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/module.h>
  21 #include <linux/types.h>
  22 #include <linux/mm.h>
  23 #include <linux/fcntl.h>
  24 #include <linux/socket.h>
  25 #include <linux/sock_diag.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/if_packet.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/gfp.h>
  32 #include <net/inet_common.h>
  33 #include <net/ip.h>
  34 #include <net/protocol.h>
  35 #include <net/netlink.h>
  36 #include <linux/skbuff.h>
  37 #include <linux/skmsg.h>
  38 #include <net/sock.h>
  39 #include <net/flow_dissector.h>
  40 #include <linux/errno.h>
  41 #include <linux/timer.h>
  42 #include <linux/uaccess.h>
  43 #include <asm/unaligned.h>
  44 #include <asm/cmpxchg.h>
  45 #include <linux/filter.h>
  46 #include <linux/ratelimit.h>
  47 #include <linux/seccomp.h>
  48 #include <linux/if_vlan.h>
  49 #include <linux/bpf.h>
  50 #include <linux/btf.h>
  51 #include <net/sch_generic.h>
  52 #include <net/cls_cgroup.h>
  53 #include <net/dst_metadata.h>
  54 #include <net/dst.h>
  55 #include <net/sock_reuseport.h>
  56 #include <net/busy_poll.h>
  57 #include <net/tcp.h>
  58 #include <net/xfrm.h>
  59 #include <net/udp.h>
  60 #include <linux/bpf_trace.h>
  61 #include <net/xdp_sock.h>
  62 #include <linux/inetdevice.h>
  63 #include <net/inet_hashtables.h>
  64 #include <net/inet6_hashtables.h>
  65 #include <net/ip_fib.h>
  66 #include <net/nexthop.h>
  67 #include <net/flow.h>
  68 #include <net/arp.h>
  69 #include <net/ipv6.h>
  70 #include <net/net_namespace.h>
  71 #include <linux/seg6_local.h>
  72 #include <net/seg6.h>
  73 #include <net/seg6_local.h>
  74 #include <net/lwtunnel.h>
  75 #include <net/ipv6_stubs.h>
  76 #include <net/bpf_sk_storage.h>
  77 #include <net/transp_v6.h>
  78 #include <linux/btf_ids.h>
  79 #include <net/tls.h>
  80
  81 static const struct bpf_func_proto *
  82 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  83
  84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  85 {
  86         if (in_compat_syscall()) {
  87                 struct compat_sock_fprog f32;
  88
  89                 if (len != sizeof(f32))
  90                         return -EINVAL;
  91                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  92                         return -EFAULT;
  93                 memset(dst, 0, sizeof(*dst));
  94                 dst->len = f32.len;
  95                 dst->filter = compat_ptr(f32.filter);
  96         } else {
  97                 if (len != sizeof(*dst))
  98                         return -EINVAL;
  99                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 100                         return -EFAULT;
 101         }
 102
 103         return 0;
 104 }
 105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 106
 107 /**
 108  *      sk_filter_trim_cap - run a packet through a socket filter
 109  *      @sk: sock associated with &sk_buff
 110  *      @skb: buffer to filter
 111  *      @cap: limit on how short the eBPF program may trim the packet
 112  *
 113  * Run the eBPF program and then cut skb->data to correct size returned by
 114  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 115  * than pkt_len we keep whole skb->data. This is the socket level
 116  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
 117  * be accepted or -EPERM if the packet should be tossed.
 118  *
 119  */
 120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 121 {
 122         int err;
 123         struct sk_filter *filter;
 124
 125         /*
 126          * If the skb was allocated from pfmemalloc reserves, only
 127          * allow SOCK_MEMALLOC sockets to use it as this socket is
 128          * helping free memory
 129          */
 130         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 131                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 132                 return -ENOMEM;
 133         }
 134         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 135         if (err)
 136                 return err;
 137
 138         err = security_sock_rcv_skb(sk, skb);
 139         if (err)
 140                 return err;
 141
 142         rcu_read_lock();
 143         filter = rcu_dereference(sk->sk_filter);
 144         if (filter) {
 145                 struct sock *save_sk = skb->sk;
 146                 unsigned int pkt_len;
 147
 148                 skb->sk = sk;
 149                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 150                 skb->sk = save_sk;
 151                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 152         }
 153         rcu_read_unlock();
 154
 155         return err;
 156 }
 157 EXPORT_SYMBOL(sk_filter_trim_cap);
 158
 159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 160 {
 161         return skb_get_poff(skb);
 162 }
 163
 164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 165 {
 166         struct nlattr *nla;
 167
 168         if (skb_is_nonlinear(skb))
 169                 return 0;
 170
 171         if (skb->len < sizeof(struct nlattr))
 172                 return 0;
 173
 174         if (a > skb->len - sizeof(struct nlattr))
 175                 return 0;
 176
 177         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 178         if (nla)
 179                 return (void *) nla - (void *) skb->data;
 180
 181         return 0;
 182 }
 183
 184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 185 {
 186         struct nlattr *nla;
 187
 188         if (skb_is_nonlinear(skb))
 189                 return 0;
 190
 191         if (skb->len < sizeof(struct nlattr))
 192                 return 0;
 193
 194         if (a > skb->len - sizeof(struct nlattr))
 195                 return 0;
 196
 197         nla = (struct nlattr *) &skb->data[a];
 198         if (nla->nla_len > skb->len - a)
 199                 return 0;
 200
 201         nla = nla_find_nested(nla, x);
 202         if (nla)
 203                 return (void *) nla - (void *) skb->data;
 204
 205         return 0;
 206 }
 207
 208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 209            data, int, headlen, int, offset)
 210 {
 211         u8 tmp, *ptr;
 212         const int len = sizeof(tmp);
 213
 214         if (offset >= 0) {
 215                 if (headlen - offset >= len)
 216                         return *(u8 *)(data + offset);
 217                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 218                         return tmp;
 219         } else {
 220                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 221                 if (likely(ptr))
 222                         return *(u8 *)ptr;
 223         }
 224
 225         return -EFAULT;
 226 }
 227
 228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 229            int, offset)
 230 {
 231         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 232                                          offset);
 233 }
 234
 235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 236            data, int, headlen, int, offset)
 237 {
 238         u16 tmp, *ptr;
 239         const int len = sizeof(tmp);
 240
 241         if (offset >= 0) {
 242                 if (headlen - offset >= len)
 243                         return get_unaligned_be16(data + offset);
 244                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 245                         return be16_to_cpu(tmp);
 246         } else {
 247                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 248                 if (likely(ptr))
 249                         return get_unaligned_be16(ptr);
 250         }
 251
 252         return -EFAULT;
 253 }
 254
 255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 256            int, offset)
 257 {
 258         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 259                                           offset);
 260 }
 261
 262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 263            data, int, headlen, int, offset)
 264 {
 265         u32 tmp, *ptr;
 266         const int len = sizeof(tmp);
 267
 268         if (likely(offset >= 0)) {
 269                 if (headlen - offset >= len)
 270                         return get_unaligned_be32(data + offset);
 271                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 272                         return be32_to_cpu(tmp);
 273         } else {
 274                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 275                 if (likely(ptr))
 276                         return get_unaligned_be32(ptr);
 277         }
 278
 279         return -EFAULT;
 280 }
 281
 282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 283            int, offset)
 284 {
 285         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 286                                           offset);
 287 }
 288
 289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 290                               struct bpf_insn *insn_buf)
 291 {
 292         struct bpf_insn *insn = insn_buf;
 293
 294         switch (skb_field) {
 295         case SKF_AD_MARK:
 296                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 297
 298                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 299                                       offsetof(struct sk_buff, mark));
 300                 break;
 301
 302         case SKF_AD_PKTTYPE:
 303                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 304                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 305 #ifdef __BIG_ENDIAN_BITFIELD
 306                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 307 #endif
 308                 break;
 309
 310         case SKF_AD_QUEUE:
 311                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 312
 313                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 314                                       offsetof(struct sk_buff, queue_mapping));
 315                 break;
 316
 317         case SKF_AD_VLAN_TAG:
 318                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 319
 320                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 321                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 322                                       offsetof(struct sk_buff, vlan_tci));
 323                 break;
 324         case SKF_AD_VLAN_TAG_PRESENT:
 325                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
 326                 if (PKT_VLAN_PRESENT_BIT)
 327                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 328                 if (PKT_VLAN_PRESENT_BIT < 7)
 329                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 330                 break;
 331         }
 332
 333         return insn - insn_buf;
 334 }
 335
 336 static bool convert_bpf_extensions(struct sock_filter *fp,
 337                                    struct bpf_insn **insnp)
 338 {
 339         struct bpf_insn *insn = *insnp;
 340         u32 cnt;
 341
 342         switch (fp->k) {
 343         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 344                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 345
 346                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 347                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 348                                       offsetof(struct sk_buff, protocol));
 349                 /* A = ntohs(A) [emitting a nop or swap16] */
 350                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 351                 break;
 352
 353         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 354                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 355                 insn += cnt - 1;
 356                 break;
 357
 358         case SKF_AD_OFF + SKF_AD_IFINDEX:
 359         case SKF_AD_OFF + SKF_AD_HATYPE:
 360                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 361                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 362
 363                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 364                                       BPF_REG_TMP, BPF_REG_CTX,
 365                                       offsetof(struct sk_buff, dev));
 366                 /* if (tmp != 0) goto pc + 1 */
 367                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 368                 *insn++ = BPF_EXIT_INSN();
 369                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 370                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 371                                             offsetof(struct net_device, ifindex));
 372                 else
 373                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 374                                             offsetof(struct net_device, type));
 375                 break;
 376
 377         case SKF_AD_OFF + SKF_AD_MARK:
 378                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 379                 insn += cnt - 1;
 380                 break;
 381
 382         case SKF_AD_OFF + SKF_AD_RXHASH:
 383                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 384
 385                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 386                                     offsetof(struct sk_buff, hash));
 387                 break;
 388
 389         case SKF_AD_OFF + SKF_AD_QUEUE:
 390                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 391                 insn += cnt - 1;
 392                 break;
 393
 394         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 395                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 396                                          BPF_REG_A, BPF_REG_CTX, insn);
 397                 insn += cnt - 1;
 398                 break;
 399
 400         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 401                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 402                                          BPF_REG_A, BPF_REG_CTX, insn);
 403                 insn += cnt - 1;
 404                 break;
 405
 406         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 407                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 408
 409                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 410                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 411                                       offsetof(struct sk_buff, vlan_proto));
 412                 /* A = ntohs(A) [emitting a nop or swap16] */
 413                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 414                 break;
 415
 416         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 417         case SKF_AD_OFF + SKF_AD_NLATTR:
 418         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 419         case SKF_AD_OFF + SKF_AD_CPU:
 420         case SKF_AD_OFF + SKF_AD_RANDOM:
 421                 /* arg1 = CTX */
 422                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 423                 /* arg2 = A */
 424                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 425                 /* arg3 = X */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 427                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 428                 switch (fp->k) {
 429                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 430                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 431                         break;
 432                 case SKF_AD_OFF + SKF_AD_NLATTR:
 433                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 434                         break;
 435                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 436                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 437                         break;
 438                 case SKF_AD_OFF + SKF_AD_CPU:
 439                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 440                         break;
 441                 case SKF_AD_OFF + SKF_AD_RANDOM:
 442                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 443                         bpf_user_rnd_init_once();
 444                         break;
 445                 }
 446                 break;
 447
 448         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 449                 /* A ^= X */
 450                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 451                 break;
 452
 453         default:
 454                 /* This is just a dummy call to avoid letting the compiler
 455                  * evict __bpf_call_base() as an optimization. Placed here
 456                  * where no-one bothers.
 457                  */
 458                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 459                 return false;
 460         }
 461
 462         *insnp = insn;
 463         return true;
 464 }
 465
 466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 467 {
 468         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 469         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 470         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 471                       BPF_SIZE(fp->code) == BPF_W;
 472         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 473         const int ip_align = NET_IP_ALIGN;
 474         struct bpf_insn *insn = *insnp;
 475         int offset = fp->k;
 476
 477         if (!indirect &&
 478             ((unaligned_ok && offset >= 0) ||
 479              (!unaligned_ok && offset >= 0 &&
 480               offset + ip_align >= 0 &&
 481               offset + ip_align % size == 0))) {
 482                 bool ldx_off_ok = offset <= S16_MAX;
 483
 484                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 485                 if (offset)
 486                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 487                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 488                                       size, 2 + endian + (!ldx_off_ok * 2));
 489                 if (ldx_off_ok) {
 490                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 491                                               BPF_REG_D, offset);
 492                 } else {
 493                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 494                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 495                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 496                                               BPF_REG_TMP, 0);
 497                 }
 498                 if (endian)
 499                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 500                 *insn++ = BPF_JMP_A(8);
 501         }
 502
 503         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 504         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 505         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 506         if (!indirect) {
 507                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 508         } else {
 509                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 510                 if (fp->k)
 511                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 512         }
 513
 514         switch (BPF_SIZE(fp->code)) {
 515         case BPF_B:
 516                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 517                 break;
 518         case BPF_H:
 519                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 520                 break;
 521         case BPF_W:
 522                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 523                 break;
 524         default:
 525                 return false;
 526         }
 527
 528         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 529         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 530         *insn   = BPF_EXIT_INSN();
 531
 532         *insnp = insn;
 533         return true;
 534 }
 535
 536 /**
 537  *      bpf_convert_filter - convert filter program
 538  *      @prog: the user passed filter program
 539  *      @len: the length of the user passed filter program
 540  *      @new_prog: allocated 'struct bpf_prog' or NULL
 541  *      @new_len: pointer to store length of converted program
 542  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 543  *
 544  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 545  * style extended BPF (eBPF).
 546  * Conversion workflow:
 547  *
 548  * 1) First pass for calculating the new program length:
 549  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 550  *
 551  * 2) 2nd pass to remap in two passes: 1st pass finds new
 552  *    jump offsets, 2nd pass remapping:
 553  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 554  */
 555 static int bpf_convert_filter(struct sock_filter *prog, int len,
 556                               struct bpf_prog *new_prog, int *new_len,
 557                               bool *seen_ld_abs)
 558 {
 559         int new_flen = 0, pass = 0, target, i, stack_off;
 560         struct bpf_insn *new_insn, *first_insn = NULL;
 561         struct sock_filter *fp;
 562         int *addrs = NULL;
 563         u8 bpf_src;
 564
 565         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 566         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 567
 568         if (len <= 0 || len > BPF_MAXINSNS)
 569                 return -EINVAL;
 570
 571         if (new_prog) {
 572                 first_insn = new_prog->insnsi;
 573                 addrs = kcalloc(len, sizeof(*addrs),
 574                                 GFP_KERNEL | __GFP_NOWARN);
 575                 if (!addrs)
 576                         return -ENOMEM;
 577         }
 578
 579 do_pass:
 580         new_insn = first_insn;
 581         fp = prog;
 582
 583         /* Classic BPF related prologue emission. */
 584         if (new_prog) {
 585                 /* Classic BPF expects A and X to be reset first. These need
 586                  * to be guaranteed to be the first two instructions.
 587                  */
 588                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 589                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 590
 591                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 592                  * In eBPF case it's done by the compiler, here we need to
 593                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 594                  */
 595                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 596                 if (*seen_ld_abs) {
 597                         /* For packet access in classic BPF, cache skb->data
 598                          * in callee-saved BPF R8 and skb->len - skb->data_len
 599                          * (headlen) in BPF R9. Since classic BPF is read-only
 600                          * on CTX, we only need to cache it once.
 601                          */
 602                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 603                                                   BPF_REG_D, BPF_REG_CTX,
 604                                                   offsetof(struct sk_buff, data));
 605                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 606                                                   offsetof(struct sk_buff, len));
 607                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, data_len));
 609                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 610                 }
 611         } else {
 612                 new_insn += 3;
 613         }
 614
 615         for (i = 0; i < len; fp++, i++) {
 616                 struct bpf_insn tmp_insns[32] = { };
 617                 struct bpf_insn *insn = tmp_insns;
 618
 619                 if (addrs)
 620                         addrs[i] = new_insn - first_insn;
 621
 622                 switch (fp->code) {
 623                 /* All arithmetic insns and skb loads map as-is. */
 624                 case BPF_ALU | BPF_ADD | BPF_X:
 625                 case BPF_ALU | BPF_ADD | BPF_K:
 626                 case BPF_ALU | BPF_SUB | BPF_X:
 627                 case BPF_ALU | BPF_SUB | BPF_K:
 628                 case BPF_ALU | BPF_AND | BPF_X:
 629                 case BPF_ALU | BPF_AND | BPF_K:
 630                 case BPF_ALU | BPF_OR | BPF_X:
 631                 case BPF_ALU | BPF_OR | BPF_K:
 632                 case BPF_ALU | BPF_LSH | BPF_X:
 633                 case BPF_ALU | BPF_LSH | BPF_K:
 634                 case BPF_ALU | BPF_RSH | BPF_X:
 635                 case BPF_ALU | BPF_RSH | BPF_K:
 636                 case BPF_ALU | BPF_XOR | BPF_X:
 637                 case BPF_ALU | BPF_XOR | BPF_K:
 638                 case BPF_ALU | BPF_MUL | BPF_X:
 639                 case BPF_ALU | BPF_MUL | BPF_K:
 640                 case BPF_ALU | BPF_DIV | BPF_X:
 641                 case BPF_ALU | BPF_DIV | BPF_K:
 642                 case BPF_ALU | BPF_MOD | BPF_X:
 643                 case BPF_ALU | BPF_MOD | BPF_K:
 644                 case BPF_ALU | BPF_NEG:
 645                 case BPF_LD | BPF_ABS | BPF_W:
 646                 case BPF_LD | BPF_ABS | BPF_H:
 647                 case BPF_LD | BPF_ABS | BPF_B:
 648                 case BPF_LD | BPF_IND | BPF_W:
 649                 case BPF_LD | BPF_IND | BPF_H:
 650                 case BPF_LD | BPF_IND | BPF_B:
 651                         /* Check for overloaded BPF extension and
 652                          * directly convert it if found, otherwise
 653                          * just move on with mapping.
 654                          */
 655                         if (BPF_CLASS(fp->code) == BPF_LD &&
 656                             BPF_MODE(fp->code) == BPF_ABS &&
 657                             convert_bpf_extensions(fp, &insn))
 658                                 break;
 659                         if (BPF_CLASS(fp->code) == BPF_LD &&
 660                             convert_bpf_ld_abs(fp, &insn)) {
 661                                 *seen_ld_abs = true;
 662                                 break;
 663                         }
 664
 665                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 666                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 667                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 668                                 /* Error with exception code on div/mod by 0.
 669                                  * For cBPF programs, this was always return 0.
 670                                  */
 671                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 672                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 673                                 *insn++ = BPF_EXIT_INSN();
 674                         }
 675
 676                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 677                         break;
 678
 679                 /* Jump transformation cannot use BPF block macros
 680                  * everywhere as offset calculation and target updates
 681                  * require a bit more work than the rest, i.e. jump
 682                  * opcodes map as-is, but offsets need adjustment.
 683                  */
 684
 685 #define BPF_EMIT_JMP                                                    \
 686         do {                                                            \
 687                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 688                 s32 off;                                                \
 689                                                                         \
 690                 if (target >= len || target < 0)                        \
 691                         goto err;                                       \
 692                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 693                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 694                 off -= insn - tmp_insns;                                \
 695                 /* Reject anything not fitting into insn->off. */       \
 696                 if (off < off_min || off > off_max)                     \
 697                         goto err;                                       \
 698                 insn->off = off;                                        \
 699         } while (0)
 700
 701                 case BPF_JMP | BPF_JA:
 702                         target = i + fp->k + 1;
 703                         insn->code = fp->code;
 704                         BPF_EMIT_JMP;
 705                         break;
 706
 707                 case BPF_JMP | BPF_JEQ | BPF_K:
 708                 case BPF_JMP | BPF_JEQ | BPF_X:
 709                 case BPF_JMP | BPF_JSET | BPF_K:
 710                 case BPF_JMP | BPF_JSET | BPF_X:
 711                 case BPF_JMP | BPF_JGT | BPF_K:
 712                 case BPF_JMP | BPF_JGT | BPF_X:
 713                 case BPF_JMP | BPF_JGE | BPF_K:
 714                 case BPF_JMP | BPF_JGE | BPF_X:
 715                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 716                                 /* BPF immediates are signed, zero extend
 717                                  * immediate into tmp register and use it
 718                                  * in compare insn.
 719                                  */
 720                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 721
 722                                 insn->dst_reg = BPF_REG_A;
 723                                 insn->src_reg = BPF_REG_TMP;
 724                                 bpf_src = BPF_X;
 725                         } else {
 726                                 insn->dst_reg = BPF_REG_A;
 727                                 insn->imm = fp->k;
 728                                 bpf_src = BPF_SRC(fp->code);
 729                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 730                         }
 731
 732                         /* Common case where 'jump_false' is next insn. */
 733                         if (fp->jf == 0) {
 734                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 735                                 target = i + fp->jt + 1;
 736                                 BPF_EMIT_JMP;
 737                                 break;
 738                         }
 739
 740                         /* Convert some jumps when 'jump_true' is next insn. */
 741                         if (fp->jt == 0) {
 742                                 switch (BPF_OP(fp->code)) {
 743                                 case BPF_JEQ:
 744                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 745                                         break;
 746                                 case BPF_JGT:
 747                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 748                                         break;
 749                                 case BPF_JGE:
 750                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 751                                         break;
 752                                 default:
 753                                         goto jmp_rest;
 754                                 }
 755
 756                                 target = i + fp->jf + 1;
 757                                 BPF_EMIT_JMP;
 758                                 break;
 759                         }
 760 jmp_rest:
 761                         /* Other jumps are mapped into two insns: Jxx and JA. */
 762                         target = i + fp->jt + 1;
 763                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 764                         BPF_EMIT_JMP;
 765                         insn++;
 766
 767                         insn->code = BPF_JMP | BPF_JA;
 768                         target = i + fp->jf + 1;
 769                         BPF_EMIT_JMP;
 770                         break;
 771
 772                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 773                 case BPF_LDX | BPF_MSH | BPF_B: {
 774                         struct sock_filter tmp = {
 775                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 776                                 .k      = fp->k,
 777                         };
 778
 779                         *seen_ld_abs = true;
 780
 781                         /* X = A */
 782                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 783                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 784                         convert_bpf_ld_abs(&tmp, &insn);
 785                         insn++;
 786                         /* A &= 0xf */
 787                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 788                         /* A <<= 2 */
 789                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 790                         /* tmp = X */
 791                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 792                         /* X = A */
 793                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 794                         /* A = tmp */
 795                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 796                         break;
 797                 }
 798                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 799                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 800                  */
 801                 case BPF_RET | BPF_A:
 802                 case BPF_RET | BPF_K:
 803                         if (BPF_RVAL(fp->code) == BPF_K)
 804                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 805                                                         0, fp->k);
 806                         *insn = BPF_EXIT_INSN();
 807                         break;
 808
 809                 /* Store to stack. */
 810                 case BPF_ST:
 811                 case BPF_STX:
 812                         stack_off = fp->k * 4  + 4;
 813                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 814                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 815                                             -stack_off);
 816                         /* check_load_and_stores() verifies that classic BPF can
 817                          * load from stack only after write, so tracking
 818                          * stack_depth for ST|STX insns is enough
 819                          */
 820                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 821                                 new_prog->aux->stack_depth = stack_off;
 822                         break;
 823
 824                 /* Load from stack. */
 825                 case BPF_LD | BPF_MEM:
 826                 case BPF_LDX | BPF_MEM:
 827                         stack_off = fp->k * 4  + 4;
 828                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 829                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 830                                             -stack_off);
 831                         break;
 832
 833                 /* A = K or X = K */
 834                 case BPF_LD | BPF_IMM:
 835                 case BPF_LDX | BPF_IMM:
 836                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 837                                               BPF_REG_A : BPF_REG_X, fp->k);
 838                         break;
 839
 840                 /* X = A */
 841                 case BPF_MISC | BPF_TAX:
 842                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 843                         break;
 844
 845                 /* A = X */
 846                 case BPF_MISC | BPF_TXA:
 847                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 848                         break;
 849
 850                 /* A = skb->len or X = skb->len */
 851                 case BPF_LD | BPF_W | BPF_LEN:
 852                 case BPF_LDX | BPF_W | BPF_LEN:
 853                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 854                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 855                                             offsetof(struct sk_buff, len));
 856                         break;
 857
 858                 /* Access seccomp_data fields. */
 859                 case BPF_LDX | BPF_ABS | BPF_W:
 860                         /* A = *(u32 *) (ctx + K) */
 861                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 862                         break;
 863
 864                 /* Unknown instruction. */
 865                 default:
 866                         goto err;
 867                 }
 868
 869                 insn++;
 870                 if (new_prog)
 871                         memcpy(new_insn, tmp_insns,
 872                                sizeof(*insn) * (insn - tmp_insns));
 873                 new_insn += insn - tmp_insns;
 874         }
 875
 876         if (!new_prog) {
 877                 /* Only calculating new length. */
 878                 *new_len = new_insn - first_insn;
 879                 if (*seen_ld_abs)
 880                         *new_len += 4; /* Prologue bits. */
 881                 return 0;
 882         }
 883
 884         pass++;
 885         if (new_flen != new_insn - first_insn) {
 886                 new_flen = new_insn - first_insn;
 887                 if (pass > 2)
 888                         goto err;
 889                 goto do_pass;
 890         }
 891
 892         kfree(addrs);
 893         BUG_ON(*new_len != new_flen);
 894         return 0;
 895 err:
 896         kfree(addrs);
 897         return -EINVAL;
 898 }
 899
 900 /* Security:
 901  *
 902  * As we dont want to clear mem[] array for each packet going through
 903  * __bpf_prog_run(), we check that filter loaded by user never try to read
 904  * a cell if not previously written, and we check all branches to be sure
 905  * a malicious user doesn't try to abuse us.
