9540                                       offsetof(struct sk_buff, sk));
9541                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9542                                       bpf_target_off(struct sock_common,
9543                                                      skc_dport,
9544                                                      2, target_size));
9545 #ifndef __BIG_ENDIAN_BITFIELD
9546                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9547 #endif
9548                 break;
9549
9550         case offsetof(struct __sk_buff, local_port):
9551                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9552
9553                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9554                                       si->dst_reg, si->src_reg,
9555                                       offsetof(struct sk_buff, sk));
9556                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9557                                       bpf_target_off(struct sock_common,
9558                                                      skc_num, 2, target_size));
9559                 break;
9560
9561         case offsetof(struct __sk_buff, tstamp):
9562                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9563
9564                 if (type == BPF_WRITE)
9565                         insn = bpf_convert_tstamp_write(prog, si, insn);
9566                 else
9567                         insn = bpf_convert_tstamp_read(prog, si, insn);
9568                 break;
9569
9570         case offsetof(struct __sk_buff, tstamp_type):
9571                 insn = bpf_convert_tstamp_type_read(si, insn);
9572                 break;
9573
9574         case offsetof(struct __sk_buff, gso_segs):
9575                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9577                                       si->dst_reg, si->dst_reg,
9578                                       bpf_target_off(struct skb_shared_info,
9579                                                      gso_segs, 2,
9580                                                      target_size));
9581                 break;
9582         case offsetof(struct __sk_buff, gso_size):
9583                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9585                                       si->dst_reg, si->dst_reg,
9586                                       bpf_target_off(struct skb_shared_info,
9587                                                      gso_size, 2,
9588                                                      target_size));
9589                 break;
9590         case offsetof(struct __sk_buff, wire_len):
9591                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9592
9593                 off = si->off;
9594                 off -= offsetof(struct __sk_buff, wire_len);
9595                 off += offsetof(struct sk_buff, cb);
9596                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9597                 *target_size = 4;
9598                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9599                 break;
9600
9601         case offsetof(struct __sk_buff, sk):
9602                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9603                                       si->dst_reg, si->src_reg,
9604                                       offsetof(struct sk_buff, sk));
9605                 break;
9606         case offsetof(struct __sk_buff, hwtstamp):
9607                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9608                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9609
9610                 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn);
9611                 *insn++ = BPF_LDX_MEM(BPF_DW,
9612                                       si->dst_reg, si->dst_reg,
9613                                       bpf_target_off(struct skb_shared_info,
9614                                                      hwtstamps, 8,
9615                                                      target_size));
9616                 break;
9617         }
9618
9619         return insn - insn_buf;
9620 }
9621
9622 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9623                                 const struct bpf_insn *si,
9624                                 struct bpf_insn *insn_buf,
9625                                 struct bpf_prog *prog, u32 *target_size)
9626 {
9627         struct bpf_insn *insn = insn_buf;
9628         int off;
9629
9630         switch (si->off) {
9631         case offsetof(struct bpf_sock, bound_dev_if):
9632                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9633
9634                 if (type == BPF_WRITE)
9635                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9636                                         offsetof(struct sock, sk_bound_dev_if));
9637                 else
9638                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9639                                       offsetof(struct sock, sk_bound_dev_if));
9640                 break;
9641
9642         case offsetof(struct bpf_sock, mark):
9643                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9644
9645                 if (type == BPF_WRITE)
9646                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9647                                         offsetof(struct sock, sk_mark));
9648                 else
9649                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9650                                       offsetof(struct sock, sk_mark));
9651                 break;
9652
9653         case offsetof(struct bpf_sock, priority):
9654                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9655
9656                 if (type == BPF_WRITE)
9657                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9658                                         offsetof(struct sock, sk_priority));
9659                 else
9660                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9661                                       offsetof(struct sock, sk_priority));
9662                 break;
9663
9664         case offsetof(struct bpf_sock, family):
9665                 *insn++ = BPF_LDX_MEM(
9666                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9667                         si->dst_reg, si->src_reg,
9668                         bpf_target_off(struct sock_common,
9669                                        skc_family,
9670                                        sizeof_field(struct sock_common,
9671                                                     skc_family),
9672                                        target_size));
9673                 break;
9674
9675         case offsetof(struct bpf_sock, type):
9676                 *insn++ = BPF_LDX_MEM(
9677                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9678                         si->dst_reg, si->src_reg,
9679                         bpf_target_off(struct sock, sk_type,
9680                                        sizeof_field(struct sock, sk_type),
9681                                        target_size));
9682                 break;
9683
9684         case offsetof(struct bpf_sock, protocol):
9685                 *insn++ = BPF_LDX_MEM(
9686                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9687                         si->dst_reg, si->src_reg,
9688                         bpf_target_off(struct sock, sk_protocol,
9689                                        sizeof_field(struct sock, sk_protocol),
9690                                        target_size));
9691                 break;
9692
9693         case offsetof(struct bpf_sock, src_ip4):
9694                 *insn++ = BPF_LDX_MEM(
9695                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9696                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9697                                        sizeof_field(struct sock_common,
9698                                                     skc_rcv_saddr),
9699                                        target_size));
9700                 break;
9701
9702         case offsetof(struct bpf_sock, dst_ip4):
9703                 *insn++ = BPF_LDX_MEM(
9704                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9705                         bpf_target_off(struct sock_common, skc_daddr,
9706                                        sizeof_field(struct sock_common,
9707                                                     skc_daddr),
9708                                        target_size));
9709                 break;
9710
9711         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9712 #if IS_ENABLED(CONFIG_IPV6)
9713                 off = si->off;
9714                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9715                 *insn++ = BPF_LDX_MEM(
9716                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9717                         bpf_target_off(
9718                                 struct sock_common,
9719                                 skc_v6_rcv_saddr.s6_addr32[0],
9720                                 sizeof_field(struct sock_common,
9721                                              skc_v6_rcv_saddr.s6_addr32[0]),
9722                                 target_size) + off);
9723 #else
9724                 (void)off;
9725                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9726 #endif
9727                 break;
9728
9729         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9730 #if IS_ENABLED(CONFIG_IPV6)
9731                 off = si->off;
9732                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9733                 *insn++ = BPF_LDX_MEM(
9734                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9735                         bpf_target_off(struct sock_common,
9736                                        skc_v6_daddr.s6_addr32[0],
9737                                        sizeof_field(struct sock_common,
9738                                                     skc_v6_daddr.s6_addr32[0]),
9739                                        target_size) + off);
9740 #else
9741                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9742                 *target_size = 4;
9743 #endif
9744                 break;
9745
9746         case offsetof(struct bpf_sock, src_port):
9747                 *insn++ = BPF_LDX_MEM(
9748                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9749                         si->dst_reg, si->src_reg,
9750                         bpf_target_off(struct sock_common, skc_num,
9751                                        sizeof_field(struct sock_common,
9752                                                     skc_num),
9753                                        target_size));
9754                 break;
9755
9756         case offsetof(struct bpf_sock, dst_port):
9757                 *insn++ = BPF_LDX_MEM(
9758                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9759                         si->dst_reg, si->src_reg,
9760                         bpf_target_off(struct sock_common, skc_dport,
9761                                        sizeof_field(struct sock_common,
9762                                                     skc_dport),
9763                                        target_size));
9764                 break;
9765
9766         case offsetof(struct bpf_sock, state):
9767                 *insn++ = BPF_LDX_MEM(
9768                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9769                         si->dst_reg, si->src_reg,
9770                         bpf_target_off(struct sock_common, skc_state,
9771                                        sizeof_field(struct sock_common,
9772                                                     skc_state),
9773                                        target_size));
9774                 break;
9775         case offsetof(struct bpf_sock, rx_queue_mapping):
9776 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9777                 *insn++ = BPF_LDX_MEM(
9778                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9779                         si->dst_reg, si->src_reg,
9780                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9781                                        sizeof_field(struct sock,
9782                                                     sk_rx_queue_mapping),
9783                                        target_size));
9784                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9785                                       1);
9786                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9787 #else
9788                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9789                 *target_size = 2;
9790 #endif
9791                 break;
9792         }
9793
9794         return insn - insn_buf;
9795 }
9796
9797 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9798                                          const struct bpf_insn *si,
9799                                          struct bpf_insn *insn_buf,
9800                                          struct bpf_prog *prog, u32 *target_size)
9801 {
9802         struct bpf_insn *insn = insn_buf;
9803
9804         switch (si->off) {
9805         case offsetof(struct __sk_buff, ifindex):
9806                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9807                                       si->dst_reg, si->src_reg,
9808                                       offsetof(struct sk_buff, dev));
9809                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9810                                       bpf_target_off(struct net_device, ifindex, 4,
9811                                                      target_size));
9812                 break;
9813         default:
9814                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9815                                               target_size);
9816         }
9817
9818         return insn - insn_buf;
9819 }
9820
9821 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9822                                   const struct bpf_insn *si,
9823                                   struct bpf_insn *insn_buf,
9824                                   struct bpf_prog *prog, u32 *target_size)
9825 {
9826         struct bpf_insn *insn = insn_buf;
9827
9828         switch (si->off) {
9829         case offsetof(struct xdp_md, data):
9830                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9831                                       si->dst_reg, si->src_reg,
9832                                       offsetof(struct xdp_buff, data));
9833                 break;
9834         case offsetof(struct xdp_md, data_meta):
9835                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9836                                       si->dst_reg, si->src_reg,
9837                                       offsetof(struct xdp_buff, data_meta));
9838                 break;
9839         case offsetof(struct xdp_md, data_end):
9840                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9841                                       si->dst_reg, si->src_reg,
9842                                       offsetof(struct xdp_buff, data_end));
9843                 break;
9844         case offsetof(struct xdp_md, ingress_ifindex):
9845                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9846                                       si->dst_reg, si->src_reg,
9847                                       offsetof(struct xdp_buff, rxq));
9848                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9849                                       si->dst_reg, si->dst_reg,
9850                                       offsetof(struct xdp_rxq_info, dev));
9851                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9852                                       offsetof(struct net_device, ifindex));
9853                 break;
9854         case offsetof(struct xdp_md, rx_queue_index):
9855                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9856                                       si->dst_reg, si->src_reg,
9857                                       offsetof(struct xdp_buff, rxq));
9858                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9859                                       offsetof(struct xdp_rxq_info,
9860                                                queue_index));
9861                 break;
9862         case offsetof(struct xdp_md, egress_ifindex):
9863                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9864                                       si->dst_reg, si->src_reg,
9865                                       offsetof(struct xdp_buff, txq));
9866                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9867                                       si->dst_reg, si->dst_reg,
9868                                       offsetof(struct xdp_txq_info, dev));
9869                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9870                                       offsetof(struct net_device, ifindex));
9871                 break;
9872         }
9873
9874         return insn - insn_buf;
9875 }
9876
9877 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9878  * context Structure, F is Field in context structure that contains a pointer
9879  * to Nested Structure of type NS that has the field NF.
9880  *
9881  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9882  * sure that SIZE is not greater than actual size of S.F.NF.
9883  *
9884  * If offset OFF is provided, the load happens from that offset relative to
9885  * offset of NF.
9886  */
9887 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9888         do {                                                                   \
9889                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9890                                       si->src_reg, offsetof(S, F));            \
9891                 *insn++ = BPF_LDX_MEM(                                         \
9892                         SIZE, si->dst_reg, si->dst_reg,                        \
9893                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9894                                        target_size)                            \
9895                                 + OFF);                                        \
9896         } while (0)
9897
9898 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9899         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9900                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9901
9902 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9903  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9904  *
9905  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9906  * "register" since two registers available in convert_ctx_access are not
9907  * enough: we can't override neither SRC, since it contains value to store, nor
9908  * DST since it contains pointer to context that may be used by later
9909  * instructions. But we need a temporary place to save pointer to nested
9910  * structure whose field we want to store to.
9911  */
9912 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9913         do {                                                                   \
9914                 int tmp_reg = BPF_REG_9;                                       \
9915                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9916                         --tmp_reg;                                             \
9917                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9918                         --tmp_reg;                                             \
9919                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9920                                       offsetof(S, TF));                        \
9921                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9922                                       si->dst_reg, offsetof(S, F));            \
9923                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9924                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9925                                        target_size)                            \
9926                                 + OFF);                                        \
9927                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9928                                       offsetof(S, TF));                        \
9929         } while (0)
9930
9931 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9932                                                       TF)                      \
9933         do {                                                                   \
9934                 if (type == BPF_WRITE) {                                       \
9935                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9936                                                          OFF, TF);             \
9937                 } else {                                                       \
9938                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9939                                 S, NS, F, NF, SIZE, OFF);  \
9940                 }                                                              \
9941         } while (0)
9942
9943 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9944         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9945                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9946
9947 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9948                                         const struct bpf_insn *si,
9949                                         struct bpf_insn *insn_buf,
9950                                         struct bpf_prog *prog, u32 *target_size)
9951 {
9952         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9953         struct bpf_insn *insn = insn_buf;
9954
9955         switch (si->off) {
9956         case offsetof(struct bpf_sock_addr, user_family):
9957                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9958                                             struct sockaddr, uaddr, sa_family);
9959                 break;
9960
9961         case offsetof(struct bpf_sock_addr, user_ip4):
9962                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9963                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9964                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9965                 break;
9966
9967         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9968                 off = si->off;
9969                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9970                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9971                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9972                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9973                         tmp_reg);
9974                 break;
9975
9976         case offsetof(struct bpf_sock_addr, user_port):
9977                 /* To get port we need to know sa_family first and then treat
9978                  * sockaddr as either sockaddr_in or sockaddr_in6.
9979                  * Though we can simplify since port field has same offset and
9980                  * size in both structures.
9981                  * Here we check this invariant and use just one of the
9982                  * structures if it's true.