 906  */
 907 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 908 {
 909         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 910         int pc, ret = 0;
 911
 912         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 913
 914         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 915         if (!masks)
 916                 return -ENOMEM;
 917
 918         memset(masks, 0xff, flen * sizeof(*masks));
 919
 920         for (pc = 0; pc < flen; pc++) {
 921                 memvalid &= masks[pc];
 922
 923                 switch (filter[pc].code) {
 924                 case BPF_ST:
 925                 case BPF_STX:
 926                         memvalid |= (1 << filter[pc].k);
 927                         break;
 928                 case BPF_LD | BPF_MEM:
 929                 case BPF_LDX | BPF_MEM:
 930                         if (!(memvalid & (1 << filter[pc].k))) {
 931                                 ret = -EINVAL;
 932                                 goto error;
 933                         }
 934                         break;
 935                 case BPF_JMP | BPF_JA:
 936                         /* A jump must set masks on target */
 937                         masks[pc + 1 + filter[pc].k] &= memvalid;
 938                         memvalid = ~0;
 939                         break;
 940                 case BPF_JMP | BPF_JEQ | BPF_K:
 941                 case BPF_JMP | BPF_JEQ | BPF_X:
 942                 case BPF_JMP | BPF_JGE | BPF_K:
 943                 case BPF_JMP | BPF_JGE | BPF_X:
 944                 case BPF_JMP | BPF_JGT | BPF_K:
 945                 case BPF_JMP | BPF_JGT | BPF_X:
 946                 case BPF_JMP | BPF_JSET | BPF_K:
 947                 case BPF_JMP | BPF_JSET | BPF_X:
 948                         /* A jump must set masks on targets */
 949                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 950                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 951                         memvalid = ~0;
 952                         break;
 953                 }
 954         }
 955 error:
 956         kfree(masks);
 957         return ret;
 958 }
 959
 960 static bool chk_code_allowed(u16 code_to_probe)
 961 {
 962         static const bool codes[] = {
 963                 /* 32 bit ALU operations */
 964                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 965                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 966                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 967                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 968                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 969                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 970                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 971                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 972                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 973                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 974                 [BPF_ALU | BPF_AND | BPF_K] = true,
 975                 [BPF_ALU | BPF_AND | BPF_X] = true,
 976                 [BPF_ALU | BPF_OR | BPF_K] = true,
 977                 [BPF_ALU | BPF_OR | BPF_X] = true,
 978                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 979                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 980                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 981                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 982                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 983                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 984                 [BPF_ALU | BPF_NEG] = true,
 985                 /* Load instructions */
 986                 [BPF_LD | BPF_W | BPF_ABS] = true,
 987                 [BPF_LD | BPF_H | BPF_ABS] = true,
 988                 [BPF_LD | BPF_B | BPF_ABS] = true,
 989                 [BPF_LD | BPF_W | BPF_LEN] = true,
 990                 [BPF_LD | BPF_W | BPF_IND] = true,
 991                 [BPF_LD | BPF_H | BPF_IND] = true,
 992                 [BPF_LD | BPF_B | BPF_IND] = true,
 993                 [BPF_LD | BPF_IMM] = true,
 994                 [BPF_LD | BPF_MEM] = true,
 995                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 996                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 997                 [BPF_LDX | BPF_IMM] = true,
 998                 [BPF_LDX | BPF_MEM] = true,
 999                 /* Store instructions */
1000                 [BPF_ST] = true,
1001                 [BPF_STX] = true,
1002                 /* Misc instructions */
1003                 [BPF_MISC | BPF_TAX] = true,
1004                 [BPF_MISC | BPF_TXA] = true,
1005                 /* Return instructions */
1006                 [BPF_RET | BPF_K] = true,
1007                 [BPF_RET | BPF_A] = true,
1008                 /* Jump instructions */
1009                 [BPF_JMP | BPF_JA] = true,
1010                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1011                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1012                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1013                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1014                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1015                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1016                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1017                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1018         };
1019
1020         if (code_to_probe >= ARRAY_SIZE(codes))
1021                 return false;
1022
1023         return codes[code_to_probe];
1024 }
1025
1026 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1027                                 unsigned int flen)
1028 {
1029         if (filter == NULL)
1030                 return false;
1031         if (flen == 0 || flen > BPF_MAXINSNS)
1032                 return false;
1033
1034         return true;
1035 }
1036
1037 /**
1038  *      bpf_check_classic - verify socket filter code
1039  *      @filter: filter to verify
1040  *      @flen: length of filter
1041  *
1042  * Check the user's filter code. If we let some ugly
1043  * filter code slip through kaboom! The filter must contain
1044  * no references or jumps that are out of range, no illegal
1045  * instructions, and must end with a RET instruction.
1046  *
1047  * All jumps are forward as they are not signed.
1048  *
1049  * Returns 0 if the rule set is legal or -EINVAL if not.
1050  */
1051 static int bpf_check_classic(const struct sock_filter *filter,
1052                              unsigned int flen)
1053 {
1054         bool anc_found;
1055         int pc;
1056
1057         /* Check the filter code now */
1058         for (pc = 0; pc < flen; pc++) {
1059                 const struct sock_filter *ftest = &filter[pc];
1060
1061                 /* May we actually operate on this code? */
1062                 if (!chk_code_allowed(ftest->code))
1063                         return -EINVAL;
1064
1065                 /* Some instructions need special checks */
1066                 switch (ftest->code) {
1067                 case BPF_ALU | BPF_DIV | BPF_K:
1068                 case BPF_ALU | BPF_MOD | BPF_K:
1069                         /* Check for division by zero */
1070                         if (ftest->k == 0)
1071                                 return -EINVAL;
1072                         break;
1073                 case BPF_ALU | BPF_LSH | BPF_K:
1074                 case BPF_ALU | BPF_RSH | BPF_K:
1075                         if (ftest->k >= 32)
1076                                 return -EINVAL;
1077                         break;
1078                 case BPF_LD | BPF_MEM:
1079                 case BPF_LDX | BPF_MEM:
1080                 case BPF_ST:
1081                 case BPF_STX:
1082                         /* Check for invalid memory addresses */
1083                         if (ftest->k >= BPF_MEMWORDS)
1084                                 return -EINVAL;
1085                         break;
1086                 case BPF_JMP | BPF_JA:
1087                         /* Note, the large ftest->k might cause loops.
1088                          * Compare this with conditional jumps below,
1089                          * where offsets are limited. --ANK (981016)
1090                          */
1091                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1092                                 return -EINVAL;
1093                         break;
1094                 case BPF_JMP | BPF_JEQ | BPF_K:
1095                 case BPF_JMP | BPF_JEQ | BPF_X:
1096                 case BPF_JMP | BPF_JGE | BPF_K:
1097                 case BPF_JMP | BPF_JGE | BPF_X:
1098                 case BPF_JMP | BPF_JGT | BPF_K:
1099                 case BPF_JMP | BPF_JGT | BPF_X:
1100                 case BPF_JMP | BPF_JSET | BPF_K:
1101                 case BPF_JMP | BPF_JSET | BPF_X:
1102                         /* Both conditionals must be safe */
1103                         if (pc + ftest->jt + 1 >= flen ||
1104                             pc + ftest->jf + 1 >= flen)
1105                                 return -EINVAL;
1106                         break;
1107                 case BPF_LD | BPF_W | BPF_ABS:
1108                 case BPF_LD | BPF_H | BPF_ABS:
1109                 case BPF_LD | BPF_B | BPF_ABS:
1110                         anc_found = false;
1111                         if (bpf_anc_helper(ftest) & BPF_ANC)
1112                                 anc_found = true;
1113                         /* Ancillary operation unknown or unsupported */
1114                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1115                                 return -EINVAL;
1116                 }
1117         }
1118
1119         /* Last instruction must be a RET code */
1120         switch (filter[flen - 1].code) {
1121         case BPF_RET | BPF_K:
1122         case BPF_RET | BPF_A:
1123                 return check_load_and_stores(filter, flen);
1124         }
1125
1126         return -EINVAL;
1127 }
1128
1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1130                                       const struct sock_fprog *fprog)
1131 {
1132         unsigned int fsize = bpf_classic_proglen(fprog);
1133         struct sock_fprog_kern *fkprog;
1134
1135         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1136         if (!fp->orig_prog)
1137                 return -ENOMEM;
1138
1139         fkprog = fp->orig_prog;
1140         fkprog->len = fprog->len;
1141
1142         fkprog->filter = kmemdup(fp->insns, fsize,
1143                                  GFP_KERNEL | __GFP_NOWARN);
1144         if (!fkprog->filter) {
1145                 kfree(fp->orig_prog);
1146                 return -ENOMEM;
1147         }
1148
1149         return 0;
1150 }
1151
1152 static void bpf_release_orig_filter(struct bpf_prog *fp)
1153 {
1154         struct sock_fprog_kern *fprog = fp->orig_prog;
1155
1156         if (fprog) {
1157                 kfree(fprog->filter);
1158                 kfree(fprog);
1159         }
1160 }
1161
1162 static void __bpf_prog_release(struct bpf_prog *prog)
1163 {
1164         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1165                 bpf_prog_put(prog);
1166         } else {
1167                 bpf_release_orig_filter(prog);
1168                 bpf_prog_free(prog);
1169         }
1170 }
1171
1172 static void __sk_filter_release(struct sk_filter *fp)
1173 {
1174         __bpf_prog_release(fp->prog);
1175         kfree(fp);
1176 }
1177
1178 /**
1179  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1180  *      @rcu: rcu_head that contains the sk_filter to free
1181  */
1182 static void sk_filter_release_rcu(struct rcu_head *rcu)
1183 {
1184         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1185
1186         __sk_filter_release(fp);
1187 }
1188
1189 /**
1190  *      sk_filter_release - release a socket filter
1191  *      @fp: filter to remove
1192  *
1193  *      Remove a filter from a socket and release its resources.
1194  */
1195 static void sk_filter_release(struct sk_filter *fp)
1196 {
1197         if (refcount_dec_and_test(&fp->refcnt))
1198                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1199 }
1200
1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1202 {
1203         u32 filter_size = bpf_prog_size(fp->prog->len);
1204
1205         atomic_sub(filter_size, &sk->sk_omem_alloc);
1206         sk_filter_release(fp);
1207 }
1208
1209 /* try to charge the socket memory if there is space available
1210  * return true on success
1211  */
1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1213 {
1214         u32 filter_size = bpf_prog_size(fp->prog->len);
1215
1216         /* same check as in sock_kmalloc() */
1217         if (filter_size <= sysctl_optmem_max &&
1218             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1219                 atomic_add(filter_size, &sk->sk_omem_alloc);
1220                 return true;
1221         }
1222         return false;
1223 }
1224
1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1226 {
1227         if (!refcount_inc_not_zero(&fp->refcnt))
1228                 return false;
1229
1230         if (!__sk_filter_charge(sk, fp)) {
1231                 sk_filter_release(fp);
1232                 return false;
1233         }
1234         return true;
1235 }
1236
1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1238 {
1239         struct sock_filter *old_prog;
1240         struct bpf_prog *old_fp;
1241         int err, new_len, old_len = fp->len;
1242         bool seen_ld_abs = false;
1243
1244         /* We are free to overwrite insns et al right here as it
1245          * won't be used at this point in time anymore internally
1246          * after the migration to the internal BPF instruction
1247          * representation.
1248          */
1249         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250                      sizeof(struct bpf_insn));
1251
1252         /* Conversion cannot happen on overlapping memory areas,
1253          * so we need to keep the user BPF around until the 2nd
1254          * pass. At this time, the user BPF is stored in fp->insns.
1255          */
1256         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257                            GFP_KERNEL | __GFP_NOWARN);
1258         if (!old_prog) {
1259                 err = -ENOMEM;
1260                 goto out_err;
1261         }
1262
1263         /* 1st pass: calculate the new program length. */
1264         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1265                                  &seen_ld_abs);
1266         if (err)
1267                 goto out_err_free;
1268
1269         /* Expand fp for appending the new filter representation. */
1270         old_fp = fp;
1271         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1272         if (!fp) {
1273                 /* The old_fp is still around in case we couldn't
1274                  * allocate new memory, so uncharge on that one.
1275                  */
1276                 fp = old_fp;
1277                 err = -ENOMEM;
1278                 goto out_err_free;
1279         }
1280
1281         fp->len = new_len;
1282
1283         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1285                                  &seen_ld_abs);
1286         if (err)
1287                 /* 2nd bpf_convert_filter() can fail only if it fails
1288                  * to allocate memory, remapping must succeed. Note,
1289                  * that at this time old_fp has already been released
1290                  * by krealloc().
1291                  */
1292                 goto out_err_free;
1293
1294         fp = bpf_prog_select_runtime(fp, &err);
1295         if (err)
1296                 goto out_err_free;
1297
1298         kfree(old_prog);
1299         return fp;
1300
1301 out_err_free:
1302         kfree(old_prog);
1303 out_err:
1304         __bpf_prog_release(fp);
1305         return ERR_PTR(err);
1306 }
1307
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309                                            bpf_aux_classic_check_t trans)
1310 {
1311         int err;
1312
1313         fp->bpf_func = NULL;
1314         fp->jited = 0;
1315
1316         err = bpf_check_classic(fp->insns, fp->len);
1317         if (err) {
1318                 __bpf_prog_release(fp);
1319                 return ERR_PTR(err);
1320         }
1321
1322         /* There might be additional checks and transformations
1323          * needed on classic filters, f.e. in case of seccomp.
1324          */
1325         if (trans) {
1326                 err = trans(fp->insns, fp->len);
1327                 if (err) {
1328                         __bpf_prog_release(fp);
1329                         return ERR_PTR(err);
1330                 }
1331         }
1332
1333         /* Probe if we can JIT compile the filter and if so, do
1334          * the compilation of the filter.
1335          */
1336         bpf_jit_compile(fp);
1337
1338         /* JIT compiler couldn't process this filter, so do the
1339          * internal BPF translation for the optimized interpreter.
1340          */
1341         if (!fp->jited)
1342                 fp = bpf_migrate_filter(fp);
1343
1344         return fp;
1345 }
1346
1347 /**
1348  *      bpf_prog_create - create an unattached filter
1349  *      @pfp: the unattached filter that is created
1350  *      @fprog: the filter program
1351  *
1352  * Create a filter independent of any socket. We first run some
1353  * sanity checks on it to make sure it does not explode on us later.
1354  * If an error occurs or there is insufficient memory for the filter
1355  * a negative errno code is returned. On success the return is zero.
1356  */
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1358 {
1359         unsigned int fsize = bpf_classic_proglen(fprog);
1360         struct bpf_prog *fp;
1361
1362         /* Make sure new filter is there and in the right amounts. */
1363         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1364                 return -EINVAL;
1365
1366         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1367         if (!fp)
1368                 return -ENOMEM;
1369
1370         memcpy(fp->insns, fprog->filter, fsize);
1371
1372         fp->len = fprog->len;
1373         /* Since unattached filters are not copied back to user
1374          * space through sk_get_filter(), we do not need to hold
1375          * a copy here, and can spare us the work.
1376          */
1377         fp->orig_prog = NULL;
1378
1379         /* bpf_prepare_filter() already takes care of freeing
1380          * memory in case something goes wrong.
1381          */
1382         fp = bpf_prepare_filter(fp, NULL);
1383         if (IS_ERR(fp))
1384                 return PTR_ERR(fp);
1385
1386         *pfp = fp;
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1390
1391 /**
1392  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1393  *      @pfp: the unattached filter that is created
1394  *      @fprog: the filter program
1395  *      @trans: post-classic verifier transformation handler
1396  *      @save_orig: save classic BPF program
1397  *
1398  * This function effectively does the same as bpf_prog_create(), only
1399  * that it builds up its insns buffer from user space provided buffer.
1400  * It also allows for passing a bpf_aux_classic_check_t handler.
1401  */
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403                               bpf_aux_classic_check_t trans, bool save_orig)
1404 {
1405         unsigned int fsize = bpf_classic_proglen(fprog);
1406         struct bpf_prog *fp;
1407         int err;
1408
1409         /* Make sure new filter is there and in the right amounts. */
1410         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1411                 return -EINVAL;
1412
1413         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1414         if (!fp)
1415                 return -ENOMEM;
1416
1417         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418                 __bpf_prog_free(fp);
1419                 return -EFAULT;
1420         }
1421
1422         fp->len = fprog->len;
1423         fp->orig_prog = NULL;
1424
1425         if (save_orig) {
1426                 err = bpf_prog_store_orig_filter(fp, fprog);
1427                 if (err) {
1428                         __bpf_prog_free(fp);
1429                         return -ENOMEM;
1430                 }
1431         }
1432
1433         /* bpf_prepare_filter() already takes care of freeing
1434          * memory in case something goes wrong.
1435          */
1436         fp = bpf_prepare_filter(fp, trans);
1437         if (IS_ERR(fp))
1438                 return PTR_ERR(fp);
1439
1440         *pfp = fp;
1441         return 0;
1442 }
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1444
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1446 {
1447         __bpf_prog_release(fp);
1448 }
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1450
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1452 {
1453         struct sk_filter *fp, *old_fp;
1454
1455         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1456         if (!fp)
1457                 return -ENOMEM;
1458
1459         fp->prog = prog;
1460
1461         if (!__sk_filter_charge(sk, fp)) {
1462                 kfree(fp);
1463                 return -ENOMEM;
1464         }
1465         refcount_set(&fp->refcnt, 1);
1466
1467         old_fp = rcu_dereference_protected(sk->sk_filter,
1468                                            lockdep_sock_is_held(sk));
1469         rcu_assign_pointer(sk->sk_filter, fp);
1470
1471         if (old_fp)
1472                 sk_filter_uncharge(sk, old_fp);
1473
1474         return 0;
1475 }
1476
1477 static
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1479 {
1480         unsigned int fsize = bpf_classic_proglen(fprog);
1481         struct bpf_prog *prog;
1482         int err;
1483
1484         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485                 return ERR_PTR(-EPERM);
1486
1487         /* Make sure new filter is there and in the right amounts. */
1488         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489                 return ERR_PTR(-EINVAL);
1490
1491         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1492         if (!prog)
1493                 return ERR_PTR(-ENOMEM);
1494
1495         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496                 __bpf_prog_free(prog);
1497                 return ERR_PTR(-EFAULT);
1498         }
1499
1500         prog->len = fprog->len;
1501
1502         err = bpf_prog_store_orig_filter(prog, fprog);
1503         if (err) {
1504                 __bpf_prog_free(prog);
1505                 return ERR_PTR(-ENOMEM);
1506         }
1507
1508         /* bpf_prepare_filter() already takes care of freeing
1509          * memory in case something goes wrong.
1510          */
1511         return bpf_prepare_filter(prog, NULL);
1512 }
1513
1514 /**
1515  *      sk_attach_filter - attach a socket filter
1516  *      @fprog: the filter program
1517  *      @sk: the socket to use
1518  *
1519  * Attach the user's filter code. We first run some sanity checks on
1520  * it to make sure it does not explode on us later. If an error
1521  * occurs or there is insufficient memory for the filter a negative
1522  * errno code is returned. On success the return is zero.
1523  */
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1525 {
1526         struct bpf_prog *prog = __get_filter(fprog, sk);
1527         int err;
1528
1529         if (IS_ERR(prog))
1530                 return PTR_ERR(prog);
1531
1532         err = __sk_attach_prog(prog, sk);
1533         if (err < 0) {
1534                 __bpf_prog_release(prog);
1535                 return err;
1536         }
1537
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1541
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1543 {
1544         struct bpf_prog *prog = __get_filter(fprog, sk);
1545         int err;
1546
1547         if (IS_ERR(prog))
1548                 return PTR_ERR(prog);
1549
1550         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1551                 err = -ENOMEM;
1552         else
1553                 err = reuseport_attach_prog(sk, prog);
1554
1555         if (err)
1556                 __bpf_prog_release(prog);
1557
1558         return err;
1559 }
1560
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1562 {
1563         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564                 return ERR_PTR(-EPERM);
1565
1566         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1567 }
1568
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1570 {
1571         struct bpf_prog *prog = __get_bpf(ufd, sk);
1572         int err;
1573
1574         if (IS_ERR(prog))
1575                 return PTR_ERR(prog);
1576
1577         err = __sk_attach_prog(prog, sk);
1578         if (err < 0) {
1579                 bpf_prog_put(prog);
1580                 return err;
1581         }
1582
1583         return 0;
1584 }
1585
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1587 {
1588         struct bpf_prog *prog;
1589         int err;
1590
1591         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1592                 return -EPERM;
1593
1594         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595         if (PTR_ERR(prog) == -EINVAL)
1596                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1597         if (IS_ERR(prog))
1598                 return PTR_ERR(prog);
1599
1600         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602                  * bpf prog (e.g. sockmap).  It depends on the
1603                  * limitation imposed by bpf_prog_load().
1604                  * Hence, sysctl_optmem_max is not checked.
1605                  */
1606                 if ((sk->sk_type != SOCK_STREAM &&
1607                      sk->sk_type != SOCK_DGRAM) ||
1608                     (sk->sk_protocol != IPPROTO_UDP &&
1609                      sk->sk_protocol != IPPROTO_TCP) ||
1610                     (sk->sk_family != AF_INET &&
1611                      sk->sk_family != AF_INET6)) {
1612                         err = -ENOTSUPP;
1613                         goto err_prog_put;
1614                 }
1615         } else {
1616                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1618                         err = -ENOMEM;
1619                         goto err_prog_put;
1620                 }
1621         }
1622
1623         err = reuseport_attach_prog(sk, prog);
1624 err_prog_put:
1625         if (err)
1626                 bpf_prog_put(prog);
1627
1628         return err;
1629 }
1630
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1632 {
1633         if (!prog)
1634                 return;
1635
1636         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1637                 bpf_prog_put(prog);
1638         else
1639                 bpf_prog_destroy(prog);
1640 }
1641
1642 struct bpf_scratchpad {
1643         union {
1644                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645                 u8     buff[MAX_BPF_STACK];
1646         };
1647 };
1648
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1650
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652                                           unsigned int write_len)
1653 {
1654         return skb_ensure_writable(skb, write_len);
1655 }
1656
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658                                         unsigned int write_len)
1659 {
1660         int err = __bpf_try_make_writable(skb, write_len);
1661
1662         bpf_compute_data_pointers(skb);
1663         return err;
1664 }
1665
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1667 {
1668         return bpf_try_make_writable(skb, skb_headlen(skb));
1669 }
1670
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1672 {
1673         if (skb_at_tc_ingress(skb))
1674                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1675 }
1676
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1678 {
1679         if (skb_at_tc_ingress(skb))
1680                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 }
1682
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684            const void *, from, u32, len, u64, flags)
1685 {
1686         void *ptr;
1687
1688         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1689                 return -EINVAL;
1690         if (unlikely(offset > 0xffff))
1691                 return -EFAULT;
1692         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1693                 return -EFAULT;
1694
1695         ptr = skb->data + offset;
1696         if (flags & BPF_F_RECOMPUTE_CSUM)
1697                 __skb_postpull_rcsum(skb, ptr, len, offset);
1698
1699         memcpy(ptr, from, len);
1700
1701         if (flags & BPF_F_RECOMPUTE_CSUM)
1702                 __skb_postpush_rcsum(skb, ptr, len, offset);
1703         if (flags & BPF_F_INVALIDATE_HASH)
1704                 skb_clear_hash(skb);
1705
1706         return 0;
1707 }
1708
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710         .func           = bpf_skb_store_bytes,
1711         .gpl_only       = false,
1712         .ret_type       = RET_INTEGER,
1713         .arg1_type      = ARG_PTR_TO_CTX,
1714         .arg2_type      = ARG_ANYTHING,
1715         .arg3_type      = ARG_PTR_TO_MEM,
1716         .arg4_type      = ARG_CONST_SIZE,
1717         .arg5_type      = ARG_ANYTHING,
1718 };
1719
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721            void *, to, u32, len)
1722 {
1723         void *ptr;
1724
1725         if (unlikely(offset > 0xffff))
1726                 goto err_clear;
1727
1728         ptr = skb_header_pointer(skb, offset, len, to);
1729         if (unlikely(!ptr))
1730                 goto err_clear;
1731         if (ptr != to)
1732                 memcpy(to, ptr, len);
1733
1734         return 0;
1735 err_clear:
1736         memset(to, 0, len);
1737         return -EFAULT;
1738 }
1739
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741         .func           = bpf_skb_load_bytes,
1742         .gpl_only       = false,
1743         .ret_type       = RET_INTEGER,
1744         .arg1_type      = ARG_PTR_TO_CTX,
1745         .arg2_type      = ARG_ANYTHING,
1746         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1747         .arg4_type      = ARG_CONST_SIZE,
1748 };
1749
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751            const struct bpf_flow_dissector *, ctx, u32, offset,
1752            void *, to, u32, len)
1753 {
1754         void *ptr;
1755
1756         if (unlikely(offset > 0xffff))
1757                 goto err_clear;
1758
1759         if (unlikely(!ctx->skb))
1760                 goto err_clear;
1761
1762         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1763         if (unlikely(!ptr))
1764                 goto err_clear;
1765         if (ptr != to)
1766                 memcpy(to, ptr, len);
1767
1768         return 0;
1769 err_clear:
1770         memset(to, 0, len);
1771         return -EFAULT;
1772 }
1773
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775         .func           = bpf_flow_dissector_load_bytes,
1776         .gpl_only       = false,
1777         .ret_type       = RET_INTEGER,
1778         .arg1_type      = ARG_PTR_TO_CTX,
1779         .arg2_type      = ARG_ANYTHING,
1780         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1781         .arg4_type      = ARG_CONST_SIZE,
1782 };
1783
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785            u32, offset, void *, to, u32, len, u32, start_header)
1786 {
1787         u8 *end = skb_tail_pointer(skb);
1788         u8 *start, *ptr;
1789
1790         if (unlikely(offset > 0xffff))
1791                 goto err_clear;
1792
1793         switch (start_header) {
1794         case BPF_HDR_START_MAC:
1795                 if (unlikely(!skb_mac_header_was_set(skb)))
1796                         goto err_clear;
1797                 start = skb_mac_header(skb);
1798                 break;
1799         case BPF_HDR_START_NET:
1800                 start = skb_network_header(skb);
1801                 break;
1802         default:
1803                 goto err_clear;
1804         }
1805
1806         ptr = start + offset;
1807
1808         if (likely(ptr + len <= end)) {
1809                 memcpy(to, ptr, len);
1810                 return 0;
1811         }
1812
1813 err_clear:
1814         memset(to, 0, len);
1815         return -EFAULT;
1816 }
1817
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819         .func           = bpf_skb_load_bytes_relative,
1820         .gpl_only       = false,
1821         .ret_type       = RET_INTEGER,
1822         .arg1_type      = ARG_PTR_TO_CTX,
1823         .arg2_type      = ARG_ANYTHING,
1824         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1825         .arg4_type      = ARG_CONST_SIZE,
1826         .arg5_type      = ARG_ANYTHING,
1827 };
1828
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1830 {
1831         /* Idea is the following: should the needed direct read/write
1832          * test fail during runtime, we can pull in more data and redo
1833          * again, since implicitly, we invalidate previous checks here.
1834          *
1835          * Or, since we know how much we need to make read/writeable,
1836          * this can be done once at the program beginning for direct
1837          * access case. By this we overcome limitations of only current
1838          * headroom being accessible.
1839          */
1840         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1841 }
1842
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844         .func           = bpf_skb_pull_data,
1845         .gpl_only       = false,
1846         .ret_type       = RET_INTEGER,
1847         .arg1_type      = ARG_PTR_TO_CTX,
1848         .arg2_type      = ARG_ANYTHING,
1849 };
1850
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1852 {
1853         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1854 }
1855
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857         .func           = bpf_sk_fullsock,
1858         .gpl_only       = false,
1859         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1860         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1861 };
1862
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864                                            unsigned int write_len)
1865 {
1866         int err = __bpf_try_make_writable(skb, write_len);
1867
1868         bpf_compute_data_end_sk_skb(skb);
1869         return err;
1870 }
1871
1872 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1873 {
1874         /* Idea is the following: should the needed direct read/write
1875          * test fail during runtime, we can pull in more data and redo
1876          * again, since implicitly, we invalidate previous checks here.
1877          *
1878          * Or, since we know how much we need to make read/writeable,
1879          * this can be done once at the program beginning for direct
1880          * access case. By this we overcome limitations of only current
1881          * headroom being accessible.