9983                  */
9984                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9985                              offsetof(struct sockaddr_in6, sin6_port));
9986                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9987                              sizeof_field(struct sockaddr_in6, sin6_port));
9988                 /* Account for sin6_port being smaller than user_port. */
9989                 port_size = min(port_size, BPF_LDST_BYTES(si));
9990                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9991                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9992                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9993                 break;
9994
9995         case offsetof(struct bpf_sock_addr, family):
9996                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9997                                             struct sock, sk, sk_family);
9998                 break;
9999
10000         case offsetof(struct bpf_sock_addr, type):
10001                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10002                                             struct sock, sk, sk_type);
10003                 break;
10004
10005         case offsetof(struct bpf_sock_addr, protocol):
10006                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
10007                                             struct sock, sk, sk_protocol);
10008                 break;
10009
10010         case offsetof(struct bpf_sock_addr, msg_src_ip4):
10011                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
10012                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10013                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
10014                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
10015                 break;
10016
10017         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
10018                                 msg_src_ip6[3]):
10019                 off = si->off;
10020                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
10021                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
10022                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
10023                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
10024                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
10025                 break;
10026         case offsetof(struct bpf_sock_addr, sk):
10027                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
10028                                       si->dst_reg, si->src_reg,
10029                                       offsetof(struct bpf_sock_addr_kern, sk));
10030                 break;
10031         }
10032
10033         return insn - insn_buf;
10034 }
10035
10036 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
10037                                        const struct bpf_insn *si,
10038                                        struct bpf_insn *insn_buf,
10039                                        struct bpf_prog *prog,
10040                                        u32 *target_size)
10041 {
10042         struct bpf_insn *insn = insn_buf;
10043         int off;
10044
10045 /* Helper macro for adding read access to tcp_sock or sock fields. */
10046 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10047         do {                                                                  \
10048                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
10049                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10050                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10051                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10052                         reg--;                                                \
10053                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10054                         reg--;                                                \
10055                 if (si->dst_reg == si->src_reg) {                             \
10056                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10057                                           offsetof(struct bpf_sock_ops_kern,  \
10058                                           temp));                             \
10059                         fullsock_reg = reg;                                   \
10060                         jmp += 2;                                             \
10061                 }                                                             \
10062                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10063                                                 struct bpf_sock_ops_kern,     \
10064                                                 is_fullsock),                 \
10065                                       fullsock_reg, si->src_reg,              \
10066                                       offsetof(struct bpf_sock_ops_kern,      \
10067                                                is_fullsock));                 \
10068                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10069                 if (si->dst_reg == si->src_reg)                               \
10070                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10071                                       offsetof(struct bpf_sock_ops_kern,      \
10072                                       temp));                                 \
10073                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10074                                                 struct bpf_sock_ops_kern, sk),\
10075                                       si->dst_reg, si->src_reg,               \
10076                                       offsetof(struct bpf_sock_ops_kern, sk));\
10077                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
10078                                                        OBJ_FIELD),            \
10079                                       si->dst_reg, si->dst_reg,               \
10080                                       offsetof(OBJ, OBJ_FIELD));              \
10081                 if (si->dst_reg == si->src_reg) {                             \
10082                         *insn++ = BPF_JMP_A(1);                               \
10083                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10084                                       offsetof(struct bpf_sock_ops_kern,      \
10085                                       temp));                                 \
10086                 }                                                             \
10087         } while (0)
10088
10089 #define SOCK_OPS_GET_SK()                                                             \
10090         do {                                                                  \
10091                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
10092                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10093                         reg--;                                                \
10094                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10095                         reg--;                                                \
10096                 if (si->dst_reg == si->src_reg) {                             \
10097                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
10098                                           offsetof(struct bpf_sock_ops_kern,  \
10099                                           temp));                             \
10100                         fullsock_reg = reg;                                   \
10101                         jmp += 2;                                             \
10102                 }                                                             \
10103                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10104                                                 struct bpf_sock_ops_kern,     \
10105                                                 is_fullsock),                 \
10106                                       fullsock_reg, si->src_reg,              \
10107                                       offsetof(struct bpf_sock_ops_kern,      \
10108                                                is_fullsock));                 \
10109                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
10110                 if (si->dst_reg == si->src_reg)                               \
10111                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10112                                       offsetof(struct bpf_sock_ops_kern,      \
10113                                       temp));                                 \
10114                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10115                                                 struct bpf_sock_ops_kern, sk),\
10116                                       si->dst_reg, si->src_reg,               \
10117                                       offsetof(struct bpf_sock_ops_kern, sk));\
10118                 if (si->dst_reg == si->src_reg) {                             \
10119                         *insn++ = BPF_JMP_A(1);                               \
10120                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
10121                                       offsetof(struct bpf_sock_ops_kern,      \
10122                                       temp));                                 \
10123                 }                                                             \
10124         } while (0)
10125
10126 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
10127                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
10128
10129 /* Helper macro for adding write access to tcp_sock or sock fields.
10130  * The macro is called with two registers, dst_reg which contains a pointer
10131  * to ctx (context) and src_reg which contains the value that should be
10132  * stored. However, we need an additional register since we cannot overwrite
10133  * dst_reg because it may be used later in the program.
10134  * Instead we "borrow" one of the other register. We first save its value
10135  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
10136  * it at the end of the macro.
10137  */
10138 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
10139         do {                                                                  \
10140                 int reg = BPF_REG_9;                                          \
10141                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
10142                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
10143                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10144                         reg--;                                                \
10145                 if (si->dst_reg == reg || si->src_reg == reg)                 \
10146                         reg--;                                                \
10147                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
10148                                       offsetof(struct bpf_sock_ops_kern,      \
10149                                                temp));                        \
10150                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10151                                                 struct bpf_sock_ops_kern,     \
10152                                                 is_fullsock),                 \
10153                                       reg, si->dst_reg,                       \
10154                                       offsetof(struct bpf_sock_ops_kern,      \
10155                                                is_fullsock));                 \
10156                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10157                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10158                                                 struct bpf_sock_ops_kern, sk),\
10159                                       reg, si->dst_reg,                       \
10160                                       offsetof(struct bpf_sock_ops_kern, sk));\
10161                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
10162                                       reg, si->src_reg,                       \
10163                                       offsetof(OBJ, OBJ_FIELD));              \
10164                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10165                                       offsetof(struct bpf_sock_ops_kern,      \
10166                                                temp));                        \
10167         } while (0)
10168
10169 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10170         do {                                                                  \
10171                 if (TYPE == BPF_WRITE)                                        \
10172                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10173                 else                                                          \
10174                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10175         } while (0)
10176
10177         switch (si->off) {
10178         case offsetof(struct bpf_sock_ops, op):
10179                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10180                                                        op),
10181                                       si->dst_reg, si->src_reg,
10182                                       offsetof(struct bpf_sock_ops_kern, op));
10183                 break;
10184
10185         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10186              offsetof(struct bpf_sock_ops, replylong[3]):
10187                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10188                              sizeof_field(struct bpf_sock_ops_kern, reply));
10189                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10190                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10191                 off = si->off;
10192                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10193                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10194                 if (type == BPF_WRITE)
10195                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10196                                               off);
10197                 else
10198                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10199                                               off);
10200                 break;
10201
10202         case offsetof(struct bpf_sock_ops, family):
10203                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10204
10205                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10206                                               struct bpf_sock_ops_kern, sk),
10207                                       si->dst_reg, si->src_reg,
10208                                       offsetof(struct bpf_sock_ops_kern, sk));
10209                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10210                                       offsetof(struct sock_common, skc_family));
10211                 break;
10212
10213         case offsetof(struct bpf_sock_ops, remote_ip4):
10214                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10215
10216                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10217                                                 struct bpf_sock_ops_kern, sk),
10218                                       si->dst_reg, si->src_reg,
10219                                       offsetof(struct bpf_sock_ops_kern, sk));
10220                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10221                                       offsetof(struct sock_common, skc_daddr));
10222                 break;
10223
10224         case offsetof(struct bpf_sock_ops, local_ip4):
10225                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10226                                           skc_rcv_saddr) != 4);
10227
10228                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10229                                               struct bpf_sock_ops_kern, sk),
10230                                       si->dst_reg, si->src_reg,
10231                                       offsetof(struct bpf_sock_ops_kern, sk));
10232                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10233                                       offsetof(struct sock_common,
10234                                                skc_rcv_saddr));
10235                 break;
10236