1882          */
1883         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1884 }
1885
1886 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1887         .func           = sk_skb_pull_data,
1888         .gpl_only       = false,
1889         .ret_type       = RET_INTEGER,
1890         .arg1_type      = ARG_PTR_TO_CTX,
1891         .arg2_type      = ARG_ANYTHING,
1892 };
1893
1894 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1895            u64, from, u64, to, u64, flags)
1896 {
1897         __sum16 *ptr;
1898
1899         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1900                 return -EINVAL;
1901         if (unlikely(offset > 0xffff || offset & 1))
1902                 return -EFAULT;
1903         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1904                 return -EFAULT;
1905
1906         ptr = (__sum16 *)(skb->data + offset);
1907         switch (flags & BPF_F_HDR_FIELD_MASK) {
1908         case 0:
1909                 if (unlikely(from != 0))
1910                         return -EINVAL;
1911
1912                 csum_replace_by_diff(ptr, to);
1913                 break;
1914         case 2:
1915                 csum_replace2(ptr, from, to);
1916                 break;
1917         case 4:
1918                 csum_replace4(ptr, from, to);
1919                 break;
1920         default:
1921                 return -EINVAL;
1922         }
1923
1924         return 0;
1925 }
1926
1927 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1928         .func           = bpf_l3_csum_replace,
1929         .gpl_only       = false,
1930         .ret_type       = RET_INTEGER,
1931         .arg1_type      = ARG_PTR_TO_CTX,
1932         .arg2_type      = ARG_ANYTHING,
1933         .arg3_type      = ARG_ANYTHING,
1934         .arg4_type      = ARG_ANYTHING,
1935         .arg5_type      = ARG_ANYTHING,
1936 };
1937
1938 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1939            u64, from, u64, to, u64, flags)
1940 {
1941         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1942         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1943         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1944         __sum16 *ptr;
1945
1946         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1947                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1948                 return -EINVAL;
1949         if (unlikely(offset > 0xffff || offset & 1))
1950                 return -EFAULT;
1951         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1952                 return -EFAULT;
1953
1954         ptr = (__sum16 *)(skb->data + offset);
1955         if (is_mmzero && !do_mforce && !*ptr)
1956                 return 0;
1957
1958         switch (flags & BPF_F_HDR_FIELD_MASK) {
1959         case 0:
1960                 if (unlikely(from != 0))
1961                         return -EINVAL;
1962
1963                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1964                 break;
1965         case 2:
1966                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1967                 break;
1968         case 4:
1969                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1970                 break;
1971         default:
1972                 return -EINVAL;
1973         }
1974
1975         if (is_mmzero && !*ptr)
1976                 *ptr = CSUM_MANGLED_0;
1977         return 0;
1978 }
1979
1980 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1981         .func           = bpf_l4_csum_replace,
1982         .gpl_only       = false,
1983         .ret_type       = RET_INTEGER,
1984         .arg1_type      = ARG_PTR_TO_CTX,
1985         .arg2_type      = ARG_ANYTHING,
1986         .arg3_type      = ARG_ANYTHING,
1987         .arg4_type      = ARG_ANYTHING,
1988         .arg5_type      = ARG_ANYTHING,
1989 };
1990
1991 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1992            __be32 *, to, u32, to_size, __wsum, seed)
1993 {
1994         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1995         u32 diff_size = from_size + to_size;
1996         int i, j = 0;
1997
1998         /* This is quite flexible, some examples:
1999          *
2000          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2001          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2002          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2003          *
2004          * Even for diffing, from_size and to_size don't need to be equal.
2005          */
2006         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2007                      diff_size > sizeof(sp->diff)))
2008                 return -EINVAL;
2009
2010         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2011                 sp->diff[j] = ~from[i];
2012         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2013                 sp->diff[j] = to[i];
2014
2015         return csum_partial(sp->diff, diff_size, seed);
2016 }
2017
2018 static const struct bpf_func_proto bpf_csum_diff_proto = {
2019         .func           = bpf_csum_diff,
2020         .gpl_only       = false,
2021         .pkt_access     = true,
2022         .ret_type       = RET_INTEGER,
2023         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
2024         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2025         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
2026         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2027         .arg5_type      = ARG_ANYTHING,
2028 };
2029
2030 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2031 {
2032         /* The interface is to be used in combination with bpf_csum_diff()
2033          * for direct packet writes. csum rotation for alignment as well
2034          * as emulating csum_sub() can be done from the eBPF program.
2035          */
2036         if (skb->ip_summed == CHECKSUM_COMPLETE)
2037                 return (skb->csum = csum_add(skb->csum, csum));
2038
2039         return -ENOTSUPP;
2040 }
2041
2042 static const struct bpf_func_proto bpf_csum_update_proto = {
2043         .func           = bpf_csum_update,
2044         .gpl_only       = false,
2045         .ret_type       = RET_INTEGER,
2046         .arg1_type      = ARG_PTR_TO_CTX,
2047         .arg2_type      = ARG_ANYTHING,
2048 };
2049
2050 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2051 {
2052         /* The interface is to be used in combination with bpf_skb_adjust_room()
2053          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2054          * is passed as flags, for example.
2055          */
2056         switch (level) {
2057         case BPF_CSUM_LEVEL_INC:
2058                 __skb_incr_checksum_unnecessary(skb);
2059                 break;
2060         case BPF_CSUM_LEVEL_DEC:
2061                 __skb_decr_checksum_unnecessary(skb);
2062                 break;
2063         case BPF_CSUM_LEVEL_RESET:
2064                 __skb_reset_checksum_unnecessary(skb);
2065                 break;
2066         case BPF_CSUM_LEVEL_QUERY:
2067                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2068                        skb->csum_level : -EACCES;
2069         default:
2070                 return -EINVAL;
2071         }
2072
2073         return 0;
2074 }
2075
2076 static const struct bpf_func_proto bpf_csum_level_proto = {
2077         .func           = bpf_csum_level,
2078         .gpl_only       = false,
2079         .ret_type       = RET_INTEGER,
2080         .arg1_type      = ARG_PTR_TO_CTX,
2081         .arg2_type      = ARG_ANYTHING,
2082 };
2083
2084 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2085 {
2086         return dev_forward_skb_nomtu(dev, skb);
2087 }
2088
2089 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2090                                       struct sk_buff *skb)
2091 {
2092         int ret = ____dev_forward_skb(dev, skb, false);
2093
2094         if (likely(!ret)) {
2095                 skb->dev = dev;
2096                 ret = netif_rx(skb);
2097         }
2098
2099         return ret;
2100 }
2101
2102 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2103 {
2104         int ret;
2105
2106         if (dev_xmit_recursion()) {
2107                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2108                 kfree_skb(skb);
2109                 return -ENETDOWN;
2110         }
2111
2112         skb->dev = dev;
2113         skb->tstamp = 0;
2114
2115         dev_xmit_recursion_inc();
2116         ret = dev_queue_xmit(skb);
2117         dev_xmit_recursion_dec();
2118
2119         return ret;
2120 }
2121
2122 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2123                                  u32 flags)
2124 {
2125         unsigned int mlen = skb_network_offset(skb);
2126
2127         if (mlen) {
2128                 __skb_pull(skb, mlen);
2129
2130                 /* At ingress, the mac header has already been pulled once.
2131                  * At egress, skb_pospull_rcsum has to be done in case that
2132                  * the skb is originated from ingress (i.e. a forwarded skb)
2133                  * to ensure that rcsum starts at net header.
2134                  */
2135                 if (!skb_at_tc_ingress(skb))
2136                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2137         }
2138         skb_pop_mac_header(skb);
2139         skb_reset_mac_len(skb);
2140         return flags & BPF_F_INGRESS ?
2141                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2142 }
2143
2144 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2145                                  u32 flags)
2146 {
2147         /* Verify that a link layer header is carried */
2148         if (unlikely(skb->mac_header >= skb->network_header)) {
2149                 kfree_skb(skb);
2150                 return -ERANGE;
2151         }
2152
2153         bpf_push_mac_rcsum(skb);
2154         return flags & BPF_F_INGRESS ?
2155                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2156 }
2157
2158 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2159                           u32 flags)
2160 {
2161         if (dev_is_mac_header_xmit(dev))
2162                 return __bpf_redirect_common(skb, dev, flags);
2163         else
2164                 return __bpf_redirect_no_mac(skb, dev, flags);
2165 }
2166
2167 #if IS_ENABLED(CONFIG_IPV6)
2168 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2169                             struct net_device *dev, struct bpf_nh_params *nh)
2170 {
2171         u32 hh_len = LL_RESERVED_SPACE(dev);
2172         const struct in6_addr *nexthop;
2173         struct dst_entry *dst = NULL;
2174         struct neighbour *neigh;
2175
2176         if (dev_xmit_recursion()) {
2177                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2178                 goto out_drop;
2179         }
2180
2181         skb->dev = dev;
2182         skb->tstamp = 0;
2183
2184         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2185                 struct sk_buff *skb2;
2186
2187                 skb2 = skb_realloc_headroom(skb, hh_len);
2188                 if (unlikely(!skb2)) {
2189                         kfree_skb(skb);
2190                         return -ENOMEM;
2191                 }
2192                 if (skb->sk)
2193                         skb_set_owner_w(skb2, skb->sk);
2194                 consume_skb(skb);
2195                 skb = skb2;
2196         }
2197
2198         rcu_read_lock_bh();
2199         if (!nh) {
2200                 dst = skb_dst(skb);
2201                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2202                                       &ipv6_hdr(skb)->daddr);
2203         } else {
2204                 nexthop = &nh->ipv6_nh;
2205         }
2206         neigh = ip_neigh_gw6(dev, nexthop);
2207         if (likely(!IS_ERR(neigh))) {
2208                 int ret;
2209
2210                 sock_confirm_neigh(skb, neigh);
2211                 dev_xmit_recursion_inc();
2212                 ret = neigh_output(neigh, skb, false);
2213                 dev_xmit_recursion_dec();
2214                 rcu_read_unlock_bh();
2215                 return ret;
2216         }
2217         rcu_read_unlock_bh();
2218         if (dst)
2219                 IP6_INC_STATS(dev_net(dst->dev),
2220                               ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2221 out_drop:
2222         kfree_skb(skb);
2223         return -ENETDOWN;
2224 }
2225
2226 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2227                                    struct bpf_nh_params *nh)
2228 {
2229         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2230         struct net *net = dev_net(dev);
2231         int err, ret = NET_XMIT_DROP;
2232
2233         if (!nh) {
2234                 struct dst_entry *dst;
2235                 struct flowi6 fl6 = {
2236                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2237                         .flowi6_mark  = skb->mark,
2238                         .flowlabel    = ip6_flowinfo(ip6h),
2239                         .flowi6_oif   = dev->ifindex,
2240                         .flowi6_proto = ip6h->nexthdr,
2241                         .daddr        = ip6h->daddr,
2242                         .saddr        = ip6h->saddr,
2243                 };
2244
2245                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2246                 if (IS_ERR(dst))
2247                         goto out_drop;
2248
2249                 skb_dst_set(skb, dst);
2250         } else if (nh->nh_family != AF_INET6) {
2251                 goto out_drop;
2252         }
2253
2254         err = bpf_out_neigh_v6(net, skb, dev, nh);
2255         if (unlikely(net_xmit_eval(err)))
2256                 dev->stats.tx_errors++;
2257         else
2258                 ret = NET_XMIT_SUCCESS;
2259         goto out_xmit;
2260 out_drop:
2261         dev->stats.tx_errors++;
2262         kfree_skb(skb);
2263 out_xmit:
2264         return ret;
2265 }
2266 #else
2267 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2268                                    struct bpf_nh_params *nh)
2269 {
2270         kfree_skb(skb);
2271         return NET_XMIT_DROP;
2272 }
2273 #endif /* CONFIG_IPV6 */
2274
2275 #if IS_ENABLED(CONFIG_INET)
2276 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2277                             struct net_device *dev, struct bpf_nh_params *nh)
2278 {
2279         u32 hh_len = LL_RESERVED_SPACE(dev);
2280         struct neighbour *neigh;
2281         bool is_v6gw = false;
2282
2283         if (dev_xmit_recursion()) {
2284                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2285                 goto out_drop;
2286         }
2287
2288         skb->dev = dev;
2289         skb->tstamp = 0;
2290
2291         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2292                 struct sk_buff *skb2;
2293
2294                 skb2 = skb_realloc_headroom(skb, hh_len);
2295                 if (unlikely(!skb2)) {
2296                         kfree_skb(skb);
2297                         return -ENOMEM;
2298                 }
2299                 if (skb->sk)
2300                         skb_set_owner_w(skb2, skb->sk);
2301                 consume_skb(skb);
2302                 skb = skb2;
2303         }
2304
2305         rcu_read_lock_bh();
2306         if (!nh) {
2307                 struct dst_entry *dst = skb_dst(skb);
2308                 struct rtable *rt = container_of(dst, struct rtable, dst);
2309
2310                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2311         } else if (nh->nh_family == AF_INET6) {
2312                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2313                 is_v6gw = true;
2314         } else if (nh->nh_family == AF_INET) {
2315                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2316         } else {
2317                 rcu_read_unlock_bh();
2318                 goto out_drop;
2319         }
2320
2321         if (likely(!IS_ERR(neigh))) {
2322                 int ret;
2323
2324                 sock_confirm_neigh(skb, neigh);
2325                 dev_xmit_recursion_inc();
2326                 ret = neigh_output(neigh, skb, is_v6gw);
2327                 dev_xmit_recursion_dec();
2328                 rcu_read_unlock_bh();
2329                 return ret;
2330         }
2331         rcu_read_unlock_bh();
2332 out_drop:
2333         kfree_skb(skb);
2334         return -ENETDOWN;
2335 }
2336
2337 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2338                                    struct bpf_nh_params *nh)
2339 {
2340         const struct iphdr *ip4h = ip_hdr(skb);
2341         struct net *net = dev_net(dev);
2342         int err, ret = NET_XMIT_DROP;
2343
2344         if (!nh) {
2345                 struct flowi4 fl4 = {
2346                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2347                         .flowi4_mark  = skb->mark,
2348                         .flowi4_tos   = RT_TOS(ip4h->tos),
2349                         .flowi4_oif   = dev->ifindex,
2350                         .flowi4_proto = ip4h->protocol,
2351                         .daddr        = ip4h->daddr,
2352                         .saddr        = ip4h->saddr,
2353                 };
2354                 struct rtable *rt;
2355
2356                 rt = ip_route_output_flow(net, &fl4, NULL);
2357                 if (IS_ERR(rt))
2358                         goto out_drop;
2359                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2360                         ip_rt_put(rt);
2361                         goto out_drop;
2362                 }
2363
2364                 skb_dst_set(skb, &rt->dst);
2365         }
2366
2367         err = bpf_out_neigh_v4(net, skb, dev, nh);
2368         if (unlikely(net_xmit_eval(err)))
2369                 dev->stats.tx_errors++;
2370         else
2371                 ret = NET_XMIT_SUCCESS;
2372         goto out_xmit;
2373 out_drop:
2374         dev->stats.tx_errors++;
2375         kfree_skb(skb);
2376 out_xmit:
2377         return ret;
2378 }
2379 #else
2380 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2381                                    struct bpf_nh_params *nh)
2382 {
2383         kfree_skb(skb);
2384         return NET_XMIT_DROP;
2385 }
2386 #endif /* CONFIG_INET */
2387
2388 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2389                                 struct bpf_nh_params *nh)
2390 {
2391         struct ethhdr *ethh = eth_hdr(skb);
2392
2393         if (unlikely(skb->mac_header >= skb->network_header))
2394                 goto out;
2395         bpf_push_mac_rcsum(skb);
2396         if (is_multicast_ether_addr(ethh->h_dest))
2397                 goto out;
2398
2399         skb_pull(skb, sizeof(*ethh));
2400         skb_unset_mac_header(skb);
2401         skb_reset_network_header(skb);
2402
2403         if (skb->protocol == htons(ETH_P_IP))
2404                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2405         else if (skb->protocol == htons(ETH_P_IPV6))
2406                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2407 out:
2408         kfree_skb(skb);
2409         return -ENOTSUPP;
2410 }
2411
2412 /* Internal, non-exposed redirect flags. */
2413 enum {
2414         BPF_F_NEIGH     = (1ULL << 1),
2415         BPF_F_PEER      = (1ULL << 2),
2416         BPF_F_NEXTHOP   = (1ULL << 3),
2417 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2418 };
2419
2420 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2421 {
2422         struct net_device *dev;
2423         struct sk_buff *clone;
2424         int ret;
2425
2426         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2427                 return -EINVAL;
2428
2429         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2430         if (unlikely(!dev))
2431                 return -EINVAL;
2432
2433         clone = skb_clone(skb, GFP_ATOMIC);
2434         if (unlikely(!clone))
2435                 return -ENOMEM;
2436
2437         /* For direct write, we need to keep the invariant that the skbs
2438          * we're dealing with need to be uncloned. Should uncloning fail
2439          * here, we need to free the just generated clone to unclone once
2440          * again.
2441          */
2442         ret = bpf_try_make_head_writable(skb);
2443         if (unlikely(ret)) {
2444                 kfree_skb(clone);
2445                 return -ENOMEM;
2446         }
2447
2448         return __bpf_redirect(clone, dev, flags);
2449 }
2450
2451 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2452         .func           = bpf_clone_redirect,
2453         .gpl_only       = false,
2454         .ret_type       = RET_INTEGER,
2455         .arg1_type      = ARG_PTR_TO_CTX,
2456         .arg2_type      = ARG_ANYTHING,
2457         .arg3_type      = ARG_ANYTHING,
2458 };
2459
2460 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2461 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2462
2463 int skb_do_redirect(struct sk_buff *skb)
2464 {
2465         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2466         struct net *net = dev_net(skb->dev);
2467         struct net_device *dev;
2468         u32 flags = ri->flags;
2469
2470         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2471         ri->tgt_index = 0;
2472         ri->flags = 0;
2473         if (unlikely(!dev))
2474                 goto out_drop;
2475         if (flags & BPF_F_PEER) {
2476                 const struct net_device_ops *ops = dev->netdev_ops;
2477
2478                 if (unlikely(!ops->ndo_get_peer_dev ||
2479                              !skb_at_tc_ingress(skb)))
2480                         goto out_drop;
2481                 dev = ops->ndo_get_peer_dev(dev);
2482                 if (unlikely(!dev ||
2483                              !(dev->flags & IFF_UP) ||
2484                              net_eq(net, dev_net(dev))))
2485                         goto out_drop;
2486                 skb->dev = dev;
2487                 return -EAGAIN;
2488         }
2489         return flags & BPF_F_NEIGH ?
2490                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2491                                     &ri->nh : NULL) :
2492                __bpf_redirect(skb, dev, flags);
2493 out_drop:
2494         kfree_skb(skb);
2495         return -EINVAL;
2496 }
2497
2498 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2499 {
2500         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2501
2502         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2503                 return TC_ACT_SHOT;
2504
2505         ri->flags = flags;
2506         ri->tgt_index = ifindex;
2507
2508         return TC_ACT_REDIRECT;
2509 }
2510
2511 static const struct bpf_func_proto bpf_redirect_proto = {
2512         .func           = bpf_redirect,
2513         .gpl_only       = false,
2514         .ret_type       = RET_INTEGER,
2515         .arg1_type      = ARG_ANYTHING,
2516         .arg2_type      = ARG_ANYTHING,
2517 };
2518
2519 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2520 {
2521         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2522
2523         if (unlikely(flags))
2524                 return TC_ACT_SHOT;
2525
2526         ri->flags = BPF_F_PEER;
2527         ri->tgt_index = ifindex;
2528
2529         return TC_ACT_REDIRECT;
2530 }
2531
2532 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2533         .func           = bpf_redirect_peer,
2534         .gpl_only       = false,
2535         .ret_type       = RET_INTEGER,
2536         .arg1_type      = ARG_ANYTHING,
2537         .arg2_type      = ARG_ANYTHING,
2538 };
2539
2540 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2541            int, plen, u64, flags)
2542 {
2543         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2544
2545         if (unlikely((plen && plen < sizeof(*params)) || flags))
2546                 return TC_ACT_SHOT;
2547
2548         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2549         ri->tgt_index = ifindex;
2550
2551         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2552         if (plen)
2553                 memcpy(&ri->nh, params, sizeof(ri->nh));
2554
2555         return TC_ACT_REDIRECT;
2556 }
2557
2558 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2559         .func           = bpf_redirect_neigh,
2560         .gpl_only       = false,
2561         .ret_type       = RET_INTEGER,
2562         .arg1_type      = ARG_ANYTHING,
2563         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
2564         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2565         .arg4_type      = ARG_ANYTHING,
2566 };
2567
2568 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2569 {
2570         msg->apply_bytes = bytes;
2571         return 0;
2572 }
2573
2574 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2575         .func           = bpf_msg_apply_bytes,
2576         .gpl_only       = false,
2577         .ret_type       = RET_INTEGER,
2578         .arg1_type      = ARG_PTR_TO_CTX,
2579         .arg2_type      = ARG_ANYTHING,
2580 };
2581
2582 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2583 {
2584         msg->cork_bytes = bytes;
2585         return 0;
2586 }
2587
2588 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2589         .func           = bpf_msg_cork_bytes,
2590         .gpl_only       = false,
2591         .ret_type       = RET_INTEGER,
2592         .arg1_type      = ARG_PTR_TO_CTX,
2593         .arg2_type      = ARG_ANYTHING,
2594 };
2595
2596 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2597            u32, end, u64, flags)
2598 {
2599         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2600         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2601         struct scatterlist *sge;
2602         u8 *raw, *to, *from;
2603         struct page *page;
2604
2605         if (unlikely(flags || end <= start))
2606                 return -EINVAL;
2607
2608         /* First find the starting scatterlist element */
2609         i = msg->sg.start;
2610         do {
2611                 offset += len;
2612                 len = sk_msg_elem(msg, i)->length;
2613                 if (start < offset + len)
2614                         break;
2615                 sk_msg_iter_var_next(i);
2616         } while (i != msg->sg.end);
2617
2618         if (unlikely(start >= offset + len))
2619                 return -EINVAL;
2620
2621         first_sge = i;
2622         /* The start may point into the sg element so we need to also
2623          * account for the headroom.
2624          */
2625         bytes_sg_total = start - offset + bytes;
2626         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2627                 goto out;
2628
2629         /* At this point we need to linearize multiple scatterlist
2630          * elements or a single shared page. Either way we need to
2631          * copy into a linear buffer exclusively owned by BPF. Then
2632          * place the buffer in the scatterlist and fixup the original
2633          * entries by removing the entries now in the linear buffer
2634          * and shifting the remaining entries. For now we do not try
2635          * to copy partial entries to avoid complexity of running out
2636          * of sg_entry slots. The downside is reading a single byte
2637          * will copy the entire sg entry.
2638          */
2639         do {
2640                 copy += sk_msg_elem(msg, i)->length;
2641                 sk_msg_iter_var_next(i);
2642                 if (bytes_sg_total <= copy)
2643                         break;
2644         } while (i != msg->sg.end);
2645         last_sge = i;
2646
2647         if (unlikely(bytes_sg_total > copy))
2648                 return -EINVAL;
2649
2650         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2651                            get_order(copy));
2652         if (unlikely(!page))
2653                 return -ENOMEM;
2654
2655         raw = page_address(page);
2656         i = first_sge;
2657         do {
2658                 sge = sk_msg_elem(msg, i);
2659                 from = sg_virt(sge);
2660                 len = sge->length;
2661                 to = raw + poffset;
2662
2663                 memcpy(to, from, len);
2664                 poffset += len;
2665                 sge->length = 0;
2666                 put_page(sg_page(sge));
2667
2668                 sk_msg_iter_var_next(i);
2669         } while (i != last_sge);
2670
2671         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2672
2673         /* To repair sg ring we need to shift entries. If we only
2674          * had a single entry though we can just replace it and
2675          * be done. Otherwise walk the ring and shift the entries.
2676          */
2677         WARN_ON_ONCE(last_sge == first_sge);
2678         shift = last_sge > first_sge ?
2679                 last_sge - first_sge - 1 :
2680                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2681         if (!shift)
2682                 goto out;
2683
2684         i = first_sge;
2685         sk_msg_iter_var_next(i);
2686         do {
2687                 u32 move_from;
2688
2689                 if (i + shift >= NR_MSG_FRAG_IDS)
2690                         move_from = i + shift - NR_MSG_FRAG_IDS;
2691                 else
2692                         move_from = i + shift;
2693                 if (move_from == msg->sg.end)
2694                         break;
2695
2696                 msg->sg.data[i] = msg->sg.data[move_from];
2697                 msg->sg.data[move_from].length = 0;
2698                 msg->sg.data[move_from].page_link = 0;
2699                 msg->sg.data[move_from].offset = 0;
2700                 sk_msg_iter_var_next(i);
2701         } while (1);
2702
2703         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2704                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2705                       msg->sg.end - shift;
2706 out:
2707         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2708         msg->data_end = msg->data + bytes;
2709         return 0;
2710 }
2711
2712 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2713         .func           = bpf_msg_pull_data,
2714         .gpl_only       = false,
2715         .ret_type       = RET_INTEGER,
2716         .arg1_type      = ARG_PTR_TO_CTX,
2717         .arg2_type      = ARG_ANYTHING,
2718         .arg3_type      = ARG_ANYTHING,
2719         .arg4_type      = ARG_ANYTHING,
2720 };
2721
2722 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2723            u32, len, u64, flags)
2724 {
2725         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2726         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2727         u8 *raw, *to, *from;
2728         struct page *page;
2729
2730         if (unlikely(flags))
2731                 return -EINVAL;
2732
2733         /* First find the starting scatterlist element */
2734         i = msg->sg.start;
2735         do {
2736                 offset += l;
2737                 l = sk_msg_elem(msg, i)->length;
2738
2739                 if (start < offset + l)
2740                         break;
2741                 sk_msg_iter_var_next(i);
2742         } while (i != msg->sg.end);
2743
2744         if (start >= offset + l)
2745                 return -EINVAL;
2746
2747         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2748
2749         /* If no space available will fallback to copy, we need at
2750          * least one scatterlist elem available to push data into
2751          * when start aligns to the beginning of an element or two
2752          * when it falls inside an element. We handle the start equals
2753          * offset case because its the common case for inserting a
2754          * header.