10237         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10238              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10239 #if IS_ENABLED(CONFIG_IPV6)
10240                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10241                                           skc_v6_daddr.s6_addr32[0]) != 4);
10242
10243                 off = si->off;
10244                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10245                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10246                                                 struct bpf_sock_ops_kern, sk),
10247                                       si->dst_reg, si->src_reg,
10248                                       offsetof(struct bpf_sock_ops_kern, sk));
10249                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10250                                       offsetof(struct sock_common,
10251                                                skc_v6_daddr.s6_addr32[0]) +
10252                                       off);
10253 #else
10254                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10255 #endif
10256                 break;
10257
10258         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10259              offsetof(struct bpf_sock_ops, local_ip6[3]):
10260 #if IS_ENABLED(CONFIG_IPV6)
10261                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10262                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10263
10264                 off = si->off;
10265                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10266                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10267                                                 struct bpf_sock_ops_kern, sk),
10268                                       si->dst_reg, si->src_reg,
10269                                       offsetof(struct bpf_sock_ops_kern, sk));
10270                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10271                                       offsetof(struct sock_common,
10272                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10273                                       off);
10274 #else
10275                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10276 #endif
10277                 break;
10278
10279         case offsetof(struct bpf_sock_ops, remote_port):
10280                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10281
10282                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10283                                                 struct bpf_sock_ops_kern, sk),
10284                                       si->dst_reg, si->src_reg,
10285                                       offsetof(struct bpf_sock_ops_kern, sk));
10286                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10287                                       offsetof(struct sock_common, skc_dport));
10288 #ifndef __BIG_ENDIAN_BITFIELD
10289                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10290 #endif
10291                 break;
10292
10293         case offsetof(struct bpf_sock_ops, local_port):
10294                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10295
10296                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10297                                                 struct bpf_sock_ops_kern, sk),
10298                                       si->dst_reg, si->src_reg,
10299                                       offsetof(struct bpf_sock_ops_kern, sk));
10300                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10301                                       offsetof(struct sock_common, skc_num));
10302                 break;
10303
10304         case offsetof(struct bpf_sock_ops, is_fullsock):
10305                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10306                                                 struct bpf_sock_ops_kern,
10307                                                 is_fullsock),
10308                                       si->dst_reg, si->src_reg,
10309                                       offsetof(struct bpf_sock_ops_kern,
10310                                                is_fullsock));
10311                 break;
10312
10313         case offsetof(struct bpf_sock_ops, state):
10314                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10315
10316                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10317                                                 struct bpf_sock_ops_kern, sk),
10318                                       si->dst_reg, si->src_reg,
10319                                       offsetof(struct bpf_sock_ops_kern, sk));
10320                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10321                                       offsetof(struct sock_common, skc_state));
10322                 break;
10323
10324         case offsetof(struct bpf_sock_ops, rtt_min):
10325                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10326                              sizeof(struct minmax));
10327                 BUILD_BUG_ON(sizeof(struct minmax) <
10328                              sizeof(struct minmax_sample));
10329
10330                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10331                                                 struct bpf_sock_ops_kern, sk),
10332                                       si->dst_reg, si->src_reg,
10333                                       offsetof(struct bpf_sock_ops_kern, sk));
10334                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10335                                       offsetof(struct tcp_sock, rtt_min) +
10336                                       sizeof_field(struct minmax_sample, t));
10337                 break;
10338
10339         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10340                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10341                                    struct tcp_sock);
10342                 break;
10343
10344         case offsetof(struct bpf_sock_ops, sk_txhash):
10345                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10346                                           struct sock, type);
10347                 break;
10348         case offsetof(struct bpf_sock_ops, snd_cwnd):
10349                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10350                 break;
10351         case offsetof(struct bpf_sock_ops, srtt_us):
10352                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10353                 break;
10354         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10355                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10356                 break;
10357         case offsetof(struct bpf_sock_ops, rcv_nxt):
10358                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10359                 break;
10360         case offsetof(struct bpf_sock_ops, snd_nxt):
10361                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10362                 break;
10363         case offsetof(struct bpf_sock_ops, snd_una):
10364                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10365                 break;
10366         case offsetof(struct bpf_sock_ops, mss_cache):
10367                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10368                 break;
10369         case offsetof(struct bpf_sock_ops, ecn_flags):
10370                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10371                 break;
10372         case offsetof(struct bpf_sock_ops, rate_delivered):
10373                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10374                 break;
10375         case offsetof(struct bpf_sock_ops, rate_interval_us):
10376                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10377                 break;
10378         case offsetof(struct bpf_sock_ops, packets_out):
10379                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10380                 break;
10381         case offsetof(struct bpf_sock_ops, retrans_out):
10382                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10383                 break;
10384         case offsetof(struct bpf_sock_ops, total_retrans):
10385                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10386                 break;
10387         case offsetof(struct bpf_sock_ops, segs_in):
10388                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10389                 break;
10390         case offsetof(struct bpf_sock_ops, data_segs_in):
10391                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10392                 break;
10393         case offsetof(struct bpf_sock_ops, segs_out):
10394                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10395                 break;
10396         case offsetof(struct bpf_sock_ops, data_segs_out):
10397                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10398                 break;
10399         case offsetof(struct bpf_sock_ops, lost_out):
10400                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10401                 break;
10402         case offsetof(struct bpf_sock_ops, sacked_out):
10403                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10404                 break;
10405         case offsetof(struct bpf_sock_ops, bytes_received):
10406                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10407                 break;
10408         case offsetof(struct bpf_sock_ops, bytes_acked):
10409                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10410                 break;
10411         case offsetof(struct bpf_sock_ops, sk):
10412                 SOCK_OPS_GET_SK();
10413                 break;
10414         case offsetof(struct bpf_sock_ops, skb_data_end):
10415                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10416                                                        skb_data_end),
10417                                       si->dst_reg, si->src_reg,
10418                                       offsetof(struct bpf_sock_ops_kern,
10419                                                skb_data_end));
10420                 break;
10421         case offsetof(struct bpf_sock_ops, skb_data):
10422                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10423                                                        skb),
10424                                       si->dst_reg, si->src_reg,
10425                                       offsetof(struct bpf_sock_ops_kern,
10426                                                skb));
10427                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10428                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10429                                       si->dst_reg, si->dst_reg,
10430                                       offsetof(struct sk_buff, data));
10431                 break;
10432         case offsetof(struct bpf_sock_ops, skb_len):
10433                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10434                                                        skb),
10435                                       si->dst_reg, si->src_reg,
10436                                       offsetof(struct bpf_sock_ops_kern,
10437                                                skb));
10438                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10439                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10440                                       si->dst_reg, si->dst_reg,
10441                                       offsetof(struct sk_buff, len));
10442                 break;
10443         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10444                 off = offsetof(struct sk_buff, cb);
10445                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10446                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10447                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10448                                                        skb),
10449                                       si->dst_reg, si->src_reg,
10450                                       offsetof(struct bpf_sock_ops_kern,
10451                                                skb));
10452                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10453                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10454                                                        tcp_flags),
10455                                       si->dst_reg, si->dst_reg, off);
10456                 break;
10457         case offsetof(struct bpf_sock_ops, skb_hwtstamp): {
10458                 struct bpf_insn *jmp_on_null_skb;
10459
10460                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10461                                                        skb),
10462                                       si->dst_reg, si->src_reg,
10463                                       offsetof(struct bpf_sock_ops_kern,
10464                                                skb));
10465                 /* Reserve one insn to test skb == NULL */
10466                 jmp_on_null_skb = insn++;
10467                 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn);
10468                 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
10469                                       bpf_target_off(struct skb_shared_info,
10470                                                      hwtstamps, 8,
10471                                                      target_size));
10472                 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0,
10473                                                insn - jmp_on_null_skb - 1);
10474                 break;
10475         }
10476         }
10477         return insn - insn_buf;
10478 }
10479
10480 /* data_end = skb->data + skb_headlen() */
10481 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10482                                                     struct bpf_insn *insn)
10483 {
10484         int reg;
10485         int temp_reg_off = offsetof(struct sk_buff, cb) +
10486                            offsetof(struct sk_skb_cb, temp_reg);
10487
10488         if (si->src_reg == si->dst_reg) {
10489                 /* We need an extra register, choose and save a register. */
10490                 reg = BPF_REG_9;
10491                 if (si->src_reg == reg || si->dst_reg == reg)
10492                         reg--;
10493                 if (si->src_reg == reg || si->dst_reg == reg)
10494                         reg--;
10495                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10496         } else {
10497                 reg = si->dst_reg;
10498         }
10499
10500         /* reg = skb->data */
10501         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10502                               reg, si->src_reg,
10503                               offsetof(struct sk_buff, data));
10504         /* AX = skb->len */
10505         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10506                               BPF_REG_AX, si->src_reg,
10507                               offsetof(struct sk_buff, len));
10508         /* reg = skb->data + skb->len */
10509         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10510         /* AX = skb->data_len */
10511         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10512                               BPF_REG_AX, si->src_reg,
10513                               offsetof(struct sk_buff, data_len));
10514
10515         /* reg = skb->data + skb->len - skb->data_len */
10516         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10517
10518         if (si->src_reg == si->dst_reg) {
10519                 /* Restore the saved register */
10520                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10521                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10522                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10523         }
10524
10525         return insn;
10526 }
10527
10528 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10529                                      const struct bpf_insn *si,
10530                                      struct bpf_insn *insn_buf,
10531                                      struct bpf_prog *prog, u32 *target_size)
10532 {
10533         struct bpf_insn *insn = insn_buf;
10534         int off;
10535
10536         switch (si->off) {
10537         case offsetof(struct __sk_buff, data_end):
10538                 insn = bpf_convert_data_end_access(si, insn);
10539                 break;
10540         case offsetof(struct __sk_buff, cb[0]) ...