2755          */
2756         if (!space || (space == 1 && start != offset))
2757                 copy = msg->sg.data[i].length;
2758
2759         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2760                            get_order(copy + len));
2761         if (unlikely(!page))
2762                 return -ENOMEM;
2763
2764         if (copy) {
2765                 int front, back;
2766
2767                 raw = page_address(page);
2768
2769                 psge = sk_msg_elem(msg, i);
2770                 front = start - offset;
2771                 back = psge->length - front;
2772                 from = sg_virt(psge);
2773
2774                 if (front)
2775                         memcpy(raw, from, front);
2776
2777                 if (back) {
2778                         from += front;
2779                         to = raw + front + len;
2780
2781                         memcpy(to, from, back);
2782                 }
2783
2784                 put_page(sg_page(psge));
2785         } else if (start - offset) {
2786                 psge = sk_msg_elem(msg, i);
2787                 rsge = sk_msg_elem_cpy(msg, i);
2788
2789                 psge->length = start - offset;
2790                 rsge.length -= psge->length;
2791                 rsge.offset += start;
2792
2793                 sk_msg_iter_var_next(i);
2794                 sg_unmark_end(psge);
2795                 sg_unmark_end(&rsge);
2796                 sk_msg_iter_next(msg, end);
2797         }
2798
2799         /* Slot(s) to place newly allocated data */
2800         new = i;
2801
2802         /* Shift one or two slots as needed */
2803         if (!copy) {
2804                 sge = sk_msg_elem_cpy(msg, i);
2805
2806                 sk_msg_iter_var_next(i);
2807                 sg_unmark_end(&sge);
2808                 sk_msg_iter_next(msg, end);
2809
2810                 nsge = sk_msg_elem_cpy(msg, i);
2811                 if (rsge.length) {
2812                         sk_msg_iter_var_next(i);
2813                         nnsge = sk_msg_elem_cpy(msg, i);
2814                 }
2815
2816                 while (i != msg->sg.end) {
2817                         msg->sg.data[i] = sge;
2818                         sge = nsge;
2819                         sk_msg_iter_var_next(i);
2820                         if (rsge.length) {
2821                                 nsge = nnsge;
2822                                 nnsge = sk_msg_elem_cpy(msg, i);
2823                         } else {
2824                                 nsge = sk_msg_elem_cpy(msg, i);
2825                         }
2826                 }
2827         }
2828
2829         /* Place newly allocated data buffer */
2830         sk_mem_charge(msg->sk, len);
2831         msg->sg.size += len;
2832         __clear_bit(new, &msg->sg.copy);
2833         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2834         if (rsge.length) {
2835                 get_page(sg_page(&rsge));
2836                 sk_msg_iter_var_next(new);
2837                 msg->sg.data[new] = rsge;
2838         }
2839
2840         sk_msg_compute_data_pointers(msg);
2841         return 0;
2842 }
2843
2844 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2845         .func           = bpf_msg_push_data,
2846         .gpl_only       = false,
2847         .ret_type       = RET_INTEGER,
2848         .arg1_type      = ARG_PTR_TO_CTX,
2849         .arg2_type      = ARG_ANYTHING,
2850         .arg3_type      = ARG_ANYTHING,
2851         .arg4_type      = ARG_ANYTHING,
2852 };
2853
2854 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2855 {
2856         int prev;
2857
2858         do {
2859                 prev = i;
2860                 sk_msg_iter_var_next(i);
2861                 msg->sg.data[prev] = msg->sg.data[i];
2862         } while (i != msg->sg.end);
2863
2864         sk_msg_iter_prev(msg, end);
2865 }
2866
2867 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2868 {
2869         struct scatterlist tmp, sge;
2870
2871         sk_msg_iter_next(msg, end);
2872         sge = sk_msg_elem_cpy(msg, i);
2873         sk_msg_iter_var_next(i);
2874         tmp = sk_msg_elem_cpy(msg, i);
2875
2876         while (i != msg->sg.end) {
2877                 msg->sg.data[i] = sge;
2878                 sk_msg_iter_var_next(i);
2879                 sge = tmp;
2880                 tmp = sk_msg_elem_cpy(msg, i);
2881         }
2882 }
2883
2884 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2885            u32, len, u64, flags)
2886 {
2887         u32 i = 0, l = 0, space, offset = 0;
2888         u64 last = start + len;
2889         int pop;
2890
2891         if (unlikely(flags))
2892                 return -EINVAL;
2893
2894         /* First find the starting scatterlist element */
2895         i = msg->sg.start;
2896         do {
2897                 offset += l;
2898                 l = sk_msg_elem(msg, i)->length;
2899
2900                 if (start < offset + l)
2901                         break;
2902                 sk_msg_iter_var_next(i);
2903         } while (i != msg->sg.end);
2904
2905         /* Bounds checks: start and pop must be inside message */
2906         if (start >= offset + l || last >= msg->sg.size)
2907                 return -EINVAL;
2908
2909         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2910
2911         pop = len;
2912         /* --------------| offset
2913          * -| start      |-------- len -------|
2914          *
2915          *  |----- a ----|-------- pop -------|----- b ----|
2916          *  |______________________________________________| length
2917          *
2918          *
2919          * a:   region at front of scatter element to save
2920          * b:   region at back of scatter element to save when length > A + pop
2921          * pop: region to pop from element, same as input 'pop' here will be
2922          *      decremented below per iteration.
2923          *
2924          * Two top-level cases to handle when start != offset, first B is non
2925          * zero and second B is zero corresponding to when a pop includes more
2926          * than one element.
2927          *
2928          * Then if B is non-zero AND there is no space allocate space and
2929          * compact A, B regions into page. If there is space shift ring to
2930          * the rigth free'ing the next element in ring to place B, leaving
2931          * A untouched except to reduce length.
2932          */
2933         if (start != offset) {
2934                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2935                 int a = start;
2936                 int b = sge->length - pop - a;
2937
2938                 sk_msg_iter_var_next(i);
2939
2940                 if (pop < sge->length - a) {
2941                         if (space) {
2942                                 sge->length = a;
2943                                 sk_msg_shift_right(msg, i);
2944                                 nsge = sk_msg_elem(msg, i);
2945                                 get_page(sg_page(sge));
2946                                 sg_set_page(nsge,
2947                                             sg_page(sge),
2948                                             b, sge->offset + pop + a);
2949                         } else {
2950                                 struct page *page, *orig;
2951                                 u8 *to, *from;
2952
2953                                 page = alloc_pages(__GFP_NOWARN |
2954                                                    __GFP_COMP   | GFP_ATOMIC,
2955                                                    get_order(a + b));
2956                                 if (unlikely(!page))
2957                                         return -ENOMEM;
2958
2959                                 sge->length = a;
2960                                 orig = sg_page(sge);
2961                                 from = sg_virt(sge);
2962                                 to = page_address(page);
2963                                 memcpy(to, from, a);
2964                                 memcpy(to + a, from + a + pop, b);
2965                                 sg_set_page(sge, page, a + b, 0);
2966                                 put_page(orig);
2967                         }
2968                         pop = 0;
2969                 } else if (pop >= sge->length - a) {
2970                         pop -= (sge->length - a);
2971                         sge->length = a;
2972                 }
2973         }
2974
2975         /* From above the current layout _must_ be as follows,
2976          *
2977          * -| offset
2978          * -| start
2979          *
2980          *  |---- pop ---|---------------- b ------------|
2981          *  |____________________________________________| length
2982          *
2983          * Offset and start of the current msg elem are equal because in the
2984          * previous case we handled offset != start and either consumed the
2985          * entire element and advanced to the next element OR pop == 0.
2986          *
2987          * Two cases to handle here are first pop is less than the length
2988          * leaving some remainder b above. Simply adjust the element's layout
2989          * in this case. Or pop >= length of the element so that b = 0. In this
2990          * case advance to next element decrementing pop.
2991          */
2992         while (pop) {
2993                 struct scatterlist *sge = sk_msg_elem(msg, i);
2994
2995                 if (pop < sge->length) {
2996                         sge->length -= pop;
2997                         sge->offset += pop;
2998                         pop = 0;
2999                 } else {
3000                         pop -= sge->length;
3001                         sk_msg_shift_left(msg, i);
3002                 }
3003                 sk_msg_iter_var_next(i);
3004         }
3005
3006         sk_mem_uncharge(msg->sk, len - pop);
3007         msg->sg.size -= (len - pop);
3008         sk_msg_compute_data_pointers(msg);
3009         return 0;
3010 }
3011
3012 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3013         .func           = bpf_msg_pop_data,
3014         .gpl_only       = false,
3015         .ret_type       = RET_INTEGER,
3016         .arg1_type      = ARG_PTR_TO_CTX,
3017         .arg2_type      = ARG_ANYTHING,
3018         .arg3_type      = ARG_ANYTHING,
3019         .arg4_type      = ARG_ANYTHING,
3020 };
3021
3022 #ifdef CONFIG_CGROUP_NET_CLASSID
3023 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3024 {
3025         return __task_get_classid(current);
3026 }
3027
3028 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3029         .func           = bpf_get_cgroup_classid_curr,
3030         .gpl_only       = false,
3031         .ret_type       = RET_INTEGER,
3032 };
3033
3034 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3035 {
3036         struct sock *sk = skb_to_full_sk(skb);
3037
3038         if (!sk || !sk_fullsock(sk))
3039                 return 0;
3040
3041         return sock_cgroup_classid(&sk->sk_cgrp_data);
3042 }
3043
3044 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3045         .func           = bpf_skb_cgroup_classid,
3046         .gpl_only       = false,
3047         .ret_type       = RET_INTEGER,
3048         .arg1_type      = ARG_PTR_TO_CTX,
3049 };
3050 #endif
3051
3052 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3053 {
3054         return task_get_classid(skb);
3055 }
3056
3057 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3058         .func           = bpf_get_cgroup_classid,
3059         .gpl_only       = false,
3060         .ret_type       = RET_INTEGER,
3061         .arg1_type      = ARG_PTR_TO_CTX,
3062 };
3063
3064 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3065 {
3066         return dst_tclassid(skb);
3067 }
3068
3069 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3070         .func           = bpf_get_route_realm,
3071         .gpl_only       = false,
3072         .ret_type       = RET_INTEGER,
3073         .arg1_type      = ARG_PTR_TO_CTX,
3074 };
3075
3076 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3077 {
3078         /* If skb_clear_hash() was called due to mangling, we can
3079          * trigger SW recalculation here. Later access to hash
3080          * can then use the inline skb->hash via context directly
3081          * instead of calling this helper again.
3082          */
3083         return skb_get_hash(skb);
3084 }
3085
3086 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3087         .func           = bpf_get_hash_recalc,
3088         .gpl_only       = false,
3089         .ret_type       = RET_INTEGER,
3090         .arg1_type      = ARG_PTR_TO_CTX,
3091 };
3092
3093 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3094 {
3095         /* After all direct packet write, this can be used once for
3096          * triggering a lazy recalc on next skb_get_hash() invocation.
3097          */
3098         skb_clear_hash(skb);
3099         return 0;
3100 }
3101
3102 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3103         .func           = bpf_set_hash_invalid,
3104         .gpl_only       = false,
3105         .ret_type       = RET_INTEGER,
3106         .arg1_type      = ARG_PTR_TO_CTX,
3107 };
3108
3109 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3110 {
3111         /* Set user specified hash as L4(+), so that it gets returned
3112          * on skb_get_hash() call unless BPF prog later on triggers a
3113          * skb_clear_hash().
3114          */
3115         __skb_set_sw_hash(skb, hash, true);
3116         return 0;
3117 }
3118
3119 static const struct bpf_func_proto bpf_set_hash_proto = {
3120         .func           = bpf_set_hash,
3121         .gpl_only       = false,
3122         .ret_type       = RET_INTEGER,
3123         .arg1_type      = ARG_PTR_TO_CTX,
3124         .arg2_type      = ARG_ANYTHING,
3125 };
3126
3127 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3128            u16, vlan_tci)
3129 {
3130         int ret;
3131
3132         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3133                      vlan_proto != htons(ETH_P_8021AD)))
3134                 vlan_proto = htons(ETH_P_8021Q);
3135
3136         bpf_push_mac_rcsum(skb);
3137         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3138         bpf_pull_mac_rcsum(skb);
3139
3140         bpf_compute_data_pointers(skb);
3141         return ret;
3142 }
3143
3144 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3145         .func           = bpf_skb_vlan_push,
3146         .gpl_only       = false,
3147         .ret_type       = RET_INTEGER,
3148         .arg1_type      = ARG_PTR_TO_CTX,
3149         .arg2_type      = ARG_ANYTHING,
3150         .arg3_type      = ARG_ANYTHING,
3151 };
3152
3153 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3154 {
3155         int ret;
3156
3157         bpf_push_mac_rcsum(skb);
3158         ret = skb_vlan_pop(skb);
3159         bpf_pull_mac_rcsum(skb);
3160
3161         bpf_compute_data_pointers(skb);
3162         return ret;
3163 }
3164
3165 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3166         .func           = bpf_skb_vlan_pop,
3167         .gpl_only       = false,
3168         .ret_type       = RET_INTEGER,
3169         .arg1_type      = ARG_PTR_TO_CTX,
3170 };
3171
3172 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3173 {
3174         /* Caller already did skb_cow() with len as headroom,
3175          * so no need to do it here.
3176          */
3177         skb_push(skb, len);
3178         memmove(skb->data, skb->data + len, off);
3179         memset(skb->data + off, 0, len);
3180
3181         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3182          * needed here as it does not change the skb->csum
3183          * result for checksum complete when summing over
3184          * zeroed blocks.
3185          */
3186         return 0;
3187 }
3188
3189 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3190 {
3191         /* skb_ensure_writable() is not needed here, as we're
3192          * already working on an uncloned skb.
3193          */
3194         if (unlikely(!pskb_may_pull(skb, off + len)))
3195                 return -ENOMEM;
3196
3197         skb_postpull_rcsum(skb, skb->data + off, len);
3198         memmove(skb->data + len, skb->data, off);
3199         __skb_pull(skb, len);
3200
3201         return 0;
3202 }
3203
3204 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3205 {
3206         bool trans_same = skb->transport_header == skb->network_header;
3207         int ret;
3208
3209         /* There's no need for __skb_push()/__skb_pull() pair to
3210          * get to the start of the mac header as we're guaranteed
3211          * to always start from here under eBPF.
3212          */
3213         ret = bpf_skb_generic_push(skb, off, len);
3214         if (likely(!ret)) {
3215                 skb->mac_header -= len;
3216                 skb->network_header -= len;
3217                 if (trans_same)
3218                         skb->transport_header = skb->network_header;
3219         }
3220
3221         return ret;
3222 }
3223
3224 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3225 {
3226         bool trans_same = skb->transport_header == skb->network_header;
3227         int ret;
3228
3229         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3230         ret = bpf_skb_generic_pop(skb, off, len);
3231         if (likely(!ret)) {
3232                 skb->mac_header += len;
3233                 skb->network_header += len;
3234                 if (trans_same)
3235                         skb->transport_header = skb->network_header;
3236         }
3237
3238         return ret;
3239 }
3240
3241 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3242 {
3243         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3244         u32 off = skb_mac_header_len(skb);
3245         int ret;
3246
3247         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3248                 return -ENOTSUPP;
3249
3250         ret = skb_cow(skb, len_diff);
3251         if (unlikely(ret < 0))
3252                 return ret;
3253
3254         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3255         if (unlikely(ret < 0))
3256                 return ret;
3257
3258         if (skb_is_gso(skb)) {
3259                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3260
3261                 /* SKB_GSO_TCPV4 needs to be changed into
3262                  * SKB_GSO_TCPV6.
3263                  */
3264                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3265                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3266                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3267                 }
3268
3269                 /* Due to IPv6 header, MSS needs to be downgraded. */
3270                 skb_decrease_gso_size(shinfo, len_diff);
3271                 /* Header must be checked, and gso_segs recomputed. */
3272                 shinfo->gso_type |= SKB_GSO_DODGY;
3273                 shinfo->gso_segs = 0;
3274         }
3275
3276         skb->protocol = htons(ETH_P_IPV6);
3277         skb_clear_hash(skb);
3278
3279         return 0;
3280 }
3281
3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3283 {
3284         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3285         u32 off = skb_mac_header_len(skb);
3286         int ret;
3287
3288         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3289                 return -ENOTSUPP;
3290
3291         ret = skb_unclone(skb, GFP_ATOMIC);
3292         if (unlikely(ret < 0))
3293                 return ret;
3294
3295         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3296         if (unlikely(ret < 0))
3297                 return ret;
3298
3299         if (skb_is_gso(skb)) {
3300                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3301
3302                 /* SKB_GSO_TCPV6 needs to be changed into
3303                  * SKB_GSO_TCPV4.
3304                  */
3305                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3306                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3307                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3308                 }
3309
3310                 /* Due to IPv4 header, MSS can be upgraded. */
3311                 skb_increase_gso_size(shinfo, len_diff);
3312                 /* Header must be checked, and gso_segs recomputed. */
3313                 shinfo->gso_type |= SKB_GSO_DODGY;
3314                 shinfo->gso_segs = 0;
3315         }
3316
3317         skb->protocol = htons(ETH_P_IP);
3318         skb_clear_hash(skb);
3319
3320         return 0;
3321 }
3322
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3324 {
3325         __be16 from_proto = skb->protocol;
3326
3327         if (from_proto == htons(ETH_P_IP) &&
3328               to_proto == htons(ETH_P_IPV6))
3329                 return bpf_skb_proto_4_to_6(skb);
3330
3331         if (from_proto == htons(ETH_P_IPV6) &&
3332               to_proto == htons(ETH_P_IP))
3333                 return bpf_skb_proto_6_to_4(skb);
3334
3335         return -ENOTSUPP;
3336 }
3337
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3339            u64, flags)
3340 {
3341         int ret;
3342
3343         if (unlikely(flags))
3344                 return -EINVAL;
3345
3346         /* General idea is that this helper does the basic groundwork
3347          * needed for changing the protocol, and eBPF program fills the
3348          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349          * and other helpers, rather than passing a raw buffer here.
3350          *
3351          * The rationale is to keep this minimal and without a need to
3352          * deal with raw packet data. F.e. even if we would pass buffers
3353          * here, the program still needs to call the bpf_lX_csum_replace()
3354          * helpers anyway. Plus, this way we keep also separation of
3355          * concerns, since f.e. bpf_skb_store_bytes() should only take
3356          * care of stores.
3357          *
3358          * Currently, additional options and extension header space are
3359          * not supported, but flags register is reserved so we can adapt
3360          * that. For offloads, we mark packet as dodgy, so that headers
3361          * need to be verified first.
3362          */
3363         ret = bpf_skb_proto_xlat(skb, proto);
3364         bpf_compute_data_pointers(skb);
3365         return ret;
3366 }
3367
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369         .func           = bpf_skb_change_proto,
3370         .gpl_only       = false,
3371         .ret_type       = RET_INTEGER,
3372         .arg1_type      = ARG_PTR_TO_CTX,
3373         .arg2_type      = ARG_ANYTHING,
3374         .arg3_type      = ARG_ANYTHING,
3375 };
3376
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3378 {
3379         /* We only allow a restricted subset to be changed for now. */
3380         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381                      !skb_pkt_type_ok(pkt_type)))
3382                 return -EINVAL;
3383
3384         skb->pkt_type = pkt_type;
3385         return 0;
3386 }
3387
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389         .func           = bpf_skb_change_type,
3390         .gpl_only       = false,
3391         .ret_type       = RET_INTEGER,
3392         .arg1_type      = ARG_PTR_TO_CTX,
3393         .arg2_type      = ARG_ANYTHING,
3394 };
3395
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3397 {
3398         switch (skb->protocol) {
3399         case htons(ETH_P_IP):
3400                 return sizeof(struct iphdr);
3401         case htons(ETH_P_IPV6):
3402                 return sizeof(struct ipv6hdr);
3403         default:
3404                 return ~0U;
3405         }
3406 }
3407
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3410
3411 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3416                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3417
3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3419                             u64 flags)
3420 {
3421         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3422         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3423         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3424         unsigned int gso_type = SKB_GSO_DODGY;
3425         int ret;
3426
3427         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3428                 /* udp gso_size delineates datagrams, only allow if fixed */
3429                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3430                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3431                         return -ENOTSUPP;
3432         }
3433
3434         ret = skb_cow_head(skb, len_diff);
3435         if (unlikely(ret < 0))
3436                 return ret;
3437
3438         if (encap) {
3439                 if (skb->protocol != htons(ETH_P_IP) &&
3440                     skb->protocol != htons(ETH_P_IPV6))
3441                         return -ENOTSUPP;
3442
3443                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3444                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3445                         return -EINVAL;
3446
3447                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3448                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3449                         return -EINVAL;
3450
3451                 if (skb->encapsulation)
3452                         return -EALREADY;
3453
3454                 mac_len = skb->network_header - skb->mac_header;
3455                 inner_net = skb->network_header;
3456                 if (inner_mac_len > len_diff)
3457                         return -EINVAL;
3458                 inner_trans = skb->transport_header;
3459         }
3460
3461         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3462         if (unlikely(ret < 0))
3463                 return ret;
3464
3465         if (encap) {
3466                 skb->inner_mac_header = inner_net - inner_mac_len;
3467                 skb->inner_network_header = inner_net;
3468                 skb->inner_transport_header = inner_trans;
3469                 skb_set_inner_protocol(skb, skb->protocol);
3470
3471                 skb->encapsulation = 1;
3472                 skb_set_network_header(skb, mac_len);
3473
3474                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3475                         gso_type |= SKB_GSO_UDP_TUNNEL;
3476                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3477                         gso_type |= SKB_GSO_GRE;
3478                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479                         gso_type |= SKB_GSO_IPXIP6;
3480                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3481                         gso_type |= SKB_GSO_IPXIP4;
3482
3483                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3484                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3485                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3486                                         sizeof(struct ipv6hdr) :
3487                                         sizeof(struct iphdr);
3488
3489                         skb_set_transport_header(skb, mac_len + nh_len);
3490                 }
3491
3492                 /* Match skb->protocol to new outer l3 protocol */
3493                 if (skb->protocol == htons(ETH_P_IP) &&
3494                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3495                         skb->protocol = htons(ETH_P_IPV6);
3496                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3497                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3498                         skb->protocol = htons(ETH_P_IP);
3499         }
3500
3501         if (skb_is_gso(skb)) {
3502                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3503
3504                 /* Due to header grow, MSS needs to be downgraded. */
3505                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3506                         skb_decrease_gso_size(shinfo, len_diff);
3507
3508                 /* Header must be checked, and gso_segs recomputed. */
3509                 shinfo->gso_type |= gso_type;
3510                 shinfo->gso_segs = 0;
3511         }
3512
3513         return 0;
3514 }
3515
3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3517                               u64 flags)
3518 {
3519         int ret;
3520
3521         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3522                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3523                 return -EINVAL;
3524
3525         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3526                 /* udp gso_size delineates datagrams, only allow if fixed */
3527                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3528                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529                         return -ENOTSUPP;
3530         }
3531
3532         ret = skb_unclone(skb, GFP_ATOMIC);
3533         if (unlikely(ret < 0))
3534                 return ret;
3535
3536         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3537         if (unlikely(ret < 0))
3538                 return ret;
3539
3540         if (skb_is_gso(skb)) {
3541                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3542
3543                 /* Due to header shrink, MSS can be upgraded. */
3544                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3545                         skb_increase_gso_size(shinfo, len_diff);
3546
3547                 /* Header must be checked, and gso_segs recomputed. */
3548                 shinfo->gso_type |= SKB_GSO_DODGY;
3549                 shinfo->gso_segs = 0;
3550         }
3551
3552         return 0;
3553 }
3554
3555 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3556
3557 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3558            u32, mode, u64, flags)
3559 {
3560         u32 len_diff_abs = abs(len_diff);
3561         bool shrink = len_diff < 0;
3562         int ret = 0;
3563
3564         if (unlikely(flags || mode))
3565                 return -EINVAL;
3566         if (unlikely(len_diff_abs > 0xfffU))
3567                 return -EFAULT;
3568
3569         if (!shrink) {
3570                 ret = skb_cow(skb, len_diff);
3571                 if (unlikely(ret < 0))
3572                         return ret;
3573                 __skb_push(skb, len_diff_abs);
3574                 memset(skb->data, 0, len_diff_abs);
3575         } else {
3576                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3577                         return -ENOMEM;
3578                 __skb_pull(skb, len_diff_abs);
3579         }
3580         bpf_compute_data_end_sk_skb(skb);
3581         if (tls_sw_has_ctx_rx(skb->sk)) {
3582                 struct strp_msg *rxm = strp_msg(skb);
3583
3584                 rxm->full_len += len_diff;
3585         }
3586         return ret;
3587 }
3588
3589 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3590         .func           = sk_skb_adjust_room,
3591         .gpl_only       = false,
3592         .ret_type       = RET_INTEGER,
3593         .arg1_type      = ARG_PTR_TO_CTX,
3594         .arg2_type      = ARG_ANYTHING,
3595         .arg3_type      = ARG_ANYTHING,
3596         .arg4_type      = ARG_ANYTHING,
3597 };
3598
3599 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3600            u32, mode, u64, flags)
3601 {
3602         u32 len_cur, len_diff_abs = abs(len_diff);
3603         u32 len_min = bpf_skb_net_base_len(skb);
3604         u32 len_max = BPF_SKB_MAX_LEN;
3605         __be16 proto = skb->protocol;
3606         bool shrink = len_diff < 0;
3607         u32 off;
3608         int ret;
3609
3610         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3611                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3612                 return -EINVAL;
3613         if (unlikely(len_diff_abs > 0xfffU))
3614                 return -EFAULT;
3615         if (unlikely(proto != htons(ETH_P_IP) &&
3616                      proto != htons(ETH_P_IPV6)))
3617                 return -ENOTSUPP;
3618
3619         off = skb_mac_header_len(skb);
3620         switch (mode) {
3621         case BPF_ADJ_ROOM_NET:
3622                 off += bpf_skb_net_base_len(skb);
3623                 break;
3624         case BPF_ADJ_ROOM_MAC:
3625                 break;
3626         default:
3627                 return -ENOTSUPP;
3628         }
3629
3630         len_cur = skb->len - skb_network_offset(skb);
3631         if ((shrink && (len_diff_abs >= len_cur ||
3632                         len_cur - len_diff_abs < len_min)) ||
3633             (!shrink && (skb->len + len_diff_abs > len_max &&
3634                          !skb_is_gso(skb))))
3635                 return -ENOTSUPP;
3636
3637         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3638                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3639         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3640                 __skb_reset_checksum_unnecessary(skb);
3641
3642         bpf_compute_data_pointers(skb);
3643         return ret;
3644 }
3645
3646 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3647         .func           = bpf_skb_adjust_room,
3648         .gpl_only       = false,
3649         .ret_type       = RET_INTEGER,
3650         .arg1_type      = ARG_PTR_TO_CTX,
3651         .arg2_type      = ARG_ANYTHING,
3652         .arg3_type      = ARG_ANYTHING,
3653         .arg4_type      = ARG_ANYTHING,
3654 };
3655
3656 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3657 {
3658         u32 min_len = skb_network_offset(skb);
3659
3660         if (skb_transport_header_was_set(skb))
3661                 min_len = skb_transport_offset(skb);
3662         if (skb->ip_summed == CHECKSUM_PARTIAL)
3663                 min_len = skb_checksum_start_offset(skb) +
3664                           skb->csum_offset + sizeof(__sum16);
3665         return min_len;
3666 }
3667
3668 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3669 {
3670         unsigned int old_len = skb->len;
3671         int ret;
3672
3673         ret = __skb_grow_rcsum(skb, new_len);
3674         if (!ret)
3675                 memset(skb->data + old_len, 0, new_len - old_len);
3676         return ret;
3677 }
3678
3679 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3680 {
3681         return __skb_trim_rcsum(skb, new_len);
3682 }
3683
3684 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3685                                         u64 flags)
3686 {
3687         u32 max_len = BPF_SKB_MAX_LEN;
3688         u32 min_len = __bpf_skb_min_len(skb);
3689         int ret;
3690
3691         if (unlikely(flags || new_len > max_len || new_len < min_len))
3692                 return -EINVAL;
3693         if (skb->encapsulation)
3694                 return -ENOTSUPP;
3695
3696         /* The basic idea of this helper is that it's performing the
3697          * needed work to either grow or trim an skb, and eBPF program
3698          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3699          * bpf_lX_csum_replace() and others rather than passing a raw
3700          * buffer here. This one is a slow path helper and intended
3701          * for replies with control messages.
3702          *
3703          * Like in bpf_skb_change_proto(), we want to keep this rather
3704          * minimal and without protocol specifics so that we are able
3705          * to separate concerns as in bpf_skb_store_bytes() should only
3706          * be the one responsible for writing buffers.
3707          *
3708          * It's really expected to be a slow path operation here for
3709          * control message replies, so we're implicitly linearizing,
3710          * uncloning and drop offloads from the skb by this.
3711          */
3712         ret = __bpf_try_make_writable(skb, skb->len);
3713         if (!ret) {
3714                 if (new_len > skb->len)
3715                         ret = bpf_skb_grow_rcsum(skb, new_len);
3716                 else if (new_len < skb->len)
3717                         ret = bpf_skb_trim_rcsum(skb, new_len);
3718                 if (!ret && skb_is_gso(skb))
3719                         skb_gso_reset(skb);
3720         }
3721         return ret;
3722 }
3723
3724 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3725            u64, flags)
3726 {
3727         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3728
3729         bpf_compute_data_pointers(skb);
3730         return ret;
3731 }
3732
3733 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3734         .func           = bpf_skb_change_tail,
3735         .gpl_only       = false,
3736         .ret_type       = RET_INTEGER,
3737         .arg1_type      = ARG_PTR_TO_CTX,
3738         .arg2_type      = ARG_ANYTHING,
3739         .arg3_type      = ARG_ANYTHING,
3740 };
3741
3742 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3743            u64, flags)
3744 {
3745         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3746
3747         bpf_compute_data_end_sk_skb(skb);
3748         return ret;
3749 }
3750
3751 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3752         .func           = sk_skb_change_tail,
3753         .gpl_only       = false,
3754         .ret_type       = RET_INTEGER,
3755         .arg1_type      = ARG_PTR_TO_CTX,
3756         .arg2_type      = ARG_ANYTHING,
3757         .arg3_type      = ARG_ANYTHING,
3758 };
3759
3760 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3761                                         u64 flags)
3762 {
3763         u32 max_len = BPF_SKB_MAX_LEN;
3764         u32 new_len = skb->len + head_room;
3765         int ret;
3766
3767         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3768                      new_len < skb->len))
3769                 return -EINVAL;
3770
3771         ret = skb_cow(skb, head_room);
3772         if (likely(!ret)) {
3773                 /* Idea for this helper is that we currently only
3774                  * allow to expand on mac header. This means that
3775                  * skb->protocol network header, etc, stay as is.
3776                  * Compared to bpf_skb_change_tail(), we're more
3777                  * flexible due to not needing to linearize or
3778                  * reset GSO. Intention for this helper is to be
3779                  * used by an L3 skb that needs to push mac header
3780                  * for redirection into L2 device.
3781                  */
3782                 __skb_push(skb, head_room);
3783                 memset(skb->data, 0, head_room);
3784                 skb_reset_mac_header(skb);
3785         }
3786
3787         return ret;
3788 }
3789
3790 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3791            u64, flags)
3792 {
3793         int ret = __bpf_skb_change_head(skb, head_room, flags);
3794
3795         bpf_compute_data_pointers(skb);
3796         return ret;
3797 }
3798
3799 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3800         .func           = bpf_skb_change_head,
3801         .gpl_only       = false,
3802         .ret_type       = RET_INTEGER,
3803         .arg1_type      = ARG_PTR_TO_CTX,
3804         .arg2_type      = ARG_ANYTHING,
3805         .arg3_type      = ARG_ANYTHING,
3806 };
3807
3808 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3809            u64, flags)
3810 {
3811         int ret = __bpf_skb_change_head(skb, head_room, flags);
3812
3813         bpf_compute_data_end_sk_skb(skb);
3814         return ret;
3815 }
3816
3817 static const struct bpf_func_proto sk_skb_change_head_proto = {
3818         .func           = sk_skb_change_head,
3819         .gpl_only       = false,
3820         .ret_type       = RET_INTEGER,
3821         .arg1_type      = ARG_PTR_TO_CTX,
3822         .arg2_type      = ARG_ANYTHING,
3823         .arg3_type      = ARG_ANYTHING,
3824 };
3825 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3826 {
3827         return xdp_data_meta_unsupported(xdp) ? 0 :
3828                xdp->data - xdp->data_meta;
3829 }
3830
3831 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3832 {
3833         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3834         unsigned long metalen = xdp_get_metalen(xdp);
3835         void *data_start = xdp_frame_end + metalen;
3836         void *data = xdp->data + offset;
3837
3838         if (unlikely(data < data_start ||
3839                      data > xdp->data_end - ETH_HLEN))
3840                 return -EINVAL;
3841
3842         if (metalen)
3843                 memmove(xdp->data_meta + offset,
3844                         xdp->data_meta, metalen);
3845         xdp->data_meta += offset;
3846         xdp->data = data;
3847
3848         return 0;
3849 }
3850
3851 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3852         .func           = bpf_xdp_adjust_head,
3853         .gpl_only       = false,
3854         .ret_type       = RET_INTEGER,
3855         .arg1_type      = ARG_PTR_TO_CTX,
3856         .arg2_type      = ARG_ANYTHING,
3857 };
3858
3859 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3860 {
3861         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3862         void *data_end = xdp->data_end + offset;
3863
3864         /* Notice that xdp_data_hard_end have reserved some tailroom */
3865         if (unlikely(data_end > data_hard_end))
3866                 return -EINVAL;
3867
3868         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3869         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3870                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3871                 return -EINVAL;
3872         }
3873
3874         if (unlikely(data_end < xdp->data + ETH_HLEN))
3875                 return -EINVAL;
3876
3877         /* Clear memory area on grow, can contain uninit kernel memory */
3878         if (offset > 0)
3879                 memset(xdp->data_end, 0, offset);
3880
3881         xdp->data_end = data_end;
3882
3883         return 0;
3884 }
3885
3886 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3887         .func           = bpf_xdp_adjust_tail,
3888         .gpl_only       = false,
3889         .ret_type       = RET_INTEGER,
3890         .arg1_type      = ARG_PTR_TO_CTX,
3891         .arg2_type      = ARG_ANYTHING,
3892 };
3893
3894 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3895 {
3896         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3897         void *meta = xdp->data_meta + offset;
3898         unsigned long metalen = xdp->data - meta;
3899
3900         if (xdp_data_meta_unsupported(xdp))
3901                 return -ENOTSUPP;
3902         if (unlikely(meta < xdp_frame_end ||
3903                      meta > xdp->data))
3904                 return -EINVAL;
3905         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3906                      (metalen > 32)))
3907                 return -EACCES;
3908
3909         xdp->data_meta = meta;
3910
3911         return 0;
3912 }
3913
3914 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3915         .func           = bpf_xdp_adjust_meta,
3916         .gpl_only       = false,
3917         .ret_type       = RET_INTEGER,
3918         .arg1_type      = ARG_PTR_TO_CTX,
3919         .arg2_type      = ARG_ANYTHING,
3920 };
3921
3922 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3923                             struct bpf_map *map, struct xdp_buff *xdp)
3924 {
3925         switch (map->map_type) {
3926         case BPF_MAP_TYPE_DEVMAP:
3927         case BPF_MAP_TYPE_DEVMAP_HASH:
3928                 return dev_map_enqueue(fwd, xdp, dev_rx);
3929         case BPF_MAP_TYPE_CPUMAP:
3930                 return cpu_map_enqueue(fwd, xdp, dev_rx);
3931         case BPF_MAP_TYPE_XSKMAP:
3932                 return __xsk_map_redirect(fwd, xdp);
3933         default:
3934                 return -EBADRQC;
3935         }
3936         return 0;
3937 }
3938
3939 void xdp_do_flush(void)
3940 {
3941         __dev_flush();
3942         __cpu_map_flush();
3943         __xsk_map_flush();
3944 }
3945 EXPORT_SYMBOL_GPL(xdp_do_flush);
3946
3947 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3948 {
3949         switch (map->map_type) {
3950         case BPF_MAP_TYPE_DEVMAP:
3951                 return __dev_map_lookup_elem(map, index);
3952         case BPF_MAP_TYPE_DEVMAP_HASH:
3953                 return __dev_map_hash_lookup_elem(map, index);
3954         case BPF_MAP_TYPE_CPUMAP:
3955                 return __cpu_map_lookup_elem(map, index);
3956         case BPF_MAP_TYPE_XSKMAP:
3957                 return __xsk_map_lookup_elem(map, index);
3958         default:
3959                 return NULL;
3960         }
3961 }
3962
3963 void bpf_clear_redirect_map(struct bpf_map *map)
3964 {
3965         struct bpf_redirect_info *ri;
3966         int cpu;
3967
3968         for_each_possible_cpu(cpu) {
3969                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3970                 /* Avoid polluting remote cacheline due to writes if
3971                  * not needed. Once we pass this test, we need the
3972                  * cmpxchg() to make sure it hasn't been changed in
3973                  * the meantime by remote CPU.
3974                  */
3975                 if (unlikely(READ_ONCE(ri->map) == map))
3976                         cmpxchg(&ri->map, map, NULL);
3977         }
3978 }
3979
3980 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3981                     struct bpf_prog *xdp_prog)
3982 {
3983         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3984         struct bpf_map *map = READ_ONCE(ri->map);
3985         u32 index = ri->tgt_index;
3986         void *fwd = ri->tgt_value;
3987         int err;
3988
3989         ri->tgt_index = 0;
3990         ri->tgt_value = NULL;
3991         WRITE_ONCE(ri->map, NULL);
3992
3993         if (unlikely(!map)) {
3994                 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3995                 if (unlikely(!fwd)) {
3996                         err = -EINVAL;
3997                         goto err;
3998                 }
3999
4000                 err = dev_xdp_enqueue(fwd, xdp, dev);
4001         } else {
4002                 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
4003         }
4004
4005         if (unlikely(err))
4006                 goto err;
4007
4008         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4009         return 0;
4010 err:
4011         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4012         return err;
4013 }
4014 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4015
4016 static int xdp_do_generic_redirect_map(struct net_device *dev,
4017                                        struct sk_buff *skb,
4018                                        struct xdp_buff *xdp,
4019                                        struct bpf_prog *xdp_prog,
4020                                        struct bpf_map *map)
4021 {
4022         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4023         u32 index = ri->tgt_index;
4024         void *fwd = ri->tgt_value;
4025         int err = 0;
4026
4027         ri->tgt_index = 0;
4028         ri->tgt_value = NULL;
4029         WRITE_ONCE(ri->map, NULL);
4030
4031         if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
4032             map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
4033                 struct bpf_dtab_netdev *dst = fwd;
4034
4035                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4036                 if (unlikely(err))
4037                         goto err;
4038         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4039                 struct xdp_sock *xs = fwd;
4040
4041                 err = xsk_generic_rcv(xs, xdp);
4042                 if (err)
4043                         goto err;
4044                 consume_skb(skb);
4045         } else {
4046                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4047                 err = -EBADRQC;
4048                 goto err;
4049         }
4050
4051         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4052         return 0;
4053 err:
4054         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4055         return err;
4056 }
4057
4058 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4059                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4060 {
4061         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4062         struct bpf_map *map = READ_ONCE(ri->map);
4063         u32 index = ri->tgt_index;
4064         struct net_device *fwd;
4065         int err = 0;
4066
4067         if (map)
4068                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4069                                                    map);
4070         ri->tgt_index = 0;
4071         fwd = dev_get_by_index_rcu(dev_net(dev), index);
4072         if (unlikely(!fwd)) {
4073                 err = -EINVAL;
4074                 goto err;
4075         }
4076
4077         err = xdp_ok_fwd_dev(fwd, skb->len);
4078         if (unlikely(err))
4079                 goto err;
4080
4081         skb->dev = fwd;
4082         _trace_xdp_redirect(dev, xdp_prog, index);
4083         generic_xdp_tx(skb, xdp_prog);
4084         return 0;
4085 err:
4086         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4087         return err;
4088 }
4089
4090 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4091 {
4092         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4093
4094         if (unlikely(flags))
4095                 return XDP_ABORTED;
4096
4097         ri->flags = flags;
4098         ri->tgt_index = ifindex;
4099         ri->tgt_value = NULL;
4100         WRITE_ONCE(ri->map, NULL);
4101
4102         return XDP_REDIRECT;
4103 }
4104
4105 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4106         .func           = bpf_xdp_redirect,
4107         .gpl_only       = false,
4108         .ret_type       = RET_INTEGER,
4109         .arg1_type      = ARG_ANYTHING,
4110         .arg2_type      = ARG_ANYTHING,
4111 };
4112
4113 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4114            u64, flags)
4115 {
4116         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4117
4118         /* Lower bits of the flags are used as return code on lookup failure */
4119         if (unlikely(flags > XDP_TX))
4120                 return XDP_ABORTED;
4121
4122         ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4123         if (unlikely(!ri->tgt_value)) {
4124                 /* If the lookup fails we want to clear out the state in the
4125                  * redirect_info struct completely, so that if an eBPF program
4126                  * performs multiple lookups, the last one always takes
4127                  * precedence.
4128                  */
4129                 WRITE_ONCE(ri->map, NULL);
4130                 return flags;
4131         }
4132
4133         ri->flags = flags;
4134         ri->tgt_index = ifindex;
4135         WRITE_ONCE(ri->map, map);
4136
4137         return XDP_REDIRECT;
4138 }
4139
4140 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4141         .func           = bpf_xdp_redirect_map,
4142         .gpl_only       = false,
4143         .ret_type       = RET_INTEGER,
4144         .arg1_type      = ARG_CONST_MAP_PTR,
4145         .arg2_type      = ARG_ANYTHING,
4146         .arg3_type      = ARG_ANYTHING,
4147 };
4148
4149 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4150                                   unsigned long off, unsigned long len)
4151 {
4152         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4153
4154         if (unlikely(!ptr))
4155                 return len;
4156         if (ptr != dst_buff)
4157                 memcpy(dst_buff, ptr, len);
4158
4159         return 0;
4160 }
4161
4162 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4163            u64, flags, void *, meta, u64, meta_size)
4164 {
4165         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4166
4167         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4168                 return -EINVAL;
4169         if (unlikely(!skb || skb_size > skb->len))
4170                 return -EFAULT;
4171
4172         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4173                                 bpf_skb_copy);
4174 }
4175
4176 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4177         .func           = bpf_skb_event_output,
4178         .gpl_only       = true,
4179         .ret_type       = RET_INTEGER,
4180         .arg1_type      = ARG_PTR_TO_CTX,
4181         .arg2_type      = ARG_CONST_MAP_PTR,
4182         .arg3_type      = ARG_ANYTHING,
4183         .arg4_type      = ARG_PTR_TO_MEM,
4184         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4185 };
4186
4187 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4188
4189 const struct bpf_func_proto bpf_skb_output_proto = {
4190         .func           = bpf_skb_event_output,
4191         .gpl_only       = true,
4192         .ret_type       = RET_INTEGER,
4193         .arg1_type      = ARG_PTR_TO_BTF_ID,
4194         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4195         .arg2_type      = ARG_CONST_MAP_PTR,
4196         .arg3_type      = ARG_ANYTHING,
4197         .arg4_type      = ARG_PTR_TO_MEM,
4198         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4199 };
4200
4201 static unsigned short bpf_tunnel_key_af(u64 flags)
4202 {
4203         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4204 }
4205
4206 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4207            u32, size, u64, flags)
4208 {
4209         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4210         u8 compat[sizeof(struct bpf_tunnel_key)];
4211         void *to_orig = to;
4212         int err;
4213
4214         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4215                 err = -EINVAL;
4216                 goto err_clear;
4217         }
4218         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4219                 err = -EPROTO;
4220                 goto err_clear;
4221         }
4222         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4223                 err = -EINVAL;
4224                 switch (size) {
4225                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4226                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4227                         goto set_compat;
4228                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4229                         /* Fixup deprecated structure layouts here, so we have
4230                          * a common path later on.
4231                          */
4232                         if (ip_tunnel_info_af(info) != AF_INET)
4233                                 goto err_clear;
4234 set_compat:
4235                         to = (struct bpf_tunnel_key *)compat;
4236                         break;
4237                 default:
4238                         goto err_clear;
4239                 }
4240         }
4241
4242         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4243         to->tunnel_tos = info->key.tos;
4244         to->tunnel_ttl = info->key.ttl;
4245         to->tunnel_ext = 0;
4246
4247         if (flags & BPF_F_TUNINFO_IPV6) {
4248                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4249                        sizeof(to->remote_ipv6));
4250                 to->tunnel_label = be32_to_cpu(info->key.label);
4251         } else {
4252                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4253                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4254                 to->tunnel_label = 0;
4255         }
4256
4257         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4258                 memcpy(to_orig, to, size);
4259
4260         return 0;
4261 err_clear:
4262         memset(to_orig, 0, size);
4263         return err;
4264 }
4265
4266 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4267         .func           = bpf_skb_get_tunnel_key,
4268         .gpl_only       = false,
4269         .ret_type       = RET_INTEGER,
4270         .arg1_type      = ARG_PTR_TO_CTX,
4271         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4272         .arg3_type      = ARG_CONST_SIZE,
4273         .arg4_type      = ARG_ANYTHING,
4274 };
4275
4276 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4277 {
4278         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4279         int err;
4280
4281         if (unlikely(!info ||
4282                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4283                 err = -ENOENT;
4284                 goto err_clear;
4285         }
4286         if (unlikely(size < info->options_len)) {
4287                 err = -ENOMEM;
4288                 goto err_clear;
4289         }
4290
4291         ip_tunnel_info_opts_get(to, info);
4292         if (size > info->options_len)
4293                 memset(to + info->options_len, 0, size - info->options_len);
4294
4295         return info->options_len;
4296 err_clear:
4297         memset(to, 0, size);
4298         return err;
4299 }
4300
4301 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4302         .func           = bpf_skb_get_tunnel_opt,
4303         .gpl_only       = false,
4304         .ret_type       = RET_INTEGER,
4305         .arg1_type      = ARG_PTR_TO_CTX,
4306         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4307         .arg3_type      = ARG_CONST_SIZE,
4308 };
4309
4310 static struct metadata_dst __percpu *md_dst;
4311
4312 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4313            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4314 {
4315         struct metadata_dst *md = this_cpu_ptr(md_dst);
4316         u8 compat[sizeof(struct bpf_tunnel_key)];
4317         struct ip_tunnel_info *info;
4318
4319         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4320                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4321                 return -EINVAL;
4322         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4323                 switch (size) {
4324                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4325                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4326                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4327                         /* Fixup deprecated structure layouts here, so we have
4328                          * a common path later on.
4329                          */
4330                         memcpy(compat, from, size);
4331                         memset(compat + size, 0, sizeof(compat) - size);
4332                         from = (const struct bpf_tunnel_key *) compat;
4333                         break;
4334                 default:
4335                         return -EINVAL;
4336                 }
4337         }
4338         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4339                      from->tunnel_ext))
4340                 return -EINVAL;
4341
4342         skb_dst_drop(skb);
4343         dst_hold((struct dst_entry *) md);
4344         skb_dst_set(skb, (struct dst_entry *) md);
4345
4346         info = &md->u.tun_info;
4347         memset(info, 0, sizeof(*info));
4348         info->mode = IP_TUNNEL_INFO_TX;
4349
4350         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4351         if (flags & BPF_F_DONT_FRAGMENT)
4352                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4353         if (flags & BPF_F_ZERO_CSUM_TX)
4354                 info->key.tun_flags &= ~TUNNEL_CSUM;
4355         if (flags & BPF_F_SEQ_NUMBER)
4356                 info->key.tun_flags |= TUNNEL_SEQ;
4357
4358         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4359         info->key.tos = from->tunnel_tos;
4360         info->key.ttl = from->tunnel_ttl;
4361
4362         if (flags & BPF_F_TUNINFO_IPV6) {
4363                 info->mode |= IP_TUNNEL_INFO_IPV6;
4364                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4365                        sizeof(from->remote_ipv6));
4366                 info->key.label = cpu_to_be32(from->tunnel_label) &
4367                                   IPV6_FLOWLABEL_MASK;
4368         } else {
4369                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4370         }
4371
4372         return 0;
4373 }
4374
4375 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4376         .func           = bpf_skb_set_tunnel_key,
4377         .gpl_only       = false,
4378         .ret_type       = RET_INTEGER,
4379         .arg1_type      = ARG_PTR_TO_CTX,
4380         .arg2_type      = ARG_PTR_TO_MEM,
4381         .arg3_type      = ARG_CONST_SIZE,
4382         .arg4_type      = ARG_ANYTHING,
4383 };
4384
4385 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4386            const u8 *, from, u32, size)
4387 {
4388         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4389         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4390
4391         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4392                 return -EINVAL;
4393         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4394                 return -ENOMEM;
4395
4396         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4397
4398         return 0;
4399 }
4400
4401 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4402         .func           = bpf_skb_set_tunnel_opt,
4403         .gpl_only       = false,
4404         .ret_type       = RET_INTEGER,
4405         .arg1_type      = ARG_PTR_TO_CTX,
4406         .arg2_type      = ARG_PTR_TO_MEM,
4407         .arg3_type      = ARG_CONST_SIZE,
4408 };
4409
4410 static const struct bpf_func_proto *
4411 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4412 {
4413         if (!md_dst) {
4414                 struct metadata_dst __percpu *tmp;
4415
4416                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4417                                                 METADATA_IP_TUNNEL,
4418                                                 GFP_KERNEL);
4419                 if (!tmp)
4420                         return NULL;
4421                 if (cmpxchg(&md_dst, NULL, tmp))
4422                         metadata_dst_free_percpu(tmp);
4423         }
4424
4425         switch (which) {
4426         case BPF_FUNC_skb_set_tunnel_key:
4427                 return &bpf_skb_set_tunnel_key_proto;
4428         case BPF_FUNC_skb_set_tunnel_opt:
4429                 return &bpf_skb_set_tunnel_opt_proto;
4430         default:
4431                 return NULL;
4432         }
4433 }
4434
4435 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4436            u32, idx)
4437 {
4438         struct bpf_array *array = container_of(map, struct bpf_array, map);
4439         struct cgroup *cgrp;
4440         struct sock *sk;
4441
4442         sk = skb_to_full_sk(skb);
4443         if (!sk || !sk_fullsock(sk))
4444                 return -ENOENT;
4445         if (unlikely(idx >= array->map.max_entries))
4446                 return -E2BIG;
4447
4448         cgrp = READ_ONCE(array->ptrs[idx]);
4449         if (unlikely(!cgrp))
4450                 return -EAGAIN;
4451
4452         return sk_under_cgroup_hierarchy(sk, cgrp);
4453 }
4454
4455 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4456         .func           = bpf_skb_under_cgroup,
4457         .gpl_only       = false,
4458         .ret_type       = RET_INTEGER,
4459         .arg1_type      = ARG_PTR_TO_CTX,
4460         .arg2_type      = ARG_CONST_MAP_PTR,
4461         .arg3_type      = ARG_ANYTHING,
4462 };
4463
4464 #ifdef CONFIG_SOCK_CGROUP_DATA
4465 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4466 {
4467         struct cgroup *cgrp;
4468
4469         sk = sk_to_full_sk(sk);
4470         if (!sk || !sk_fullsock(sk))
4471                 return 0;
4472
4473         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4474         return cgroup_id(cgrp);
4475 }
4476
4477 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4478 {
4479         return __bpf_sk_cgroup_id(skb->sk);
4480 }
4481
4482 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4483         .func           = bpf_skb_cgroup_id,
4484         .gpl_only       = false,
4485         .ret_type       = RET_INTEGER,
4486         .arg1_type      = ARG_PTR_TO_CTX,
4487 };
4488
4489 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4490                                               int ancestor_level)
4491 {
4492         struct cgroup *ancestor;
4493         struct cgroup *cgrp;
4494
4495         sk = sk_to_full_sk(sk);
4496         if (!sk || !sk_fullsock(sk))
4497                 return 0;
4498
4499         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4500         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4501         if (!ancestor)
4502                 return 0;
4503
4504         return cgroup_id(ancestor);
4505 }
4506
4507 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4508            ancestor_level)
4509 {
4510         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4511 }
4512
4513 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4514         .func           = bpf_skb_ancestor_cgroup_id,
4515         .gpl_only       = false,
4516         .ret_type       = RET_INTEGER,
4517         .arg1_type      = ARG_PTR_TO_CTX,
4518         .arg2_type      = ARG_ANYTHING,
4519 };
4520
4521 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4522 {
4523         return __bpf_sk_cgroup_id(sk);
4524 }
4525
4526 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4527         .func           = bpf_sk_cgroup_id,
4528         .gpl_only       = false,
4529         .ret_type       = RET_INTEGER,
4530         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4531 };
4532
4533 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4534 {
4535         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4536 }
4537
4538 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4539         .func           = bpf_sk_ancestor_cgroup_id,
4540         .gpl_only       = false,
4541         .ret_type       = RET_INTEGER,
4542         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4543         .arg2_type      = ARG_ANYTHING,
4544 };
4545 #endif
4546
4547 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4548                                   unsigned long off, unsigned long len)
4549 {
4550         memcpy(dst_buff, src_buff + off, len);
4551         return 0;
4552 }
4553
4554 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4555            u64, flags, void *, meta, u64, meta_size)
4556 {
4557         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4558
4559         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4560                 return -EINVAL;
4561         if (unlikely(!xdp ||
4562                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4563                 return -EFAULT;
4564
4565         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4566                                 xdp_size, bpf_xdp_copy);
4567 }
4568
4569 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4570         .func           = bpf_xdp_event_output,
4571         .gpl_only       = true,
4572         .ret_type       = RET_INTEGER,
4573         .arg1_type      = ARG_PTR_TO_CTX,
4574         .arg2_type      = ARG_CONST_MAP_PTR,
4575         .arg3_type      = ARG_ANYTHING,
4576         .arg4_type      = ARG_PTR_TO_MEM,
4577         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4578 };
4579
4580 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4581
4582 const struct bpf_func_proto bpf_xdp_output_proto = {
4583         .func           = bpf_xdp_event_output,
4584         .gpl_only       = true,
4585         .ret_type       = RET_INTEGER,
4586         .arg1_type      = ARG_PTR_TO_BTF_ID,
4587         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4588         .arg2_type      = ARG_CONST_MAP_PTR,
4589         .arg3_type      = ARG_ANYTHING,
4590         .arg4_type      = ARG_PTR_TO_MEM,
4591         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4592 };
4593
4594 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4595 {
4596         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4597 }
4598
4599 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4600         .func           = bpf_get_socket_cookie,
4601         .gpl_only       = false,
4602         .ret_type       = RET_INTEGER,
4603         .arg1_type      = ARG_PTR_TO_CTX,
4604 };
4605
4606 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4607 {
4608         return __sock_gen_cookie(ctx->sk);
4609 }
4610
4611 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4612         .func           = bpf_get_socket_cookie_sock_addr,
4613         .gpl_only       = false,
4614         .ret_type       = RET_INTEGER,
4615         .arg1_type      = ARG_PTR_TO_CTX,
4616 };
4617
4618 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4619 {
4620         return __sock_gen_cookie(ctx);
4621 }
4622
4623 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4624         .func           = bpf_get_socket_cookie_sock,
4625         .gpl_only       = false,
4626         .ret_type       = RET_INTEGER,
4627         .arg1_type      = ARG_PTR_TO_CTX,
4628 };
4629
4630 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4631 {
4632         return sk ? sock_gen_cookie(sk) : 0;
4633 }
4634
4635 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4636         .func           = bpf_get_socket_ptr_cookie,
4637         .gpl_only       = false,
4638         .ret_type       = RET_INTEGER,
4639         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4640 };
4641
4642 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4643 {
4644         return __sock_gen_cookie(ctx->sk);
4645 }
4646
4647 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4648         .func           = bpf_get_socket_cookie_sock_ops,
4649         .gpl_only       = false,
4650         .ret_type       = RET_INTEGER,
4651         .arg1_type      = ARG_PTR_TO_CTX,
4652 };
4653
4654 static u64 __bpf_get_netns_cookie(struct sock *sk)
4655 {
4656         const struct net *net = sk ? sock_net(sk) : &init_net;
4657
4658         return net->net_cookie;
4659 }
4660
4661 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4662 {
4663         return __bpf_get_netns_cookie(ctx);
4664 }
4665
4666 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4667         .func           = bpf_get_netns_cookie_sock,
4668         .gpl_only       = false,
4669         .ret_type       = RET_INTEGER,
4670         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4671 };
4672
4673 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4674 {
4675         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4676 }
4677
4678 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4679         .func           = bpf_get_netns_cookie_sock_addr,
4680         .gpl_only       = false,
4681         .ret_type       = RET_INTEGER,
4682         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4683 };
4684
4685 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4686 {
4687         struct sock *sk = sk_to_full_sk(skb->sk);
4688         kuid_t kuid;
4689
4690         if (!sk || !sk_fullsock(sk))
4691                 return overflowuid;
4692         kuid = sock_net_uid(sock_net(sk), sk);
4693         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4694 }
4695
4696 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4697         .func           = bpf_get_socket_uid,
4698         .gpl_only       = false,
4699         .ret_type       = RET_INTEGER,
4700         .arg1_type      = ARG_PTR_TO_CTX,
4701 };
4702
4703 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4704                            char *optval, int optlen)
4705 {
4706         char devname[IFNAMSIZ];
4707         int val, valbool;
4708         struct net *net;
4709         int ifindex;
4710         int ret = 0;
4711
4712         if (!sk_fullsock(sk))
4713                 return -EINVAL;
4714
4715         sock_owned_by_me(sk);
4716
4717         if (level == SOL_SOCKET) {
4718                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4719                         return -EINVAL;
4720                 val = *((int *)optval);
4721                 valbool = val ? 1 : 0;
4722
4723                 /* Only some socketops are supported */
4724                 switch (optname) {
4725                 case SO_RCVBUF:
4726                         val = min_t(u32, val, sysctl_rmem_max);
4727                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4728                         WRITE_ONCE(sk->sk_rcvbuf,
4729                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4730                         break;
4731                 case SO_SNDBUF:
4732                         val = min_t(u32, val, sysctl_wmem_max);
4733                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4734                         WRITE_ONCE(sk->sk_sndbuf,
4735                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4736                         break;
4737                 case SO_MAX_PACING_RATE: /* 32bit version */
4738                         if (val != ~0U)
4739                                 cmpxchg(&sk->sk_pacing_status,
4740                                         SK_PACING_NONE,
4741                                         SK_PACING_NEEDED);
4742                         sk->sk_max_pacing_rate = (val == ~0U) ?