10541              offsetofend(struct __sk_buff, cb[4]) - 1:
10542                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10543                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10544                               offsetof(struct sk_skb_cb, data)) %
10545                              sizeof(__u64));
10546
10547                 prog->cb_access = 1;
10548                 off  = si->off;
10549                 off -= offsetof(struct __sk_buff, cb[0]);
10550                 off += offsetof(struct sk_buff, cb);
10551                 off += offsetof(struct sk_skb_cb, data);
10552                 if (type == BPF_WRITE)
10553                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10554                                               si->src_reg, off);
10555                 else
10556                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10557                                               si->src_reg, off);
10558                 break;
10559
10560
10561         default:
10562                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10563                                               target_size);
10564         }
10565
10566         return insn - insn_buf;
10567 }
10568
10569 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
10570                                      const struct bpf_insn *si,
10571                                      struct bpf_insn *insn_buf,
10572                                      struct bpf_prog *prog, u32 *target_size)
10573 {
10574         struct bpf_insn *insn = insn_buf;
10575 #if IS_ENABLED(CONFIG_IPV6)
10576         int off;
10577 #endif
10578
10579         /* convert ctx uses the fact sg element is first in struct */
10580         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10581
10582         switch (si->off) {
10583         case offsetof(struct sk_msg_md, data):
10584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10585                                       si->dst_reg, si->src_reg,
10586                                       offsetof(struct sk_msg, data));
10587                 break;
10588         case offsetof(struct sk_msg_md, data_end):
10589                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10590                                       si->dst_reg, si->src_reg,
10591                                       offsetof(struct sk_msg, data_end));
10592                 break;
10593         case offsetof(struct sk_msg_md, family):
10594                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10595
10596                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10597                                               struct sk_msg, sk),
10598                                       si->dst_reg, si->src_reg,
10599                                       offsetof(struct sk_msg, sk));
10600                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10601                                       offsetof(struct sock_common, skc_family));
10602                 break;
10603
10604         case offsetof(struct sk_msg_md, remote_ip4):
10605                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10606
10607                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10608                                                 struct sk_msg, sk),
10609                                       si->dst_reg, si->src_reg,
10610                                       offsetof(struct sk_msg, sk));
10611                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10612                                       offsetof(struct sock_common, skc_daddr));
10613                 break;
10614
10615         case offsetof(struct sk_msg_md, local_ip4):
10616                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10617                                           skc_rcv_saddr) != 4);
10618
10619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10620                                               struct sk_msg, sk),
10621                                       si->dst_reg, si->src_reg,
10622                                       offsetof(struct sk_msg, sk));
10623                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10624                                       offsetof(struct sock_common,
10625                                                skc_rcv_saddr));
10626                 break;
10627
10628         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10629              offsetof(struct sk_msg_md, remote_ip6[3]):
10630 #if IS_ENABLED(CONFIG_IPV6)
10631                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10632                                           skc_v6_daddr.s6_addr32[0]) != 4);
10633
10634                 off = si->off;
10635                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
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_W, si->dst_reg, si->dst_reg,
10641                                       offsetof(struct sock_common,
10642                                                skc_v6_daddr.s6_addr32[0]) +
10643                                       off);
10644 #else
10645                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10646 #endif
10647                 break;
10648
10649         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10650              offsetof(struct sk_msg_md, local_ip6[3]):
10651 #if IS_ENABLED(CONFIG_IPV6)
10652                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10653                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10654
10655                 off = si->off;
10656                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10657                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10658                                                 struct sk_msg, sk),
10659                                       si->dst_reg, si->src_reg,
10660                                       offsetof(struct sk_msg, sk));
10661                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10662                                       offsetof(struct sock_common,
10663                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10664                                       off);
10665 #else
10666                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10667 #endif
10668                 break;
10669
10670         case offsetof(struct sk_msg_md, remote_port):
10671                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10672
10673                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10674                                                 struct sk_msg, sk),
10675                                       si->dst_reg, si->src_reg,
10676                                       offsetof(struct sk_msg, sk));
10677                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10678                                       offsetof(struct sock_common, skc_dport));
10679 #ifndef __BIG_ENDIAN_BITFIELD
10680                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10681 #endif
10682                 break;
10683
10684         case offsetof(struct sk_msg_md, local_port):
10685                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10686
10687                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10688                                                 struct sk_msg, sk),
10689                                       si->dst_reg, si->src_reg,
10690                                       offsetof(struct sk_msg, sk));
10691                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10692                                       offsetof(struct sock_common, skc_num));
10693                 break;
10694
10695         case offsetof(struct sk_msg_md, size):
10696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10697                                       si->dst_reg, si->src_reg,
10698                                       offsetof(struct sk_msg_sg, size));
10699                 break;
10700
10701         case offsetof(struct sk_msg_md, sk):
10702                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10703                                       si->dst_reg, si->src_reg,
10704                                       offsetof(struct sk_msg, sk));
10705                 break;
10706         }
10707
10708         return insn - insn_buf;
10709 }
10710
10711 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10712         .get_func_proto         = sk_filter_func_proto,
10713         .is_valid_access        = sk_filter_is_valid_access,
10714         .convert_ctx_access     = bpf_convert_ctx_access,
10715         .gen_ld_abs             = bpf_gen_ld_abs,
10716 };
10717
10718 const struct bpf_prog_ops sk_filter_prog_ops = {
10719         .test_run               = bpf_prog_test_run_skb,
10720 };
10721
10722 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10723         .get_func_proto         = tc_cls_act_func_proto,
10724         .is_valid_access        = tc_cls_act_is_valid_access,
10725         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10726         .gen_prologue           = tc_cls_act_prologue,
10727         .gen_ld_abs             = bpf_gen_ld_abs,
10728         .btf_struct_access      = tc_cls_act_btf_struct_access,
10729 };
10730
10731 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10732         .test_run               = bpf_prog_test_run_skb,
10733 };
10734
10735 const struct bpf_verifier_ops xdp_verifier_ops = {
10736         .get_func_proto         = xdp_func_proto,
10737         .is_valid_access        = xdp_is_valid_access,
10738         .convert_ctx_access     = xdp_convert_ctx_access,
10739         .gen_prologue           = bpf_noop_prologue,
10740         .btf_struct_access      = xdp_btf_struct_access,
10741 };
10742
10743 const struct bpf_prog_ops xdp_prog_ops = {
10744         .test_run               = bpf_prog_test_run_xdp,
10745 };
10746
10747 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10748         .get_func_proto         = cg_skb_func_proto,
10749         .is_valid_access        = cg_skb_is_valid_access,
10750         .convert_ctx_access     = bpf_convert_ctx_access,
10751 };
10752
10753 const struct bpf_prog_ops cg_skb_prog_ops = {
10754         .test_run               = bpf_prog_test_run_skb,
10755 };
10756
10757 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10758         .get_func_proto         = lwt_in_func_proto,
10759         .is_valid_access        = lwt_is_valid_access,
10760         .convert_ctx_access     = bpf_convert_ctx_access,
10761 };
10762
10763 const struct bpf_prog_ops lwt_in_prog_ops = {
10764         .test_run               = bpf_prog_test_run_skb,
10765 };
10766
10767 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10768         .get_func_proto         = lwt_out_func_proto,
10769         .is_valid_access        = lwt_is_valid_access,
10770         .convert_ctx_access     = bpf_convert_ctx_access,
10771 };
10772
10773 const struct bpf_prog_ops lwt_out_prog_ops = {
10774         .test_run               = bpf_prog_test_run_skb,
10775 };
10776
10777 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10778         .get_func_proto         = lwt_xmit_func_proto,
10779         .is_valid_access        = lwt_is_valid_access,
10780         .convert_ctx_access     = bpf_convert_ctx_access,
10781         .gen_prologue           = tc_cls_act_prologue,
10782 };
10783
10784 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10785         .test_run               = bpf_prog_test_run_skb,
10786 };
10787
10788 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10789         .get_func_proto         = lwt_seg6local_func_proto,
10790         .is_valid_access        = lwt_is_valid_access,
10791         .convert_ctx_access     = bpf_convert_ctx_access,
10792 };
10793
10794 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10795         .test_run               = bpf_prog_test_run_skb,
10796 };
10797
10798 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10799         .get_func_proto         = sock_filter_func_proto,
10800         .is_valid_access        = sock_filter_is_valid_access,
10801         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10802 };
10803
10804 const struct bpf_prog_ops cg_sock_prog_ops = {
10805 };
10806
10807 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10808         .get_func_proto         = sock_addr_func_proto,
10809         .is_valid_access        = sock_addr_is_valid_access,
10810         .convert_ctx_access     = sock_addr_convert_ctx_access,
10811 };
10812
10813 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10814 };
10815
10816 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10817         .get_func_proto         = sock_ops_func_proto,
10818         .is_valid_access        = sock_ops_is_valid_access,
10819         .convert_ctx_access     = sock_ops_convert_ctx_access,
10820 };
10821
10822 const struct bpf_prog_ops sock_ops_prog_ops = {
10823 };
10824
10825 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10826         .get_func_proto         = sk_skb_func_proto,
10827         .is_valid_access        = sk_skb_is_valid_access,
10828         .convert_ctx_access     = sk_skb_convert_ctx_access,
10829         .gen_prologue           = sk_skb_prologue,
10830 };
10831
10832 const struct bpf_prog_ops sk_skb_prog_ops = {
10833 };
10834
10835 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10836         .get_func_proto         = sk_msg_func_proto,
10837         .is_valid_access        = sk_msg_is_valid_access,
10838         .convert_ctx_access     = sk_msg_convert_ctx_access,
10839         .gen_prologue           = bpf_noop_prologue,
10840 };
10841
10842 const struct bpf_prog_ops sk_msg_prog_ops = {
10843 };
10844
10845 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10846         .get_func_proto         = flow_dissector_func_proto,
10847         .is_valid_access        = flow_dissector_is_valid_access,
10848         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10849 };
10850
10851 const struct bpf_prog_ops flow_dissector_prog_ops = {
10852         .test_run               = bpf_prog_test_run_flow_dissector,
10853 };
10854
10855 int sk_detach_filter(struct sock *sk)
10856 {
10857         int ret = -ENOENT;
10858         struct sk_filter *filter;
10859
10860         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10861                 return -EPERM;
10862
10863         filter = rcu_dereference_protected(sk->sk_filter,
10864                                            lockdep_sock_is_held(sk));
10865         if (filter) {
10866                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10867                 sk_filter_uncharge(sk, filter);
10868                 ret = 0;
10869         }
10870
10871         return ret;
10872 }
10873 EXPORT_SYMBOL_GPL(sk_detach_filter);
10874
10875 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10876 {
10877         struct sock_fprog_kern *fprog;
10878         struct sk_filter *filter;
10879         int ret = 0;
10880
10881         sockopt_lock_sock(sk);
10882         filter = rcu_dereference_protected(sk->sk_filter,
10883                                            lockdep_sock_is_held(sk));
10884         if (!filter)
10885                 goto out;
10886
10887         /* We're copying the filter that has been originally attached,
10888          * so no conversion/decode needed anymore. eBPF programs that
10889          * have no original program cannot be dumped through this.
10890          */
10891         ret = -EACCES;
10892         fprog = filter->prog->orig_prog;
10893         if (!fprog)
10894                 goto out;
10895
10896         ret = fprog->len;
10897         if (!len)
10898                 /* User space only enquires number of filter blocks. */
10899                 goto out;
10900
10901         ret = -EINVAL;
10902         if (len < fprog->len)
10903                 goto out;
10904
10905         ret = -EFAULT;
10906         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10907                 goto out;
10908
10909         /* Instead of bytes, the API requests to return the number
10910          * of filter blocks.
10911          */
10912         ret = fprog->len;
10913 out:
10914         sockopt_release_sock(sk);
10915         return ret;
10916 }
10917
10918 #ifdef CONFIG_INET
10919 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10920                                     struct sock_reuseport *reuse,
10921                                     struct sock *sk, struct sk_buff *skb,
10922                                     struct sock *migrating_sk,
10923                                     u32 hash)
10924 {
10925         reuse_kern->skb = skb;
10926         reuse_kern->sk = sk;
10927         reuse_kern->selected_sk = NULL;
10928         reuse_kern->migrating_sk = migrating_sk;
10929         reuse_kern->data_end = skb->data + skb_headlen(skb);
10930         reuse_kern->hash = hash;
10931         reuse_kern->reuseport_id = reuse->reuseport_id;
10932         reuse_kern->bind_inany = reuse->bind_inany;
10933 }
10934
10935 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10936                                   struct bpf_prog *prog, struct sk_buff *skb,
10937                                   struct sock *migrating_sk,
10938                                   u32 hash)
10939 {
10940         struct sk_reuseport_kern reuse_kern;
10941         enum sk_action action;
10942
10943         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10944         action = bpf_prog_run(prog, &reuse_kern);
10945
10946         if (action == SK_PASS)
10947                 return reuse_kern.selected_sk;
10948         else
10949                 return ERR_PTR(-ECONNREFUSED);
10950 }
10951
10952 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10953            struct bpf_map *, map, void *, key, u32, flags)
10954 {
10955         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10956         struct sock_reuseport *reuse;
10957         struct sock *selected_sk;
10958
10959         selected_sk = map->ops->map_lookup_elem(map, key);
10960         if (!selected_sk)
10961                 return -ENOENT;
10962
10963         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10964         if (!reuse) {
10965                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10966                 if (sk_is_refcounted(selected_sk))
10967                         sock_put(selected_sk);
10968
10969                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10970                  * The only (!reuse) case here is - the sk has already been
10971                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10972                  *
10973                  * Other maps (e.g. sock_map) do not provide this guarantee and
10974                  * the sk may never be in the reuseport group to begin with.