4743                                                  ~0UL : (unsigned int)val;
4744                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4745                                                  sk->sk_max_pacing_rate);
4746                         break;
4747                 case SO_PRIORITY:
4748                         sk->sk_priority = val;
4749                         break;
4750                 case SO_RCVLOWAT:
4751                         if (val < 0)
4752                                 val = INT_MAX;
4753                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4754                         break;
4755                 case SO_MARK:
4756                         if (sk->sk_mark != val) {
4757                                 sk->sk_mark = val;
4758                                 sk_dst_reset(sk);
4759                         }
4760                         break;
4761                 case SO_BINDTODEVICE:
4762                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4763                         strncpy(devname, optval, optlen);
4764                         devname[optlen] = 0;
4765
4766                         ifindex = 0;
4767                         if (devname[0] != '\0') {
4768                                 struct net_device *dev;
4769
4770                                 ret = -ENODEV;
4771
4772                                 net = sock_net(sk);
4773                                 dev = dev_get_by_name(net, devname);
4774                                 if (!dev)
4775                                         break;
4776                                 ifindex = dev->ifindex;
4777                                 dev_put(dev);
4778                         }
4779                         fallthrough;
4780                 case SO_BINDTOIFINDEX:
4781                         if (optname == SO_BINDTOIFINDEX)
4782                                 ifindex = val;
4783                         ret = sock_bindtoindex(sk, ifindex, false);
4784                         break;
4785                 case SO_KEEPALIVE:
4786                         if (sk->sk_prot->keepalive)
4787                                 sk->sk_prot->keepalive(sk, valbool);
4788                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4789                         break;
4790                 default:
4791                         ret = -EINVAL;
4792                 }
4793 #ifdef CONFIG_INET
4794         } else if (level == SOL_IP) {
4795                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4796                         return -EINVAL;
4797
4798                 val = *((int *)optval);
4799                 /* Only some options are supported */
4800                 switch (optname) {
4801                 case IP_TOS:
4802                         if (val < -1 || val > 0xff) {
4803                                 ret = -EINVAL;
4804                         } else {
4805                                 struct inet_sock *inet = inet_sk(sk);
4806
4807                                 if (val == -1)
4808                                         val = 0;
4809                                 inet->tos = val;
4810                         }
4811                         break;
4812                 default:
4813                         ret = -EINVAL;
4814                 }
4815 #if IS_ENABLED(CONFIG_IPV6)
4816         } else if (level == SOL_IPV6) {
4817                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4818                         return -EINVAL;
4819
4820                 val = *((int *)optval);
4821                 /* Only some options are supported */
4822                 switch (optname) {
4823                 case IPV6_TCLASS:
4824                         if (val < -1 || val > 0xff) {
4825                                 ret = -EINVAL;
4826                         } else {
4827                                 struct ipv6_pinfo *np = inet6_sk(sk);
4828
4829                                 if (val == -1)
4830                                         val = 0;
4831                                 np->tclass = val;
4832                         }
4833                         break;
4834                 default:
4835                         ret = -EINVAL;
4836                 }
4837 #endif
4838         } else if (level == SOL_TCP &&
4839                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4840                 if (optname == TCP_CONGESTION) {
4841                         char name[TCP_CA_NAME_MAX];
4842
4843                         strncpy(name, optval, min_t(long, optlen,
4844                                                     TCP_CA_NAME_MAX-1));
4845                         name[TCP_CA_NAME_MAX-1] = 0;
4846                         ret = tcp_set_congestion_control(sk, name, false, true);
4847                 } else {
4848                         struct inet_connection_sock *icsk = inet_csk(sk);
4849                         struct tcp_sock *tp = tcp_sk(sk);
4850                         unsigned long timeout;
4851
4852                         if (optlen != sizeof(int))
4853                                 return -EINVAL;
4854
4855                         val = *((int *)optval);
4856                         /* Only some options are supported */
4857                         switch (optname) {
4858                         case TCP_BPF_IW:
4859                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4860                                         ret = -EINVAL;
4861                                 else
4862                                         tp->snd_cwnd = val;
4863                                 break;
4864                         case TCP_BPF_SNDCWND_CLAMP:
4865                                 if (val <= 0) {
4866                                         ret = -EINVAL;
4867                                 } else {
4868                                         tp->snd_cwnd_clamp = val;
4869                                         tp->snd_ssthresh = val;
4870                                 }
4871                                 break;
4872                         case TCP_BPF_DELACK_MAX:
4873                                 timeout = usecs_to_jiffies(val);
4874                                 if (timeout > TCP_DELACK_MAX ||
4875                                     timeout < TCP_TIMEOUT_MIN)
4876                                         return -EINVAL;
4877                                 inet_csk(sk)->icsk_delack_max = timeout;
4878                                 break;
4879                         case TCP_BPF_RTO_MIN:
4880                                 timeout = usecs_to_jiffies(val);
4881                                 if (timeout > TCP_RTO_MIN ||
4882                                     timeout < TCP_TIMEOUT_MIN)
4883                                         return -EINVAL;
4884                                 inet_csk(sk)->icsk_rto_min = timeout;
4885                                 break;
4886                         case TCP_SAVE_SYN:
4887                                 if (val < 0 || val > 1)
4888                                         ret = -EINVAL;
4889                                 else
4890                                         tp->save_syn = val;
4891                                 break;
4892                         case TCP_KEEPIDLE:
4893                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4894                                 break;
4895                         case TCP_KEEPINTVL:
4896                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4897                                         ret = -EINVAL;
4898                                 else
4899                                         tp->keepalive_intvl = val * HZ;
4900                                 break;
4901                         case TCP_KEEPCNT:
4902                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4903                                         ret = -EINVAL;
4904                                 else
4905                                         tp->keepalive_probes = val;
4906                                 break;
4907                         case TCP_SYNCNT:
4908                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4909                                         ret = -EINVAL;
4910                                 else
4911                                         icsk->icsk_syn_retries = val;
4912                                 break;
4913                         case TCP_USER_TIMEOUT:
4914                                 if (val < 0)
4915                                         ret = -EINVAL;
4916                                 else
4917                                         icsk->icsk_user_timeout = val;
4918                                 break;
4919                         case TCP_NOTSENT_LOWAT:
4920                                 tp->notsent_lowat = val;
4921                                 sk->sk_write_space(sk);
4922                                 break;
4923                         case TCP_WINDOW_CLAMP:
4924                                 ret = tcp_set_window_clamp(sk, val);
4925                                 break;
4926                         default:
4927                                 ret = -EINVAL;
4928                         }
4929                 }
4930 #endif
4931         } else {
4932                 ret = -EINVAL;
4933         }
4934         return ret;
4935 }
4936
4937 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4938                            char *optval, int optlen)
4939 {
4940         if (!sk_fullsock(sk))
4941                 goto err_clear;
4942
4943         sock_owned_by_me(sk);
4944
4945         if (level == SOL_SOCKET) {
4946                 if (optlen != sizeof(int))
4947                         goto err_clear;
4948
4949                 switch (optname) {
4950                 case SO_MARK:
4951                         *((int *)optval) = sk->sk_mark;
4952                         break;
4953                 case SO_PRIORITY:
4954                         *((int *)optval) = sk->sk_priority;
4955                         break;
4956                 case SO_BINDTOIFINDEX:
4957                         *((int *)optval) = sk->sk_bound_dev_if;
4958                         break;
4959                 default:
4960                         goto err_clear;
4961                 }
4962 #ifdef CONFIG_INET
4963         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4964                 struct inet_connection_sock *icsk;
4965                 struct tcp_sock *tp;
4966
4967                 switch (optname) {
4968                 case TCP_CONGESTION:
4969                         icsk = inet_csk(sk);
4970
4971                         if (!icsk->icsk_ca_ops || optlen <= 1)
4972                                 goto err_clear;
4973                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4974                         optval[optlen - 1] = 0;
4975                         break;
4976                 case TCP_SAVED_SYN:
4977                         tp = tcp_sk(sk);
4978
4979                         if (optlen <= 0 || !tp->saved_syn ||
4980                             optlen > tcp_saved_syn_len(tp->saved_syn))
4981                                 goto err_clear;
4982                         memcpy(optval, tp->saved_syn->data, optlen);
4983                         break;
4984                 default:
4985                         goto err_clear;
4986                 }
4987         } else if (level == SOL_IP) {
4988                 struct inet_sock *inet = inet_sk(sk);
4989
4990                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4991                         goto err_clear;
4992
4993                 /* Only some options are supported */
4994                 switch (optname) {
4995                 case IP_TOS:
4996                         *((int *)optval) = (int)inet->tos;
4997                         break;
4998                 default:
4999                         goto err_clear;
5000                 }
5001 #if IS_ENABLED(CONFIG_IPV6)
5002         } else if (level == SOL_IPV6) {
5003                 struct ipv6_pinfo *np = inet6_sk(sk);
5004
5005                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
5006                         goto err_clear;
5007
5008                 /* Only some options are supported */
5009                 switch (optname) {
5010                 case IPV6_TCLASS:
5011                         *((int *)optval) = (int)np->tclass;
5012                         break;
5013                 default:
5014                         goto err_clear;
5015                 }
5016 #endif
5017 #endif
5018         } else {
5019                 goto err_clear;
5020         }
5021         return 0;
5022 err_clear:
5023         memset(optval, 0, optlen);
5024         return -EINVAL;
5025 }
5026
5027 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5028            int, level, int, optname, char *, optval, int, optlen)
5029 {
5030         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5031 }
5032
5033 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5034         .func           = bpf_sock_addr_setsockopt,
5035         .gpl_only       = false,
5036         .ret_type       = RET_INTEGER,
5037         .arg1_type      = ARG_PTR_TO_CTX,
5038         .arg2_type      = ARG_ANYTHING,
5039         .arg3_type      = ARG_ANYTHING,
5040         .arg4_type      = ARG_PTR_TO_MEM,
5041         .arg5_type      = ARG_CONST_SIZE,
5042 };
5043
5044 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5045            int, level, int, optname, char *, optval, int, optlen)
5046 {
5047         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5048 }
5049
5050 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5051         .func           = bpf_sock_addr_getsockopt,
5052         .gpl_only       = false,
5053         .ret_type       = RET_INTEGER,
5054         .arg1_type      = ARG_PTR_TO_CTX,
5055         .arg2_type      = ARG_ANYTHING,
5056         .arg3_type      = ARG_ANYTHING,
5057         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5058         .arg5_type      = ARG_CONST_SIZE,
5059 };
5060
5061 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5062            int, level, int, optname, char *, optval, int, optlen)
5063 {
5064         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5065 }
5066
5067 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5068         .func           = bpf_sock_ops_setsockopt,
5069         .gpl_only       = false,
5070         .ret_type       = RET_INTEGER,
5071         .arg1_type      = ARG_PTR_TO_CTX,
5072         .arg2_type      = ARG_ANYTHING,
5073         .arg3_type      = ARG_ANYTHING,
5074         .arg4_type      = ARG_PTR_TO_MEM,
5075         .arg5_type      = ARG_CONST_SIZE,
5076 };
5077
5078 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5079                                 int optname, const u8 **start)
5080 {
5081         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5082         const u8 *hdr_start;
5083         int ret;
5084
5085         if (syn_skb) {
5086                 /* sk is a request_sock here */
5087
5088                 if (optname == TCP_BPF_SYN) {
5089                         hdr_start = syn_skb->data;
5090                         ret = tcp_hdrlen(syn_skb);
5091                 } else if (optname == TCP_BPF_SYN_IP) {
5092                         hdr_start = skb_network_header(syn_skb);
5093                         ret = skb_network_header_len(syn_skb) +
5094                                 tcp_hdrlen(syn_skb);
5095                 } else {
5096                         /* optname == TCP_BPF_SYN_MAC */
5097                         hdr_start = skb_mac_header(syn_skb);
5098                         ret = skb_mac_header_len(syn_skb) +
5099                                 skb_network_header_len(syn_skb) +
5100                                 tcp_hdrlen(syn_skb);
5101                 }
5102         } else {
5103                 struct sock *sk = bpf_sock->sk;
5104                 struct saved_syn *saved_syn;
5105
5106                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5107                         /* synack retransmit. bpf_sock->syn_skb will
5108                          * not be available.  It has to resort to
5109                          * saved_syn (if it is saved).
5110                          */
5111                         saved_syn = inet_reqsk(sk)->saved_syn;
5112                 else
5113                         saved_syn = tcp_sk(sk)->saved_syn;
5114
5115                 if (!saved_syn)
5116                         return -ENOENT;
5117
5118                 if (optname == TCP_BPF_SYN) {
5119                         hdr_start = saved_syn->data +
5120                                 saved_syn->mac_hdrlen +
5121                                 saved_syn->network_hdrlen;
5122                         ret = saved_syn->tcp_hdrlen;
5123                 } else if (optname == TCP_BPF_SYN_IP) {
5124                         hdr_start = saved_syn->data +
5125                                 saved_syn->mac_hdrlen;
5126                         ret = saved_syn->network_hdrlen +
5127                                 saved_syn->tcp_hdrlen;
5128                 } else {
5129                         /* optname == TCP_BPF_SYN_MAC */
5130
5131                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5132                         if (!saved_syn->mac_hdrlen)
5133                                 return -ENOENT;
5134
5135                         hdr_start = saved_syn->data;
5136                         ret = saved_syn->mac_hdrlen +
5137                                 saved_syn->network_hdrlen +
5138                                 saved_syn->tcp_hdrlen;
5139                 }
5140         }
5141
5142         *start = hdr_start;
5143         return ret;
5144 }
5145
5146 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5147            int, level, int, optname, char *, optval, int, optlen)
5148 {
5149         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5150             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5151                 int ret, copy_len = 0;
5152                 const u8 *start;
5153
5154                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5155                 if (ret > 0) {
5156                         copy_len = ret;
5157                         if (optlen < copy_len) {
5158                                 copy_len = optlen;
5159                                 ret = -ENOSPC;
5160                         }
5161
5162                         memcpy(optval, start, copy_len);
5163                 }
5164
5165                 /* Zero out unused buffer at the end */
5166                 memset(optval + copy_len, 0, optlen - copy_len);
5167
5168                 return ret;
5169         }
5170
5171         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5172 }
5173
5174 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5175         .func           = bpf_sock_ops_getsockopt,
5176         .gpl_only       = false,
5177         .ret_type       = RET_INTEGER,
5178         .arg1_type      = ARG_PTR_TO_CTX,
5179         .arg2_type      = ARG_ANYTHING,
5180         .arg3_type      = ARG_ANYTHING,
5181         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5182         .arg5_type      = ARG_CONST_SIZE,
5183 };
5184
5185 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5186            int, argval)
5187 {
5188         struct sock *sk = bpf_sock->sk;
5189         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5190
5191         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5192                 return -EINVAL;
5193
5194         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5195
5196         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5197 }
5198
5199 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5200         .func           = bpf_sock_ops_cb_flags_set,
5201         .gpl_only       = false,
5202         .ret_type       = RET_INTEGER,
5203         .arg1_type      = ARG_PTR_TO_CTX,
5204         .arg2_type      = ARG_ANYTHING,
5205 };
5206
5207 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5208 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5209
5210 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5211            int, addr_len)
5212 {
5213 #ifdef CONFIG_INET
5214         struct sock *sk = ctx->sk;
5215         u32 flags = BIND_FROM_BPF;
5216         int err;
5217
5218         err = -EINVAL;
5219         if (addr_len < offsetofend(struct sockaddr, sa_family))
5220                 return err;
5221         if (addr->sa_family == AF_INET) {
5222                 if (addr_len < sizeof(struct sockaddr_in))
5223                         return err;
5224                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5225                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5226                 return __inet_bind(sk, addr, addr_len, flags);
5227 #if IS_ENABLED(CONFIG_IPV6)
5228         } else if (addr->sa_family == AF_INET6) {
5229                 if (addr_len < SIN6_LEN_RFC2133)
5230                         return err;
5231                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5232                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5233                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5234                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5235                  */
5236                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5237 #endif /* CONFIG_IPV6 */
5238         }
5239 #endif /* CONFIG_INET */
5240
5241         return -EAFNOSUPPORT;
5242 }
5243
5244 static const struct bpf_func_proto bpf_bind_proto = {
5245         .func           = bpf_bind,
5246         .gpl_only       = false,
5247         .ret_type       = RET_INTEGER,
5248         .arg1_type      = ARG_PTR_TO_CTX,
5249         .arg2_type      = ARG_PTR_TO_MEM,
5250         .arg3_type      = ARG_CONST_SIZE,
5251 };
5252
5253 #ifdef CONFIG_XFRM
5254 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5255            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5256 {
5257         const struct sec_path *sp = skb_sec_path(skb);
5258         const struct xfrm_state *x;
5259
5260         if (!sp || unlikely(index >= sp->len || flags))
5261                 goto err_clear;
5262
5263         x = sp->xvec[index];
5264
5265         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5266                 goto err_clear;
5267
5268         to->reqid = x->props.reqid;
5269         to->spi = x->id.spi;
5270         to->family = x->props.family;
5271         to->ext = 0;
5272
5273         if (to->family == AF_INET6) {
5274                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5275                        sizeof(to->remote_ipv6));
5276         } else {
5277                 to->remote_ipv4 = x->props.saddr.a4;
5278                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5279         }
5280
5281         return 0;
5282 err_clear:
5283         memset(to, 0, size);
5284         return -EINVAL;
5285 }
5286
5287 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5288         .func           = bpf_skb_get_xfrm_state,
5289         .gpl_only       = false,
5290         .ret_type       = RET_INTEGER,
5291         .arg1_type      = ARG_PTR_TO_CTX,
5292         .arg2_type      = ARG_ANYTHING,
5293         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5294         .arg4_type      = ARG_CONST_SIZE,
5295         .arg5_type      = ARG_ANYTHING,
5296 };
5297 #endif
5298
5299 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5300 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5301                                   const struct neighbour *neigh,
5302                                   const struct net_device *dev, u32 mtu)
5303 {
5304         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5305         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5306         params->h_vlan_TCI = 0;
5307         params->h_vlan_proto = 0;
5308         if (mtu)
5309                 params->mtu_result = mtu; /* union with tot_len */
5310
5311         return 0;
5312 }
5313 #endif
5314
5315 #if IS_ENABLED(CONFIG_INET)
5316 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5317                                u32 flags, bool check_mtu)
5318 {
5319         struct fib_nh_common *nhc;
5320         struct in_device *in_dev;
5321         struct neighbour *neigh;
5322         struct net_device *dev;
5323         struct fib_result res;
5324         struct flowi4 fl4;
5325         u32 mtu = 0;
5326         int err;
5327
5328         dev = dev_get_by_index_rcu(net, params->ifindex);
5329         if (unlikely(!dev))
5330                 return -ENODEV;
5331
5332         /* verify forwarding is enabled on this interface */
5333         in_dev = __in_dev_get_rcu(dev);
5334         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5335                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5336
5337         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5338                 fl4.flowi4_iif = 1;
5339                 fl4.flowi4_oif = params->ifindex;
5340         } else {
5341                 fl4.flowi4_iif = params->ifindex;
5342                 fl4.flowi4_oif = 0;
5343         }
5344         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5345         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5346         fl4.flowi4_flags = 0;
5347
5348         fl4.flowi4_proto = params->l4_protocol;
5349         fl4.daddr = params->ipv4_dst;
5350         fl4.saddr = params->ipv4_src;
5351         fl4.fl4_sport = params->sport;
5352         fl4.fl4_dport = params->dport;
5353         fl4.flowi4_multipath_hash = 0;
5354
5355         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5356                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5357                 struct fib_table *tb;
5358
5359                 tb = fib_get_table(net, tbid);
5360                 if (unlikely(!tb))
5361                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5362
5363                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5364         } else {
5365                 fl4.flowi4_mark = 0;
5366                 fl4.flowi4_secid = 0;
5367                 fl4.flowi4_tun_key.tun_id = 0;
5368                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5369
5370                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5371         }
5372
5373         if (err) {
5374                 /* map fib lookup errors to RTN_ type */
5375                 if (err == -EINVAL)
5376                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5377                 if (err == -EHOSTUNREACH)
5378                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5379                 if (err == -EACCES)
5380                         return BPF_FIB_LKUP_RET_PROHIBIT;
5381
5382                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5383         }
5384
5385         if (res.type != RTN_UNICAST)
5386                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5387
5388         if (fib_info_num_path(res.fi) > 1)
5389                 fib_select_path(net, &res, &fl4, NULL);
5390
5391         if (check_mtu) {
5392                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5393                 if (params->tot_len > mtu) {
5394                         params->mtu_result = mtu; /* union with tot_len */
5395                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5396                 }
5397         }
5398
5399         nhc = res.nhc;
5400
5401         /* do not handle lwt encaps right now */
5402         if (nhc->nhc_lwtstate)
5403                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5404
5405         dev = nhc->nhc_dev;
5406
5407         params->rt_metric = res.fi->fib_priority;
5408         params->ifindex = dev->ifindex;
5409
5410         /* xdp and cls_bpf programs are run in RCU-bh so
5411          * rcu_read_lock_bh is not needed here
5412          */
5413         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5414                 if (nhc->nhc_gw_family)
5415                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5416
5417                 neigh = __ipv4_neigh_lookup_noref(dev,
5418                                                  (__force u32)params->ipv4_dst);
5419         } else {
5420                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5421
5422                 params->family = AF_INET6;
5423                 *dst = nhc->nhc_gw.ipv6;
5424                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5425         }
5426
5427         if (!neigh)
5428                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5429
5430         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5431 }
5432 #endif
5433
5434 #if IS_ENABLED(CONFIG_IPV6)
5435 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5436                                u32 flags, bool check_mtu)
5437 {
5438         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5439         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5440         struct fib6_result res = {};
5441         struct neighbour *neigh;
5442         struct net_device *dev;
5443         struct inet6_dev *idev;
5444         struct flowi6 fl6;
5445         int strict = 0;
5446         int oif, err;
5447         u32 mtu = 0;
5448
5449         /* link local addresses are never forwarded */
5450         if (rt6_need_strict(dst) || rt6_need_strict(src))
5451                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5452
5453         dev = dev_get_by_index_rcu(net, params->ifindex);
5454         if (unlikely(!dev))
5455                 return -ENODEV;
5456
5457         idev = __in6_dev_get_safely(dev);
5458         if (unlikely(!idev || !idev->cnf.forwarding))
5459                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5460
5461         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5462                 fl6.flowi6_iif = 1;
5463                 oif = fl6.flowi6_oif = params->ifindex;
5464         } else {
5465                 oif = fl6.flowi6_iif = params->ifindex;
5466                 fl6.flowi6_oif = 0;
5467                 strict = RT6_LOOKUP_F_HAS_SADDR;
5468         }
5469         fl6.flowlabel = params->flowinfo;
5470         fl6.flowi6_scope = 0;
5471         fl6.flowi6_flags = 0;
5472         fl6.mp_hash = 0;
5473
5474         fl6.flowi6_proto = params->l4_protocol;
5475         fl6.daddr = *dst;
5476         fl6.saddr = *src;
5477         fl6.fl6_sport = params->sport;
5478         fl6.fl6_dport = params->dport;
5479
5480         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5481                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5482                 struct fib6_table *tb;
5483
5484                 tb = ipv6_stub->fib6_get_table(net, tbid);
5485                 if (unlikely(!tb))
5486                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5487
5488                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5489                                                    strict);
5490         } else {
5491                 fl6.flowi6_mark = 0;
5492                 fl6.flowi6_secid = 0;
5493                 fl6.flowi6_tun_key.tun_id = 0;
5494                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5495
5496                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5497         }
5498
5499         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5500                      res.f6i == net->ipv6.fib6_null_entry))
5501                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5502
5503         switch (res.fib6_type) {
5504         /* only unicast is forwarded */
5505         case RTN_UNICAST:
5506                 break;
5507         case RTN_BLACKHOLE:
5508                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5509         case RTN_UNREACHABLE:
5510                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5511         case RTN_PROHIBIT:
5512                 return BPF_FIB_LKUP_RET_PROHIBIT;
5513         default:
5514                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5515         }
5516
5517         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5518                                     fl6.flowi6_oif != 0, NULL, strict);
5519
5520         if (check_mtu) {
5521                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5522                 if (params->tot_len > mtu) {
5523                         params->mtu_result = mtu; /* union with tot_len */
5524                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5525                 }
5526         }
5527
5528         if (res.nh->fib_nh_lws)
5529                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5530
5531         if (res.nh->fib_nh_gw_family)
5532                 *dst = res.nh->fib_nh_gw6;
5533
5534         dev = res.nh->fib_nh_dev;
5535         params->rt_metric = res.f6i->fib6_metric;
5536         params->ifindex = dev->ifindex;
5537
5538         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5539          * not needed here.
5540          */
5541         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5542         if (!neigh)
5543                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5544
5545         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5546 }
5547 #endif
5548
5549 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5550            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5551 {
5552         if (plen < sizeof(*params))
5553                 return -EINVAL;
5554
5555         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5556                 return -EINVAL;
5557
5558         switch (params->family) {
5559 #if IS_ENABLED(CONFIG_INET)
5560         case AF_INET:
5561                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5562                                            flags, true);
5563 #endif
5564 #if IS_ENABLED(CONFIG_IPV6)
5565         case AF_INET6:
5566                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5567                                            flags, true);
5568 #endif
5569         }
5570         return -EAFNOSUPPORT;
5571 }
5572
5573 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5574         .func           = bpf_xdp_fib_lookup,
5575         .gpl_only       = true,
5576         .ret_type       = RET_INTEGER,
5577         .arg1_type      = ARG_PTR_TO_CTX,
5578         .arg2_type      = ARG_PTR_TO_MEM,
5579         .arg3_type      = ARG_CONST_SIZE,
5580         .arg4_type      = ARG_ANYTHING,
5581 };
5582
5583 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5584            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5585 {
5586         struct net *net = dev_net(skb->dev);
5587         int rc = -EAFNOSUPPORT;
5588         bool check_mtu = false;
5589
5590         if (plen < sizeof(*params))
5591                 return -EINVAL;
5592
5593         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5594                 return -EINVAL;
5595
5596         if (params->tot_len)
5597                 check_mtu = true;
5598
5599         switch (params->family) {
5600 #if IS_ENABLED(CONFIG_INET)
5601         case AF_INET:
5602                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5603                 break;
5604 #endif
5605 #if IS_ENABLED(CONFIG_IPV6)
5606         case AF_INET6:
5607                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5608                 break;
5609 #endif
5610         }
5611
5612         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5613                 struct net_device *dev;
5614
5615                 /* When tot_len isn't provided by user, check skb
5616                  * against MTU of FIB lookup resulting net_device
5617                  */
5618                 dev = dev_get_by_index_rcu(net, params->ifindex);
5619                 if (!is_skb_forwardable(dev, skb))
5620                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5621
5622                 params->mtu_result = dev->mtu; /* union with tot_len */
5623         }
5624
5625         return rc;
5626 }
5627
5628 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5629         .func           = bpf_skb_fib_lookup,
5630         .gpl_only       = true,
5631         .ret_type       = RET_INTEGER,
5632         .arg1_type      = ARG_PTR_TO_CTX,
5633         .arg2_type      = ARG_PTR_TO_MEM,
5634         .arg3_type      = ARG_CONST_SIZE,
5635         .arg4_type      = ARG_ANYTHING,
5636 };
5637
5638 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5639                                             u32 ifindex)
5640 {
5641         struct net *netns = dev_net(dev_curr);
5642
5643         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5644         if (ifindex == 0)
5645                 return dev_curr;
5646
5647         return dev_get_by_index_rcu(netns, ifindex);
5648 }
5649
5650 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5651            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5652 {
5653         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5654         struct net_device *dev = skb->dev;
5655         int skb_len, dev_len;
5656         int mtu;
5657
5658         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5659                 return -EINVAL;
5660
5661         if (unlikely(flags & BPF_MTU_CHK_SEGS && len_diff))
5662                 return -EINVAL;
5663
5664         dev = __dev_via_ifindex(dev, ifindex);
5665         if (unlikely(!dev))
5666                 return -ENODEV;
5667
5668         mtu = READ_ONCE(dev->mtu);
5669
5670         dev_len = mtu + dev->hard_header_len;
5671         skb_len = skb->len + len_diff; /* minus result pass check */
5672         if (skb_len <= dev_len) {
5673                 ret = BPF_MTU_CHK_RET_SUCCESS;
5674                 goto out;
5675         }
5676         /* At this point, skb->len exceed MTU, but as it include length of all
5677          * segments, it can still be below MTU.  The SKB can possibly get
5678          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5679          * must choose if segs are to be MTU checked.
5680          */
5681         if (skb_is_gso(skb)) {
5682                 ret = BPF_MTU_CHK_RET_SUCCESS;
5683
5684                 if (flags & BPF_MTU_CHK_SEGS &&
5685                     !skb_gso_validate_network_len(skb, mtu))
5686                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5687         }
5688 out:
5689         /* BPF verifier guarantees valid pointer */
5690         *mtu_len = mtu;
5691
5692         return ret;
5693 }
5694
5695 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5696            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5697 {
5698         struct net_device *dev = xdp->rxq->dev;
5699         int xdp_len = xdp->data_end - xdp->data;
5700         int ret = BPF_MTU_CHK_RET_SUCCESS;
5701         int mtu, dev_len;
5702
5703         /* XDP variant doesn't support multi-buffer segment check (yet) */
5704         if (unlikely(flags))
5705                 return -EINVAL;
5706
5707         dev = __dev_via_ifindex(dev, ifindex);
5708         if (unlikely(!dev))
5709                 return -ENODEV;
5710
5711         mtu = READ_ONCE(dev->mtu);
5712
5713         /* Add L2-header as dev MTU is L3 size */
5714         dev_len = mtu + dev->hard_header_len;
5715
5716         xdp_len += len_diff; /* minus result pass check */
5717         if (xdp_len > dev_len)
5718                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5719
5720         /* BPF verifier guarantees valid pointer */
5721         *mtu_len = mtu;
5722
5723         return ret;
5724 }
5725
5726 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5727         .func           = bpf_skb_check_mtu,
5728         .gpl_only       = true,
5729         .ret_type       = RET_INTEGER,
5730         .arg1_type      = ARG_PTR_TO_CTX,
5731         .arg2_type      = ARG_ANYTHING,
5732         .arg3_type      = ARG_PTR_TO_INT,
5733         .arg4_type      = ARG_ANYTHING,
5734         .arg5_type      = ARG_ANYTHING,
5735 };
5736
5737 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5738         .func           = bpf_xdp_check_mtu,
5739         .gpl_only       = true,
5740         .ret_type       = RET_INTEGER,
5741         .arg1_type      = ARG_PTR_TO_CTX,
5742         .arg2_type      = ARG_ANYTHING,
5743         .arg3_type      = ARG_PTR_TO_INT,
5744         .arg4_type      = ARG_ANYTHING,
5745         .arg5_type      = ARG_ANYTHING,
5746 };
5747
5748 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5749 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5750 {
5751         int err;
5752         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5753
5754         if (!seg6_validate_srh(srh, len, false))
5755                 return -EINVAL;
5756
5757         switch (type) {
5758         case BPF_LWT_ENCAP_SEG6_INLINE:
5759                 if (skb->protocol != htons(ETH_P_IPV6))
5760                         return -EBADMSG;
5761
5762                 err = seg6_do_srh_inline(skb, srh);
5763                 break;
5764         case BPF_LWT_ENCAP_SEG6:
5765                 skb_reset_inner_headers(skb);
5766                 skb->encapsulation = 1;
5767                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5768                 break;
5769         default:
5770                 return -EINVAL;
5771         }
5772
5773         bpf_compute_data_pointers(skb);
5774         if (err)
5775                 return err;
5776
5777         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5778         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5779
5780         return seg6_lookup_nexthop(skb, NULL, 0);
5781 }
5782 #endif /* CONFIG_IPV6_SEG6_BPF */
5783
5784 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5785 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5786                              bool ingress)
5787 {
5788         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5789 }
5790 #endif
5791
5792 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5793            u32, len)
5794 {
5795         switch (type) {
5796 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5797         case BPF_LWT_ENCAP_SEG6:
5798         case BPF_LWT_ENCAP_SEG6_INLINE:
5799                 return bpf_push_seg6_encap(skb, type, hdr, len);
5800 #endif
5801 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5802         case BPF_LWT_ENCAP_IP:
5803                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5804 #endif
5805         default:
5806                 return -EINVAL;
5807         }
5808 }
5809
5810 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5811            void *, hdr, u32, len)
5812 {
5813         switch (type) {
5814 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5815         case BPF_LWT_ENCAP_IP:
5816                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5817 #endif
5818         default:
5819                 return -EINVAL;
5820         }
5821 }
5822
5823 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5824         .func           = bpf_lwt_in_push_encap,
5825         .gpl_only       = false,
5826         .ret_type       = RET_INTEGER,
5827         .arg1_type      = ARG_PTR_TO_CTX,
5828         .arg2_type      = ARG_ANYTHING,
5829         .arg3_type      = ARG_PTR_TO_MEM,
5830         .arg4_type      = ARG_CONST_SIZE
5831 };
5832
5833 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5834         .func           = bpf_lwt_xmit_push_encap,
5835         .gpl_only       = false,
5836         .ret_type       = RET_INTEGER,
5837         .arg1_type      = ARG_PTR_TO_CTX,
5838         .arg2_type      = ARG_ANYTHING,
5839         .arg3_type      = ARG_PTR_TO_MEM,
5840         .arg4_type      = ARG_CONST_SIZE
5841 };
5842
5843 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5844 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5845            const void *, from, u32, len)
5846 {
5847         struct seg6_bpf_srh_state *srh_state =
5848                 this_cpu_ptr(&seg6_bpf_srh_states);
5849         struct ipv6_sr_hdr *srh = srh_state->srh;
5850         void *srh_tlvs, *srh_end, *ptr;
5851         int srhoff = 0;
5852
5853         if (srh == NULL)
5854                 return -EINVAL;
5855
5856         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5857         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5858
5859         ptr = skb->data + offset;
5860         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5861                 srh_state->valid = false;
5862         else if (ptr < (void *)&srh->flags ||
5863                  ptr + len > (void *)&srh->segments)
5864                 return -EFAULT;
5865
5866         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5867                 return -EFAULT;
5868         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5869                 return -EINVAL;
5870         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5871
5872         memcpy(skb->data + offset, from, len);
5873         return 0;
5874 }
5875
5876 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5877         .func           = bpf_lwt_seg6_store_bytes,
5878         .gpl_only       = false,
5879         .ret_type       = RET_INTEGER,
5880         .arg1_type      = ARG_PTR_TO_CTX,
5881         .arg2_type      = ARG_ANYTHING,
5882         .arg3_type      = ARG_PTR_TO_MEM,
5883         .arg4_type      = ARG_CONST_SIZE
5884 };
5885
5886 static void bpf_update_srh_state(struct sk_buff *skb)
5887 {
5888         struct seg6_bpf_srh_state *srh_state =
5889                 this_cpu_ptr(&seg6_bpf_srh_states);
5890         int srhoff = 0;
5891
5892         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5893                 srh_state->srh = NULL;
5894         } else {
5895                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5896                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5897                 srh_state->valid = true;
5898         }
5899 }
5900
5901 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5902            u32, action, void *, param, u32, param_len)
5903 {
5904         struct seg6_bpf_srh_state *srh_state =
5905                 this_cpu_ptr(&seg6_bpf_srh_states);
5906         int hdroff = 0;
5907         int err;
5908
5909         switch (action) {
5910         case SEG6_LOCAL_ACTION_END_X:
5911                 if (!seg6_bpf_has_valid_srh(skb))
5912                         return -EBADMSG;
5913                 if (param_len != sizeof(struct in6_addr))
5914                         return -EINVAL;
5915                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5916         case SEG6_LOCAL_ACTION_END_T:
5917                 if (!seg6_bpf_has_valid_srh(skb))
5918                         return -EBADMSG;
5919                 if (param_len != sizeof(int))
5920                         return -EINVAL;
5921                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5922         case SEG6_LOCAL_ACTION_END_DT6:
5923                 if (!seg6_bpf_has_valid_srh(skb))
5924                         return -EBADMSG;
5925                 if (param_len != sizeof(int))
5926                         return -EINVAL;
5927
5928                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5929                         return -EBADMSG;
5930                 if (!pskb_pull(skb, hdroff))
5931                         return -EBADMSG;
5932
5933                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5934                 skb_reset_network_header(skb);
5935                 skb_reset_transport_header(skb);
5936                 skb->encapsulation = 0;
5937
5938                 bpf_compute_data_pointers(skb);
5939                 bpf_update_srh_state(skb);
5940                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5941         case SEG6_LOCAL_ACTION_END_B6:
5942                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5943                         return -EBADMSG;
5944                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5945                                           param, param_len);
5946                 if (!err)
5947                         bpf_update_srh_state(skb);
5948
5949                 return err;
5950         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5951                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5952                         return -EBADMSG;
5953                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5954                                           param, param_len);
5955                 if (!err)
5956                         bpf_update_srh_state(skb);
5957
5958                 return err;
5959         default:
5960                 return -EINVAL;
5961         }
5962 }
5963
5964 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5965         .func           = bpf_lwt_seg6_action,
5966         .gpl_only       = false,
5967         .ret_type       = RET_INTEGER,
5968         .arg1_type      = ARG_PTR_TO_CTX,
5969         .arg2_type      = ARG_ANYTHING,
5970         .arg3_type      = ARG_PTR_TO_MEM,
5971         .arg4_type      = ARG_CONST_SIZE
5972 };
5973
5974 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5975            s32, len)
5976 {
5977         struct seg6_bpf_srh_state *srh_state =
5978                 this_cpu_ptr(&seg6_bpf_srh_states);
5979         struct ipv6_sr_hdr *srh = srh_state->srh;
5980         void *srh_end, *srh_tlvs, *ptr;
5981         struct ipv6hdr *hdr;
5982         int srhoff = 0;
5983         int ret;
5984
5985         if (unlikely(srh == NULL))
5986                 return -EINVAL;
5987
5988         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5989                         ((srh->first_segment + 1) << 4));
5990         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5991                         srh_state->hdrlen);
5992         ptr = skb->data + offset;
5993
5994         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5995                 return -EFAULT;
5996         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5997                 return -EFAULT;
5998
5999         if (len > 0) {
6000                 ret = skb_cow_head(skb, len);
6001                 if (unlikely(ret < 0))
6002                         return ret;
6003
6004                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6005         } else {
6006                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6007         }
6008
6009         bpf_compute_data_pointers(skb);
6010         if (unlikely(ret < 0))
6011                 return ret;
6012
6013         hdr = (struct ipv6hdr *)skb->data;
6014         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6015
6016         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6017                 return -EINVAL;
6018         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6019         srh_state->hdrlen += len;
6020         srh_state->valid = false;
6021         return 0;
6022 }
6023
6024 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6025         .func           = bpf_lwt_seg6_adjust_srh,
6026         .gpl_only       = false,
6027         .ret_type       = RET_INTEGER,
6028         .arg1_type      = ARG_PTR_TO_CTX,
6029         .arg2_type      = ARG_ANYTHING,
6030         .arg3_type      = ARG_ANYTHING,
6031 };
6032 #endif /* CONFIG_IPV6_SEG6_BPF */
6033
6034 #ifdef CONFIG_INET
6035 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6036                               int dif, int sdif, u8 family, u8 proto)
6037 {
6038         bool refcounted = false;
6039         struct sock *sk = NULL;
6040
6041         if (family == AF_INET) {
6042                 __be32 src4 = tuple->ipv4.saddr;
6043                 __be32 dst4 = tuple->ipv4.daddr;
6044
6045                 if (proto == IPPROTO_TCP)
6046                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6047                                            src4, tuple->ipv4.sport,
6048                                            dst4, tuple->ipv4.dport,
6049                                            dif, sdif, &refcounted);
6050                 else
6051                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6052                                                dst4, tuple->ipv4.dport,
6053                                                dif, sdif, &udp_table, NULL);
6054 #if IS_ENABLED(CONFIG_IPV6)
6055         } else {
6056                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6057                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6058
6059                 if (proto == IPPROTO_TCP)
6060                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6061                                             src6, tuple->ipv6.sport,
6062                                             dst6, ntohs(tuple->ipv6.dport),
6063                                             dif, sdif, &refcounted);
6064                 else if (likely(ipv6_bpf_stub))
6065                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6066                                                             src6, tuple->ipv6.sport,
6067                                                             dst6, tuple->ipv6.dport,
6068                                                             dif, sdif,
6069                                                             &udp_table, NULL);
6070 #endif
6071         }
6072
6073         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6074                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6075                 sk = NULL;
6076         }
6077         return sk;
6078 }
6079
6080 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6081  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6082  * Returns the socket as an 'unsigned long' to simplify the casting in the
6083  * callers to satisfy BPF_CALL declarations.
6084  */
6085 static struct sock *
6086 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6087                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6088                  u64 flags)
6089 {
6090         struct sock *sk = NULL;
6091         u8 family = AF_UNSPEC;
6092         struct net *net;
6093         int sdif;
6094
6095         if (len == sizeof(tuple->ipv4))
6096                 family = AF_INET;
6097         else if (len == sizeof(tuple->ipv6))
6098                 family = AF_INET6;
6099         else
6100                 return NULL;
6101
6102         if (unlikely(family == AF_UNSPEC || flags ||
6103                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6104                 goto out;
6105
6106         if (family == AF_INET)
6107                 sdif = inet_sdif(skb);
6108         else
6109                 sdif = inet6_sdif(skb);
6110
6111         if ((s32)netns_id < 0) {
6112                 net = caller_net;
6113                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6114         } else {
6115                 net = get_net_ns_by_id(caller_net, netns_id);
6116                 if (unlikely(!net))
6117                         goto out;
6118                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6119                 put_net(net);
6120         }
6121
6122 out:
6123         return sk;
6124 }
6125
6126 static struct sock *
6127 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6128                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6129                 u64 flags)
6130 {
6131         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6132                                            ifindex, proto, netns_id, flags);
6133
6134         if (sk) {
6135                 sk = sk_to_full_sk(sk);
6136                 if (!sk_fullsock(sk)) {
6137                         sock_gen_put(sk);
6138                         return NULL;
6139                 }
6140         }
6141
6142         return sk;
6143 }
6144
6145 static struct sock *
6146 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6147                u8 proto, u64 netns_id, u64 flags)
6148 {
6149         struct net *caller_net;
6150         int ifindex;
6151
6152         if (skb->dev) {
6153                 caller_net = dev_net(skb->dev);
6154                 ifindex = skb->dev->ifindex;
6155         } else {
6156                 caller_net = sock_net(skb->sk);
6157                 ifindex = 0;
6158         }
6159
6160         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6161                                 netns_id, flags);
6162 }
6163
6164 static struct sock *
6165 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6166               u8 proto, u64 netns_id, u64 flags)
6167 {
6168         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6169                                          flags);
6170
6171         if (sk) {
6172                 sk = sk_to_full_sk(sk);
6173                 if (!sk_fullsock(sk)) {
6174                         sock_gen_put(sk);
6175                         return NULL;
6176                 }
6177         }
6178
6179         return sk;
6180 }
6181
6182 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6183            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6184 {
6185         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6186                                              netns_id, flags);
6187 }
6188
6189 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6190         .func           = bpf_skc_lookup_tcp,
6191         .gpl_only       = false,
6192         .pkt_access     = true,
6193         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6194         .arg1_type      = ARG_PTR_TO_CTX,
6195         .arg2_type      = ARG_PTR_TO_MEM,
6196         .arg3_type      = ARG_CONST_SIZE,
6197         .arg4_type      = ARG_ANYTHING,
6198         .arg5_type      = ARG_ANYTHING,
6199 };
6200
6201 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6202            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6203 {
6204         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6205                                             netns_id, flags);
6206 }
6207
6208 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6209         .func           = bpf_sk_lookup_tcp,
6210         .gpl_only       = false,
6211         .pkt_access     = true,
6212         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6213         .arg1_type      = ARG_PTR_TO_CTX,
6214         .arg2_type      = ARG_PTR_TO_MEM,
6215         .arg3_type      = ARG_CONST_SIZE,
6216         .arg4_type      = ARG_ANYTHING,
6217         .arg5_type      = ARG_ANYTHING,
6218 };
6219
6220 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6221            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6222 {
6223         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6224                                             netns_id, flags);
6225 }
6226
6227 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6228         .func           = bpf_sk_lookup_udp,
6229         .gpl_only       = false,
6230         .pkt_access     = true,
6231         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6232         .arg1_type      = ARG_PTR_TO_CTX,
6233         .arg2_type      = ARG_PTR_TO_MEM,
6234         .arg3_type      = ARG_CONST_SIZE,
6235         .arg4_type      = ARG_ANYTHING,
6236         .arg5_type      = ARG_ANYTHING,
6237 };
6238
6239 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6240 {
6241         if (sk && sk_is_refcounted(sk))
6242                 sock_gen_put(sk);
6243         return 0;
6244 }
6245
6246 static const struct bpf_func_proto bpf_sk_release_proto = {
6247         .func           = bpf_sk_release,
6248         .gpl_only       = false,
6249         .ret_type       = RET_INTEGER,
6250         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6251 };
6252
6253 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6254            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6255 {
6256         struct net *caller_net = dev_net(ctx->rxq->dev);
6257         int ifindex = ctx->rxq->dev->ifindex;
6258
6259         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6260                                               ifindex, IPPROTO_UDP, netns_id,
6261                                               flags);
6262 }
6263
6264 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6265         .func           = bpf_xdp_sk_lookup_udp,
6266         .gpl_only       = false,
6267         .pkt_access     = true,
6268         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6269         .arg1_type      = ARG_PTR_TO_CTX,
6270         .arg2_type      = ARG_PTR_TO_MEM,
6271         .arg3_type      = ARG_CONST_SIZE,
6272         .arg4_type      = ARG_ANYTHING,
6273         .arg5_type      = ARG_ANYTHING,
6274 };
6275
6276 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6277            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6278 {
6279         struct net *caller_net = dev_net(ctx->rxq->dev);
6280         int ifindex = ctx->rxq->dev->ifindex;
6281
6282         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6283                                                ifindex, IPPROTO_TCP, netns_id,
6284                                                flags);
6285 }
6286
6287 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6288         .func           = bpf_xdp_skc_lookup_tcp,
6289         .gpl_only       = false,
6290         .pkt_access     = true,
6291         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6292         .arg1_type      = ARG_PTR_TO_CTX,
6293         .arg2_type      = ARG_PTR_TO_MEM,
6294         .arg3_type      = ARG_CONST_SIZE,
6295         .arg4_type      = ARG_ANYTHING,
6296         .arg5_type      = ARG_ANYTHING,
6297 };
6298
6299 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6300            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6301 {
6302         struct net *caller_net = dev_net(ctx->rxq->dev);
6303         int ifindex = ctx->rxq->dev->ifindex;
6304
6305         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6306                                               ifindex, IPPROTO_TCP, netns_id,
6307                                               flags);
6308 }
6309
6310 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6311         .func           = bpf_xdp_sk_lookup_tcp,
6312         .gpl_only       = false,
6313         .pkt_access     = true,
6314         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6315         .arg1_type      = ARG_PTR_TO_CTX,
6316         .arg2_type      = ARG_PTR_TO_MEM,
6317         .arg3_type      = ARG_CONST_SIZE,
6318         .arg4_type      = ARG_ANYTHING,
6319         .arg5_type      = ARG_ANYTHING,
6320 };
6321
6322 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6323            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6324 {
6325         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6326                                                sock_net(ctx->sk), 0,
6327                                                IPPROTO_TCP, netns_id, flags);
6328 }
6329
6330 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6331         .func           = bpf_sock_addr_skc_lookup_tcp,
6332         .gpl_only       = false,
6333         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6334         .arg1_type      = ARG_PTR_TO_CTX,
6335         .arg2_type      = ARG_PTR_TO_MEM,
6336         .arg3_type      = ARG_CONST_SIZE,
6337         .arg4_type      = ARG_ANYTHING,
6338         .arg5_type      = ARG_ANYTHING,
6339 };
6340
6341 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6342            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6343 {
6344         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6345                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6346                                               netns_id, flags);
6347 }
6348
6349 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6350         .func           = bpf_sock_addr_sk_lookup_tcp,
6351         .gpl_only       = false,
6352         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6353         .arg1_type      = ARG_PTR_TO_CTX,
6354         .arg2_type      = ARG_PTR_TO_MEM,
6355         .arg3_type      = ARG_CONST_SIZE,
6356         .arg4_type      = ARG_ANYTHING,
6357         .arg5_type      = ARG_ANYTHING,
6358 };
6359
6360 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6361            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6362 {
6363         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6364                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6365                                               netns_id, flags);
6366 }
6367
6368 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6369         .func           = bpf_sock_addr_sk_lookup_udp,
6370         .gpl_only       = false,
6371         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6372         .arg1_type      = ARG_PTR_TO_CTX,
6373         .arg2_type      = ARG_PTR_TO_MEM,
6374         .arg3_type      = ARG_CONST_SIZE,
6375         .arg4_type      = ARG_ANYTHING,
6376         .arg5_type      = ARG_ANYTHING,
6377 };
6378
6379 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6380                                   struct bpf_insn_access_aux *info)
6381 {
6382         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6383                                           icsk_retransmits))
6384                 return false;
6385
6386         if (off % size != 0)
6387                 return false;
6388
6389         switch (off) {
6390         case offsetof(struct bpf_tcp_sock, bytes_received):
6391         case offsetof(struct bpf_tcp_sock, bytes_acked):
6392                 return size == sizeof(__u64);
6393         default:
6394                 return size == sizeof(__u32);
6395         }
6396 }
6397
6398 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6399                                     const struct bpf_insn *si,
6400                                     struct bpf_insn *insn_buf,
6401                                     struct bpf_prog *prog, u32 *target_size)
6402 {
6403         struct bpf_insn *insn = insn_buf;
6404
6405 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6406         do {                                                            \
6407                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6408                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6409                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6410                                       si->dst_reg, si->src_reg,         \
6411                                       offsetof(struct tcp_sock, FIELD)); \
6412         } while (0)
6413
6414 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6415         do {                                                            \
6416                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6417                                           FIELD) >                      \
6418                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6419                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6420                                         struct inet_connection_sock,    \
6421                                         FIELD),                         \
6422                                       si->dst_reg, si->src_reg,         \
6423                                       offsetof(                         \
6424                                         struct inet_connection_sock,    \
6425                                         FIELD));                        \
6426         } while (0)
6427
6428         if (insn > insn_buf)
6429                 return insn - insn_buf;
6430
6431         switch (si->off) {
6432         case offsetof(struct bpf_tcp_sock, rtt_min):
6433                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6434                              sizeof(struct minmax));
6435                 BUILD_BUG_ON(sizeof(struct minmax) <
6436                              sizeof(struct minmax_sample));
6437
6438                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6439                                       offsetof(struct tcp_sock, rtt_min) +
6440                                       offsetof(struct minmax_sample, v));
6441                 break;
6442         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6443                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6444                 break;
6445         case offsetof(struct bpf_tcp_sock, srtt_us):
6446                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6447                 break;
6448         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6449                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6450                 break;
6451         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6452                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6453                 break;
6454         case offsetof(struct bpf_tcp_sock, snd_nxt):
6455                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6456                 break;
6457         case offsetof(struct bpf_tcp_sock, snd_una):
6458                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6459                 break;
6460         case offsetof(struct bpf_tcp_sock, mss_cache):
6461                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6462                 break;
6463         case offsetof(struct bpf_tcp_sock, ecn_flags):
6464                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6465                 break;
6466         case offsetof(struct bpf_tcp_sock, rate_delivered):
6467                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6468                 break;
6469         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6470                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6471                 break;
6472         case offsetof(struct bpf_tcp_sock, packets_out):
6473                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6474                 break;
6475         case offsetof(struct bpf_tcp_sock, retrans_out):
6476                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6477                 break;
6478         case offsetof(struct bpf_tcp_sock, total_retrans):
6479                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6480                 break;
6481         case offsetof(struct bpf_tcp_sock, segs_in):
6482                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6483                 break;
6484         case offsetof(struct bpf_tcp_sock, data_segs_in):
6485                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6486                 break;
6487         case offsetof(struct bpf_tcp_sock, segs_out):
6488                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6489                 break;
6490         case offsetof(struct bpf_tcp_sock, data_segs_out):
6491                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6492                 break;
6493         case offsetof(struct bpf_tcp_sock, lost_out):
6494                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6495                 break;
6496         case offsetof(struct bpf_tcp_sock, sacked_out):
6497                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6498                 break;
6499         case offsetof(struct bpf_tcp_sock, bytes_received):
6500                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6501                 break;
6502         case offsetof(struct bpf_tcp_sock, bytes_acked):
6503                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6504                 break;
6505         case offsetof(struct bpf_tcp_sock, dsack_dups):
6506                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6507                 break;
6508         case offsetof(struct bpf_tcp_sock, delivered):
6509                 BPF_TCP_SOCK_GET_COMMON(delivered);
6510                 break;
6511         case offsetof(struct bpf_tcp_sock, delivered_ce):
6512                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6513                 break;
6514         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6515                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6516                 break;
6517         }
6518
6519         return insn - insn_buf;
6520 }
6521
6522 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6523 {
6524         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6525                 return (unsigned long)sk;
6526
6527         return (unsigned long)NULL;
6528 }
6529
6530 const struct bpf_func_proto bpf_tcp_sock_proto = {
6531         .func           = bpf_tcp_sock,
6532         .gpl_only       = false,
6533         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6534         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6535 };
6536
6537 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6538 {
6539         sk = sk_to_full_sk(sk);
6540
6541         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6542                 return (unsigned long)sk;
6543
6544         return (unsigned long)NULL;
6545 }
6546
6547 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6548         .func           = bpf_get_listener_sock,
6549         .gpl_only       = false,
6550         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6551         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6552 };
6553
6554 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6555 {
6556         unsigned int iphdr_len;
6557
6558         switch (skb_protocol(skb, true)) {
6559         case cpu_to_be16(ETH_P_IP):
6560                 iphdr_len = sizeof(struct iphdr);
6561                 break;
6562         case cpu_to_be16(ETH_P_IPV6):
6563                 iphdr_len = sizeof(struct ipv6hdr);
6564                 break;
6565         default:
6566                 return 0;
6567         }
6568
6569         if (skb_headlen(skb) < iphdr_len)
6570                 return 0;
6571
6572         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6573                 return 0;
6574
6575         return INET_ECN_set_ce(skb);
6576 }
6577
6578 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6579                                   struct bpf_insn_access_aux *info)
6580 {
6581         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6582                 return false;
6583
6584         if (off % size != 0)
6585                 return false;
6586
6587         switch (off) {
6588         default:
6589                 return size == sizeof(__u32);
6590         }
6591 }
6592
6593 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6594                                     const struct bpf_insn *si,
6595                                     struct bpf_insn *insn_buf,
6596                                     struct bpf_prog *prog, u32 *target_size)
6597 {
6598         struct bpf_insn *insn = insn_buf;
6599
6600 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6601         do {                                                            \
6602                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6603                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6604                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6605                                       si->dst_reg, si->src_reg,         \
6606                                       offsetof(struct xdp_sock, FIELD)); \
6607         } while (0)
6608
6609         switch (si->off) {
6610         case offsetof(struct bpf_xdp_sock, queue_id):
6611                 BPF_XDP_SOCK_GET(queue_id);
6612                 break;
6613         }
6614
6615         return insn - insn_buf;
6616 }
6617
6618 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6619         .func           = bpf_skb_ecn_set_ce,
6620         .gpl_only       = false,
6621         .ret_type       = RET_INTEGER,
6622         .arg1_type      = ARG_PTR_TO_CTX,
6623 };
6624
6625 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6626            struct tcphdr *, th, u32, th_len)
6627 {
6628 #ifdef CONFIG_SYN_COOKIES
6629         u32 cookie;
6630         int ret;
6631
6632         if (unlikely(!sk || th_len < sizeof(*th)))
6633                 return -EINVAL;
6634
6635         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6636         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6637                 return -EINVAL;
6638
6639         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6640                 return -EINVAL;
6641
6642         if (!th->ack || th->rst || th->syn)
6643                 return -ENOENT;
6644
6645         if (tcp_synq_no_recent_overflow(sk))
6646                 return -ENOENT;
6647
6648         cookie = ntohl(th->ack_seq) - 1;
6649
6650         switch (sk->sk_family) {
6651         case AF_INET:
6652                 if (unlikely(iph_len < sizeof(struct iphdr)))
6653                         return -EINVAL;
6654
6655                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6656                 break;
6657
6658 #if IS_BUILTIN(CONFIG_IPV6)
6659         case AF_INET6:
6660                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6661                         return -EINVAL;
6662
6663                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6664                 break;
6665 #endif /* CONFIG_IPV6 */
6666
6667         default:
6668                 return -EPROTONOSUPPORT;
6669         }
6670
6671         if (ret > 0)
6672                 return 0;
6673
6674         return -ENOENT;
6675 #else
6676         return -ENOTSUPP;
6677 #endif
6678 }
6679
6680 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6681         .func           = bpf_tcp_check_syncookie,
6682         .gpl_only       = true,
6683         .pkt_access     = true,
6684         .ret_type       = RET_INTEGER,
6685         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6686         .arg2_type      = ARG_PTR_TO_MEM,
6687         .arg3_type      = ARG_CONST_SIZE,
6688         .arg4_type      = ARG_PTR_TO_MEM,
6689         .arg5_type      = ARG_CONST_SIZE,
6690 };
6691
6692 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6693            struct tcphdr *, th, u32, th_len)
6694 {
6695 #ifdef CONFIG_SYN_COOKIES
6696         u32 cookie;
6697         u16 mss;
6698
6699         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6700                 return -EINVAL;
6701
6702         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6703                 return -EINVAL;
6704
6705         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6706                 return -ENOENT;
6707
6708         if (!th->syn || th->ack || th->fin || th->rst)
6709                 return -EINVAL;
6710
6711         if (unlikely(iph_len < sizeof(struct iphdr)))
6712                 return -EINVAL;
6713
6714         /* Both struct iphdr and struct ipv6hdr have the version field at the
6715          * same offset so we can cast to the shorter header (struct iphdr).