10975                  */
10976                 return is_sockarray ? -ENOENT : -EINVAL;
10977         }
10978
10979         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10980                 struct sock *sk = reuse_kern->sk;
10981
10982                 if (sk->sk_protocol != selected_sk->sk_protocol)
10983                         return -EPROTOTYPE;
10984                 else if (sk->sk_family != selected_sk->sk_family)
10985                         return -EAFNOSUPPORT;
10986
10987                 /* Catch all. Likely bound to a different sockaddr. */
10988                 return -EBADFD;
10989         }
10990
10991         reuse_kern->selected_sk = selected_sk;
10992
10993         return 0;
10994 }
10995
10996 static const struct bpf_func_proto sk_select_reuseport_proto = {
10997         .func           = sk_select_reuseport,
10998         .gpl_only       = false,
10999         .ret_type       = RET_INTEGER,
11000         .arg1_type      = ARG_PTR_TO_CTX,
11001         .arg2_type      = ARG_CONST_MAP_PTR,
11002         .arg3_type      = ARG_PTR_TO_MAP_KEY,
11003         .arg4_type      = ARG_ANYTHING,
11004 };
11005
11006 BPF_CALL_4(sk_reuseport_load_bytes,
11007            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11008            void *, to, u32, len)
11009 {
11010         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
11011 }
11012
11013 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
11014         .func           = sk_reuseport_load_bytes,
11015         .gpl_only       = false,
11016         .ret_type       = RET_INTEGER,
11017         .arg1_type      = ARG_PTR_TO_CTX,
11018         .arg2_type      = ARG_ANYTHING,
11019         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11020         .arg4_type      = ARG_CONST_SIZE,
11021 };
11022
11023 BPF_CALL_5(sk_reuseport_load_bytes_relative,
11024            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
11025            void *, to, u32, len, u32, start_header)
11026 {
11027         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
11028                                                len, start_header);
11029 }
11030
11031 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
11032         .func           = sk_reuseport_load_bytes_relative,
11033         .gpl_only       = false,
11034         .ret_type       = RET_INTEGER,
11035         .arg1_type      = ARG_PTR_TO_CTX,
11036         .arg2_type      = ARG_ANYTHING,
11037         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
11038         .arg4_type      = ARG_CONST_SIZE,
11039         .arg5_type      = ARG_ANYTHING,
11040 };
11041
11042 static const struct bpf_func_proto *
11043 sk_reuseport_func_proto(enum bpf_func_id func_id,
11044                         const struct bpf_prog *prog)
11045 {
11046         switch (func_id) {
11047         case BPF_FUNC_sk_select_reuseport:
11048                 return &sk_select_reuseport_proto;
11049         case BPF_FUNC_skb_load_bytes:
11050                 return &sk_reuseport_load_bytes_proto;
11051         case BPF_FUNC_skb_load_bytes_relative:
11052                 return &sk_reuseport_load_bytes_relative_proto;
11053         case BPF_FUNC_get_socket_cookie:
11054                 return &bpf_get_socket_ptr_cookie_proto;
11055         case BPF_FUNC_ktime_get_coarse_ns:
11056                 return &bpf_ktime_get_coarse_ns_proto;
11057         default:
11058                 return bpf_base_func_proto(func_id);
11059         }
11060 }
11061
11062 static bool
11063 sk_reuseport_is_valid_access(int off, int size,
11064                              enum bpf_access_type type,
11065                              const struct bpf_prog *prog,
11066                              struct bpf_insn_access_aux *info)
11067 {
11068         const u32 size_default = sizeof(__u32);
11069
11070         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
11071             off % size || type != BPF_READ)
11072                 return false;
11073
11074         switch (off) {
11075         case offsetof(struct sk_reuseport_md, data):
11076                 info->reg_type = PTR_TO_PACKET;
11077                 return size == sizeof(__u64);
11078
11079         case offsetof(struct sk_reuseport_md, data_end):
11080                 info->reg_type = PTR_TO_PACKET_END;
11081                 return size == sizeof(__u64);
11082
11083         case offsetof(struct sk_reuseport_md, hash):
11084                 return size == size_default;
11085
11086         case offsetof(struct sk_reuseport_md, sk):
11087                 info->reg_type = PTR_TO_SOCKET;
11088                 return size == sizeof(__u64);
11089
11090         case offsetof(struct sk_reuseport_md, migrating_sk):
11091                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
11092                 return size == sizeof(__u64);
11093
11094         /* Fields that allow narrowing */
11095         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
11096                 if (size < sizeof_field(struct sk_buff, protocol))
11097                         return false;
11098                 fallthrough;
11099         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
11100         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
11101         case bpf_ctx_range(struct sk_reuseport_md, len):
11102                 bpf_ctx_record_field_size(info, size_default);
11103                 return bpf_ctx_narrow_access_ok(off, size, size_default);
11104
11105         default:
11106                 return false;
11107         }
11108 }
11109
11110 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
11111         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
11112                               si->dst_reg, si->src_reg,                 \
11113                               bpf_target_off(struct sk_reuseport_kern, F, \
11114                                              sizeof_field(struct sk_reuseport_kern, F), \
11115                                              target_size));             \
11116         })
11117
11118 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
11119         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11120                                     struct sk_buff,                     \
11121                                     skb,                                \
11122                                     SKB_FIELD)
11123
11124 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
11125         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
11126                                     struct sock,                        \
11127                                     sk,                                 \
11128                                     SK_FIELD)
11129
11130 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
11131                                            const struct bpf_insn *si,
11132                                            struct bpf_insn *insn_buf,
11133                                            struct bpf_prog *prog,
11134                                            u32 *target_size)
11135 {
11136         struct bpf_insn *insn = insn_buf;
11137
11138         switch (si->off) {
11139         case offsetof(struct sk_reuseport_md, data):
11140                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
11141                 break;
11142
11143         case offsetof(struct sk_reuseport_md, len):
11144                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
11145                 break;
11146
11147         case offsetof(struct sk_reuseport_md, eth_protocol):
11148                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
11149                 break;
11150
11151         case offsetof(struct sk_reuseport_md, ip_protocol):
11152                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
11153                 break;
11154
11155         case offsetof(struct sk_reuseport_md, data_end):
11156                 SK_REUSEPORT_LOAD_FIELD(data_end);
11157                 break;
11158
11159         case offsetof(struct sk_reuseport_md, hash):
11160                 SK_REUSEPORT_LOAD_FIELD(hash);
11161                 break;
11162
11163         case offsetof(struct sk_reuseport_md, bind_inany):
11164                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
11165                 break;
11166
11167         case offsetof(struct sk_reuseport_md, sk):
11168                 SK_REUSEPORT_LOAD_FIELD(sk);
11169                 break;
11170
11171         case offsetof(struct sk_reuseport_md, migrating_sk):
11172                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11173                 break;
11174         }
11175
11176         return insn - insn_buf;
11177 }
11178
11179 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11180         .get_func_proto         = sk_reuseport_func_proto,
11181         .is_valid_access        = sk_reuseport_is_valid_access,
11182         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11183 };
11184
11185 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11186 };
11187
11188 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11189 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11190
11191 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11192            struct sock *, sk, u64, flags)
11193 {
11194         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11195                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11196                 return -EINVAL;
11197         if (unlikely(sk && sk_is_refcounted(sk)))
11198                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11199         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11200                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11201         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11202                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11203
11204         /* Check if socket is suitable for packet L3/L4 protocol */
11205         if (sk && sk->sk_protocol != ctx->protocol)
11206                 return -EPROTOTYPE;
11207         if (sk && sk->sk_family != ctx->family &&
11208             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11209                 return -EAFNOSUPPORT;
11210
11211         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11212                 return -EEXIST;
11213
11214         /* Select socket as lookup result */
11215         ctx->selected_sk = sk;
11216         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11217         return 0;
11218 }
11219
11220 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11221         .func           = bpf_sk_lookup_assign,
11222         .gpl_only       = false,
11223         .ret_type       = RET_INTEGER,
11224         .arg1_type      = ARG_PTR_TO_CTX,
11225         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11226         .arg3_type      = ARG_ANYTHING,
11227 };
11228
11229 static const struct bpf_func_proto *
11230 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11231 {
11232         switch (func_id) {
11233         case BPF_FUNC_perf_event_output:
11234                 return &bpf_event_output_data_proto;
11235         case BPF_FUNC_sk_assign:
11236                 return &bpf_sk_lookup_assign_proto;
11237         case BPF_FUNC_sk_release:
11238                 return &bpf_sk_release_proto;
11239         default:
11240                 return bpf_sk_base_func_proto(func_id);
11241         }
11242 }
11243
11244 static bool sk_lookup_is_valid_access(int off, int size,
11245                                       enum bpf_access_type type,
11246                                       const struct bpf_prog *prog,
11247                                       struct bpf_insn_access_aux *info)
11248 {
11249         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11250                 return false;
11251         if (off % size != 0)
11252                 return false;
11253         if (type != BPF_READ)
11254                 return false;
11255
11256         switch (off) {
11257         case offsetof(struct bpf_sk_lookup, sk):
11258                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11259                 return size == sizeof(__u64);
11260
11261         case bpf_ctx_range(struct bpf_sk_lookup, family):
11262         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11263         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11264         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11265         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11266         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11267         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11268         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11269                 bpf_ctx_record_field_size(info, sizeof(__u32));
11270                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11271
11272         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11273                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11274                 if (size == sizeof(__u32))
11275                         return true;
11276                 bpf_ctx_record_field_size(info, sizeof(__be16));
11277                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11278
11279         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11280              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11281                 /* Allow access to zero padding for backward compatibility */
11282                 bpf_ctx_record_field_size(info, sizeof(__u16));
11283                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11284
11285         default:
11286                 return false;
11287         }
11288 }
11289
11290 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11291                                         const struct bpf_insn *si,
11292                                         struct bpf_insn *insn_buf,
11293                                         struct bpf_prog *prog,
11294                                         u32 *target_size)
11295 {
11296         struct bpf_insn *insn = insn_buf;
11297
11298         switch (si->off) {
11299         case offsetof(struct bpf_sk_lookup, sk):
11300                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11301                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11302                 break;
11303
11304         case offsetof(struct bpf_sk_lookup, family):
11305                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11306                                       bpf_target_off(struct bpf_sk_lookup_kern,
11307                                                      family, 2, target_size));
11308                 break;
11309
11310         case offsetof(struct bpf_sk_lookup, protocol):
11311                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11312                                       bpf_target_off(struct bpf_sk_lookup_kern,
11313                                                      protocol, 2, target_size));
11314                 break;
11315
11316         case offsetof(struct bpf_sk_lookup, remote_ip4):
11317                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11318                                       bpf_target_off(struct bpf_sk_lookup_kern,
11319                                                      v4.saddr, 4, target_size));
11320                 break;
11321
11322         case offsetof(struct bpf_sk_lookup, local_ip4):
11323                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11324                                       bpf_target_off(struct bpf_sk_lookup_kern,
11325                                                      v4.daddr, 4, target_size));
11326                 break;
11327
11328         case bpf_ctx_range_till(struct bpf_sk_lookup,
11329                                 remote_ip6[0], remote_ip6[3]): {
11330 #if IS_ENABLED(CONFIG_IPV6)
11331                 int off = si->off;
11332
11333                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11334                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11335                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11336                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11337                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11338                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11339 #else
11340                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11341 #endif
11342                 break;
11343         }
11344         case bpf_ctx_range_till(struct bpf_sk_lookup,
11345                                 local_ip6[0], local_ip6[3]): {
11346 #if IS_ENABLED(CONFIG_IPV6)
11347                 int off = si->off;
11348
11349                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11350                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11351                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11352                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11353                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11354                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11355 #else
11356                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11357 #endif
11358                 break;
11359         }
11360         case offsetof(struct bpf_sk_lookup, remote_port):
11361                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11362                                       bpf_target_off(struct bpf_sk_lookup_kern,
11363                                                      sport, 2, target_size));
11364                 break;
11365
11366         case offsetofend(struct bpf_sk_lookup, remote_port):
11367                 *target_size = 2;
11368                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11369                 break;
11370
11371         case offsetof(struct bpf_sk_lookup, local_port):
11372                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11373                                       bpf_target_off(struct bpf_sk_lookup_kern,
11374                                                      dport, 2, target_size));
11375                 break;
11376
11377         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11378                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11379                                       bpf_target_off(struct bpf_sk_lookup_kern,
11380                                                      ingress_ifindex, 4, target_size));
11381                 break;
11382         }
11383
11384         return insn - insn_buf;
11385 }
11386
11387 const struct bpf_prog_ops sk_lookup_prog_ops = {
11388         .test_run = bpf_prog_test_run_sk_lookup,
11389 };
11390
11391 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11392         .get_func_proto         = sk_lookup_func_proto,
11393         .is_valid_access        = sk_lookup_is_valid_access,
11394         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11395 };
11396
11397 #endif /* CONFIG_INET */
11398
11399 DEFINE_BPF_DISPATCHER(xdp)
11400
11401 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11402 {
11403         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11404 }
11405
11406 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11407 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11408 BTF_SOCK_TYPE_xxx
11409 #undef BTF_SOCK_TYPE
11410
11411 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11412 {
11413         /* tcp6_sock type is not generated in dwarf and hence btf,
11414          * trigger an explicit type generation here.
11415          */
11416         BTF_TYPE_EMIT(struct tcp6_sock);
11417         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11418             sk->sk_family == AF_INET6)
11419                 return (unsigned long)sk;
11420
11421         return (unsigned long)NULL;
11422 }
11423
11424 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11425         .func                   = bpf_skc_to_tcp6_sock,
11426         .gpl_only               = false,
11427         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11428         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11429         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11430 };
11431
11432 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11433 {
11434         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11435                 return (unsigned long)sk;
11436
11437         return (unsigned long)NULL;
11438 }
11439
11440 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11441         .func                   = bpf_skc_to_tcp_sock,
11442         .gpl_only               = false,
11443         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11444         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11445         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11446 };
11447
11448 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11449 {
11450         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11451          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11452          */
11453         BTF_TYPE_EMIT(struct inet_timewait_sock);
11454         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11455
11456 #ifdef CONFIG_INET
11457         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11458                 return (unsigned long)sk;
11459 #endif
11460
11461 #if IS_BUILTIN(CONFIG_IPV6)
11462         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11463                 return (unsigned long)sk;
11464 #endif
11465
11466         return (unsigned long)NULL;
11467 }
11468
11469 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11470         .func                   = bpf_skc_to_tcp_timewait_sock,
11471         .gpl_only               = false,
11472         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11473         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11474         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11475 };
11476
11477 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11478 {
11479 #ifdef CONFIG_INET
11480         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11481                 return (unsigned long)sk;
11482 #endif
11483
11484 #if IS_BUILTIN(CONFIG_IPV6)
11485         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11486                 return (unsigned long)sk;
11487 #endif
11488
11489         return (unsigned long)NULL;
11490 }
11491
11492 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11493         .func                   = bpf_skc_to_tcp_request_sock,
11494         .gpl_only               = false,
11495         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11496         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11497         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11498 };
11499
11500 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11501 {
11502         /* udp6_sock type is not generated in dwarf and hence btf,
11503          * trigger an explicit type generation here.
11504          */
11505         BTF_TYPE_EMIT(struct udp6_sock);
11506         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11507             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11508                 return (unsigned long)sk;
11509
11510         return (unsigned long)NULL;
11511 }
11512
11513 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11514         .func                   = bpf_skc_to_udp6_sock,
11515         .gpl_only               = false,
11516         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11517         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11518         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11519 };
11520
11521 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11522 {
11523         /* unix_sock type is not generated in dwarf and hence btf,
11524          * trigger an explicit type generation here.
11525          */
11526         BTF_TYPE_EMIT(struct unix_sock);
11527         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11528                 return (unsigned long)sk;
11529
11530         return (unsigned long)NULL;
11531 }
11532
11533 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11534         .func                   = bpf_skc_to_unix_sock,
11535         .gpl_only               = false,
11536         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11537         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11538         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11539 };
11540
11541 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11542 {
11543         BTF_TYPE_EMIT(struct mptcp_sock);
11544         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11545 }
11546
11547 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11548         .func           = bpf_skc_to_mptcp_sock,
11549         .gpl_only       = false,
11550         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11551         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11552         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11553 };
11554
11555 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11556 {
11557         return (unsigned long)sock_from_file(file);
11558 }
11559
11560 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11561 BTF_ID(struct, socket)
11562 BTF_ID(struct, file)
11563
11564 const struct bpf_func_proto bpf_sock_from_file_proto = {
11565         .func           = bpf_sock_from_file,
11566         .gpl_only       = false,
11567         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11568         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11569         .arg1_type      = ARG_PTR_TO_BTF_ID,
11570         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11571 };
11572
11573 static const struct bpf_func_proto *
11574 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11575 {
11576         const struct bpf_func_proto *func;
11577
11578         switch (func_id) {
11579         case BPF_FUNC_skc_to_tcp6_sock:
11580                 func = &bpf_skc_to_tcp6_sock_proto;
11581                 break;
11582         case BPF_FUNC_skc_to_tcp_sock:
11583                 func = &bpf_skc_to_tcp_sock_proto;
11584                 break;
11585         case BPF_FUNC_skc_to_tcp_timewait_sock:
11586                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11587                 break;
11588         case BPF_FUNC_skc_to_tcp_request_sock:
11589                 func = &bpf_skc_to_tcp_request_sock_proto;
11590                 break;
11591         case BPF_FUNC_skc_to_udp6_sock:
11592                 func = &bpf_skc_to_udp6_sock_proto;
11593                 break;
11594         case BPF_FUNC_skc_to_unix_sock:
11595                 func = &bpf_skc_to_unix_sock_proto;
11596                 break;
11597         case BPF_FUNC_skc_to_mptcp_sock:
11598                 func = &bpf_skc_to_mptcp_sock_proto;
11599                 break;
11600         case BPF_FUNC_ktime_get_coarse_ns:
11601                 return &bpf_ktime_get_coarse_ns_proto;
11602         default:
11603                 return bpf_base_func_proto(func_id);
11604         }
11605
11606         if (!perfmon_capable())
11607                 return NULL;
11608
11609         return func;
11610 }