6716          */
6717         switch (((struct iphdr *)iph)->version) {
6718         case 4:
6719                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6720                         return -EINVAL;
6721
6722                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6723                 break;
6724
6725 #if IS_BUILTIN(CONFIG_IPV6)
6726         case 6:
6727                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6728                         return -EINVAL;
6729
6730                 if (sk->sk_family != AF_INET6)
6731                         return -EINVAL;
6732
6733                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6734                 break;
6735 #endif /* CONFIG_IPV6 */
6736
6737         default:
6738                 return -EPROTONOSUPPORT;
6739         }
6740         if (mss == 0)
6741                 return -ENOENT;
6742
6743         return cookie | ((u64)mss << 32);
6744 #else
6745         return -EOPNOTSUPP;
6746 #endif /* CONFIG_SYN_COOKIES */
6747 }
6748
6749 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6750         .func           = bpf_tcp_gen_syncookie,
6751         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6752         .pkt_access     = true,
6753         .ret_type       = RET_INTEGER,
6754         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6755         .arg2_type      = ARG_PTR_TO_MEM,
6756         .arg3_type      = ARG_CONST_SIZE,
6757         .arg4_type      = ARG_PTR_TO_MEM,
6758         .arg5_type      = ARG_CONST_SIZE,
6759 };
6760
6761 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6762 {
6763         if (!sk || flags != 0)
6764                 return -EINVAL;
6765         if (!skb_at_tc_ingress(skb))
6766                 return -EOPNOTSUPP;
6767         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6768                 return -ENETUNREACH;
6769         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6770                 return -ESOCKTNOSUPPORT;
6771         if (sk_is_refcounted(sk) &&
6772             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6773                 return -ENOENT;
6774
6775         skb_orphan(skb);
6776         skb->sk = sk;
6777         skb->destructor = sock_pfree;
6778
6779         return 0;
6780 }
6781
6782 static const struct bpf_func_proto bpf_sk_assign_proto = {
6783         .func           = bpf_sk_assign,
6784         .gpl_only       = false,
6785         .ret_type       = RET_INTEGER,
6786         .arg1_type      = ARG_PTR_TO_CTX,
6787         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6788         .arg3_type      = ARG_ANYTHING,
6789 };
6790
6791 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6792                                     u8 search_kind, const u8 *magic,
6793                                     u8 magic_len, bool *eol)
6794 {
6795         u8 kind, kind_len;
6796
6797         *eol = false;
6798
6799         while (op < opend) {
6800                 kind = op[0];
6801
6802                 if (kind == TCPOPT_EOL) {
6803                         *eol = true;
6804                         return ERR_PTR(-ENOMSG);
6805                 } else if (kind == TCPOPT_NOP) {
6806                         op++;
6807                         continue;
6808                 }
6809
6810                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6811                         /* Something is wrong in the received header.
6812                          * Follow the TCP stack's tcp_parse_options()
6813                          * and just bail here.
6814                          */
6815                         return ERR_PTR(-EFAULT);
6816
6817                 kind_len = op[1];
6818                 if (search_kind == kind) {
6819                         if (!magic_len)
6820                                 return op;
6821
6822                         if (magic_len > kind_len - 2)
6823                                 return ERR_PTR(-ENOMSG);
6824
6825                         if (!memcmp(&op[2], magic, magic_len))
6826                                 return op;
6827                 }
6828
6829                 op += kind_len;
6830         }
6831
6832         return ERR_PTR(-ENOMSG);
6833 }
6834
6835 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6836            void *, search_res, u32, len, u64, flags)
6837 {
6838         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6839         const u8 *op, *opend, *magic, *search = search_res;
6840         u8 search_kind, search_len, copy_len, magic_len;
6841         int ret;
6842
6843         /* 2 byte is the minimal option len except TCPOPT_NOP and
6844          * TCPOPT_EOL which are useless for the bpf prog to learn
6845          * and this helper disallow loading them also.
6846          */
6847         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6848                 return -EINVAL;
6849
6850         search_kind = search[0];
6851         search_len = search[1];
6852
6853         if (search_len > len || search_kind == TCPOPT_NOP ||
6854             search_kind == TCPOPT_EOL)
6855                 return -EINVAL;
6856
6857         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6858                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6859                 if (search_len != 4 && search_len != 6)
6860                         return -EINVAL;
6861                 magic = &search[2];
6862                 magic_len = search_len - 2;
6863         } else {
6864                 if (search_len)
6865                         return -EINVAL;
6866                 magic = NULL;
6867                 magic_len = 0;
6868         }
6869
6870         if (load_syn) {
6871                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6872                 if (ret < 0)
6873                         return ret;
6874
6875                 opend = op + ret;
6876                 op += sizeof(struct tcphdr);
6877         } else {
6878                 if (!bpf_sock->skb ||
6879                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6880                         /* This bpf_sock->op cannot call this helper */
6881                         return -EPERM;
6882
6883                 opend = bpf_sock->skb_data_end;
6884                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6885         }
6886
6887         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6888                                 &eol);
6889         if (IS_ERR(op))
6890                 return PTR_ERR(op);
6891
6892         copy_len = op[1];
6893         ret = copy_len;
6894         if (copy_len > len) {
6895                 ret = -ENOSPC;
6896                 copy_len = len;
6897         }
6898
6899         memcpy(search_res, op, copy_len);
6900         return ret;
6901 }
6902
6903 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6904         .func           = bpf_sock_ops_load_hdr_opt,
6905         .gpl_only       = false,
6906         .ret_type       = RET_INTEGER,
6907         .arg1_type      = ARG_PTR_TO_CTX,
6908         .arg2_type      = ARG_PTR_TO_MEM,
6909         .arg3_type      = ARG_CONST_SIZE,
6910         .arg4_type      = ARG_ANYTHING,
6911 };
6912
6913 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6914            const void *, from, u32, len, u64, flags)
6915 {
6916         u8 new_kind, new_kind_len, magic_len = 0, *opend;
6917         const u8 *op, *new_op, *magic = NULL;
6918         struct sk_buff *skb;
6919         bool eol;
6920
6921         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6922                 return -EPERM;
6923
6924         if (len < 2 || flags)
6925                 return -EINVAL;
6926
6927         new_op = from;
6928         new_kind = new_op[0];
6929         new_kind_len = new_op[1];
6930
6931         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6932             new_kind == TCPOPT_EOL)
6933                 return -EINVAL;
6934
6935         if (new_kind_len > bpf_sock->remaining_opt_len)
6936                 return -ENOSPC;
6937
6938         /* 253 is another experimental kind */
6939         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
6940                 if (new_kind_len < 4)
6941                         return -EINVAL;
6942                 /* Match for the 2 byte magic also.
6943                  * RFC 6994: the magic could be 2 or 4 bytes.
6944                  * Hence, matching by 2 byte only is on the
6945                  * conservative side but it is the right
6946                  * thing to do for the 'search-for-duplication'
6947                  * purpose.
6948                  */
6949                 magic = &new_op[2];
6950                 magic_len = 2;
6951         }
6952
6953         /* Check for duplication */
6954         skb = bpf_sock->skb;
6955         op = skb->data + sizeof(struct tcphdr);
6956         opend = bpf_sock->skb_data_end;
6957
6958         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6959                                 &eol);
6960         if (!IS_ERR(op))
6961                 return -EEXIST;
6962
6963         if (PTR_ERR(op) != -ENOMSG)
6964                 return PTR_ERR(op);
6965
6966         if (eol)
6967                 /* The option has been ended.  Treat it as no more
6968                  * header option can be written.
6969                  */
6970                 return -ENOSPC;
6971
6972         /* No duplication found.  Store the header option. */
6973         memcpy(opend, from, new_kind_len);
6974
6975         bpf_sock->remaining_opt_len -= new_kind_len;
6976         bpf_sock->skb_data_end += new_kind_len;
6977
6978         return 0;
6979 }
6980
6981 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6982         .func           = bpf_sock_ops_store_hdr_opt,
6983         .gpl_only       = false,
6984         .ret_type       = RET_INTEGER,
6985         .arg1_type      = ARG_PTR_TO_CTX,
6986         .arg2_type      = ARG_PTR_TO_MEM,
6987         .arg3_type      = ARG_CONST_SIZE,
6988         .arg4_type      = ARG_ANYTHING,
6989 };
6990
6991 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6992            u32, len, u64, flags)
6993 {
6994         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6995                 return -EPERM;
6996
6997         if (flags || len < 2)
6998                 return -EINVAL;
6999
7000         if (len > bpf_sock->remaining_opt_len)
7001                 return -ENOSPC;
7002
7003         bpf_sock->remaining_opt_len -= len;
7004
7005         return 0;
7006 }
7007
7008 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7009         .func           = bpf_sock_ops_reserve_hdr_opt,
7010         .gpl_only       = false,
7011         .ret_type       = RET_INTEGER,
7012         .arg1_type      = ARG_PTR_TO_CTX,
7013         .arg2_type      = ARG_ANYTHING,
7014         .arg3_type      = ARG_ANYTHING,
7015 };
7016
7017 #endif /* CONFIG_INET */
7018
7019 bool bpf_helper_changes_pkt_data(void *func)
7020 {
7021         if (func == bpf_skb_vlan_push ||
7022             func == bpf_skb_vlan_pop ||
7023             func == bpf_skb_store_bytes ||
7024             func == bpf_skb_change_proto ||
7025             func == bpf_skb_change_head ||
7026             func == sk_skb_change_head ||
7027             func == bpf_skb_change_tail ||
7028             func == sk_skb_change_tail ||
7029             func == bpf_skb_adjust_room ||
7030             func == sk_skb_adjust_room ||
7031             func == bpf_skb_pull_data ||
7032             func == sk_skb_pull_data ||
7033             func == bpf_clone_redirect ||
7034             func == bpf_l3_csum_replace ||
7035             func == bpf_l4_csum_replace ||
7036             func == bpf_xdp_adjust_head ||
7037             func == bpf_xdp_adjust_meta ||
7038             func == bpf_msg_pull_data ||
7039             func == bpf_msg_push_data ||
7040             func == bpf_msg_pop_data ||
7041             func == bpf_xdp_adjust_tail ||
7042 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7043             func == bpf_lwt_seg6_store_bytes ||
7044             func == bpf_lwt_seg6_adjust_srh ||
7045             func == bpf_lwt_seg6_action ||
7046 #endif
7047 #ifdef CONFIG_INET
7048             func == bpf_sock_ops_store_hdr_opt ||
7049 #endif
7050             func == bpf_lwt_in_push_encap ||
7051             func == bpf_lwt_xmit_push_encap)
7052                 return true;
7053
7054         return false;
7055 }
7056
7057 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7058 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7059
7060 static const struct bpf_func_proto *
7061 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7062 {
7063         switch (func_id) {
7064         /* inet and inet6 sockets are created in a process
7065          * context so there is always a valid uid/gid
7066          */
7067         case BPF_FUNC_get_current_uid_gid:
7068                 return &bpf_get_current_uid_gid_proto;
7069         case BPF_FUNC_get_local_storage:
7070                 return &bpf_get_local_storage_proto;
7071         case BPF_FUNC_get_socket_cookie:
7072                 return &bpf_get_socket_cookie_sock_proto;
7073         case BPF_FUNC_get_netns_cookie:
7074                 return &bpf_get_netns_cookie_sock_proto;
7075         case BPF_FUNC_perf_event_output:
7076                 return &bpf_event_output_data_proto;
7077         case BPF_FUNC_get_current_pid_tgid:
7078                 return &bpf_get_current_pid_tgid_proto;
7079         case BPF_FUNC_get_current_comm:
7080                 return &bpf_get_current_comm_proto;
7081 #ifdef CONFIG_CGROUPS
7082         case BPF_FUNC_get_current_cgroup_id:
7083                 return &bpf_get_current_cgroup_id_proto;
7084         case BPF_FUNC_get_current_ancestor_cgroup_id:
7085                 return &bpf_get_current_ancestor_cgroup_id_proto;
7086 #endif
7087 #ifdef CONFIG_CGROUP_NET_CLASSID
7088         case BPF_FUNC_get_cgroup_classid:
7089                 return &bpf_get_cgroup_classid_curr_proto;
7090 #endif
7091         case BPF_FUNC_sk_storage_get:
7092                 return &bpf_sk_storage_get_cg_sock_proto;
7093         default:
7094                 return bpf_base_func_proto(func_id);
7095         }
7096 }
7097
7098 static const struct bpf_func_proto *
7099 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7100 {
7101         switch (func_id) {
7102         /* inet and inet6 sockets are created in a process
7103          * context so there is always a valid uid/gid
7104          */
7105         case BPF_FUNC_get_current_uid_gid:
7106                 return &bpf_get_current_uid_gid_proto;
7107         case BPF_FUNC_bind:
7108                 switch (prog->expected_attach_type) {
7109                 case BPF_CGROUP_INET4_CONNECT:
7110                 case BPF_CGROUP_INET6_CONNECT:
7111                         return &bpf_bind_proto;
7112                 default:
7113                         return NULL;
7114                 }
7115         case BPF_FUNC_get_socket_cookie:
7116                 return &bpf_get_socket_cookie_sock_addr_proto;
7117         case BPF_FUNC_get_netns_cookie:
7118                 return &bpf_get_netns_cookie_sock_addr_proto;
7119         case BPF_FUNC_get_local_storage:
7120                 return &bpf_get_local_storage_proto;
7121         case BPF_FUNC_perf_event_output:
7122                 return &bpf_event_output_data_proto;
7123         case BPF_FUNC_get_current_pid_tgid:
7124                 return &bpf_get_current_pid_tgid_proto;
7125         case BPF_FUNC_get_current_comm:
7126                 return &bpf_get_current_comm_proto;
7127 #ifdef CONFIG_CGROUPS
7128         case BPF_FUNC_get_current_cgroup_id:
7129                 return &bpf_get_current_cgroup_id_proto;
7130         case BPF_FUNC_get_current_ancestor_cgroup_id:
7131                 return &bpf_get_current_ancestor_cgroup_id_proto;
7132 #endif
7133 #ifdef CONFIG_CGROUP_NET_CLASSID
7134         case BPF_FUNC_get_cgroup_classid:
7135                 return &bpf_get_cgroup_classid_curr_proto;
7136 #endif
7137 #ifdef CONFIG_INET
7138         case BPF_FUNC_sk_lookup_tcp:
7139                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7140         case BPF_FUNC_sk_lookup_udp:
7141                 return &bpf_sock_addr_sk_lookup_udp_proto;
7142         case BPF_FUNC_sk_release:
7143                 return &bpf_sk_release_proto;
7144         case BPF_FUNC_skc_lookup_tcp:
7145                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7146 #endif /* CONFIG_INET */
7147         case BPF_FUNC_sk_storage_get:
7148                 return &bpf_sk_storage_get_proto;
7149         case BPF_FUNC_sk_storage_delete:
7150                 return &bpf_sk_storage_delete_proto;
7151         case BPF_FUNC_setsockopt:
7152                 switch (prog->expected_attach_type) {
7153                 case BPF_CGROUP_INET4_BIND:
7154                 case BPF_CGROUP_INET6_BIND:
7155                 case BPF_CGROUP_INET4_CONNECT:
7156                 case BPF_CGROUP_INET6_CONNECT:
7157                 case BPF_CGROUP_UDP4_RECVMSG:
7158                 case BPF_CGROUP_UDP6_RECVMSG:
7159                 case BPF_CGROUP_UDP4_SENDMSG:
7160                 case BPF_CGROUP_UDP6_SENDMSG:
7161                 case BPF_CGROUP_INET4_GETPEERNAME:
7162                 case BPF_CGROUP_INET6_GETPEERNAME:
7163                 case BPF_CGROUP_INET4_GETSOCKNAME:
7164                 case BPF_CGROUP_INET6_GETSOCKNAME:
7165                         return &bpf_sock_addr_setsockopt_proto;
7166                 default:
7167                         return NULL;
7168                 }
7169         case BPF_FUNC_getsockopt:
7170                 switch (prog->expected_attach_type) {
7171                 case BPF_CGROUP_INET4_BIND:
7172                 case BPF_CGROUP_INET6_BIND:
7173                 case BPF_CGROUP_INET4_CONNECT:
7174                 case BPF_CGROUP_INET6_CONNECT:
7175                 case BPF_CGROUP_UDP4_RECVMSG:
7176                 case BPF_CGROUP_UDP6_RECVMSG:
7177                 case BPF_CGROUP_UDP4_SENDMSG:
7178                 case BPF_CGROUP_UDP6_SENDMSG:
7179                 case BPF_CGROUP_INET4_GETPEERNAME:
7180                 case BPF_CGROUP_INET6_GETPEERNAME:
7181                 case BPF_CGROUP_INET4_GETSOCKNAME:
7182                 case BPF_CGROUP_INET6_GETSOCKNAME:
7183                         return &bpf_sock_addr_getsockopt_proto;
7184                 default:
7185                         return NULL;
7186                 }
7187         default:
7188                 return bpf_sk_base_func_proto(func_id);
7189         }
7190 }
7191
7192 static const struct bpf_func_proto *
7193 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7194 {
7195         switch (func_id) {
7196         case BPF_FUNC_skb_load_bytes:
7197                 return &bpf_skb_load_bytes_proto;
7198         case BPF_FUNC_skb_load_bytes_relative:
7199                 return &bpf_skb_load_bytes_relative_proto;
7200         case BPF_FUNC_get_socket_cookie:
7201                 return &bpf_get_socket_cookie_proto;
7202         case BPF_FUNC_get_socket_uid:
7203                 return &bpf_get_socket_uid_proto;
7204         case BPF_FUNC_perf_event_output:
7205                 return &bpf_skb_event_output_proto;
7206         default:
7207                 return bpf_sk_base_func_proto(func_id);
7208         }
7209 }
7210
7211 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7212 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7213
7214 static const struct bpf_func_proto *
7215 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7216 {
7217         switch (func_id) {
7218         case BPF_FUNC_get_local_storage:
7219                 return &bpf_get_local_storage_proto;
7220         case BPF_FUNC_sk_fullsock:
7221                 return &bpf_sk_fullsock_proto;
7222         case BPF_FUNC_sk_storage_get:
7223                 return &bpf_sk_storage_get_proto;
7224         case BPF_FUNC_sk_storage_delete:
7225                 return &bpf_sk_storage_delete_proto;
7226         case BPF_FUNC_perf_event_output:
7227                 return &bpf_skb_event_output_proto;
7228 #ifdef CONFIG_SOCK_CGROUP_DATA
7229         case BPF_FUNC_skb_cgroup_id:
7230                 return &bpf_skb_cgroup_id_proto;
7231         case BPF_FUNC_skb_ancestor_cgroup_id:
7232                 return &bpf_skb_ancestor_cgroup_id_proto;
7233         case BPF_FUNC_sk_cgroup_id:
7234                 return &bpf_sk_cgroup_id_proto;
7235         case BPF_FUNC_sk_ancestor_cgroup_id:
7236                 return &bpf_sk_ancestor_cgroup_id_proto;
7237 #endif
7238 #ifdef CONFIG_INET
7239         case BPF_FUNC_sk_lookup_tcp:
7240                 return &bpf_sk_lookup_tcp_proto;
7241         case BPF_FUNC_sk_lookup_udp:
7242                 return &bpf_sk_lookup_udp_proto;
7243         case BPF_FUNC_sk_release:
7244                 return &bpf_sk_release_proto;
7245         case BPF_FUNC_skc_lookup_tcp:
7246                 return &bpf_skc_lookup_tcp_proto;
7247         case BPF_FUNC_tcp_sock:
7248                 return &bpf_tcp_sock_proto;
7249         case BPF_FUNC_get_listener_sock:
7250                 return &bpf_get_listener_sock_proto;
7251         case BPF_FUNC_skb_ecn_set_ce:
7252                 return &bpf_skb_ecn_set_ce_proto;
7253 #endif
7254         default:
7255                 return sk_filter_func_proto(func_id, prog);
7256         }
7257 }
7258
7259 static const struct bpf_func_proto *
7260 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7261 {
7262         switch (func_id) {
7263         case BPF_FUNC_skb_store_bytes:
7264                 return &bpf_skb_store_bytes_proto;
7265         case BPF_FUNC_skb_load_bytes:
7266                 return &bpf_skb_load_bytes_proto;
7267         case BPF_FUNC_skb_load_bytes_relative:
7268                 return &bpf_skb_load_bytes_relative_proto;
7269         case BPF_FUNC_skb_pull_data:
7270                 return &bpf_skb_pull_data_proto;
7271         case BPF_FUNC_csum_diff:
7272                 return &bpf_csum_diff_proto;
7273         case BPF_FUNC_csum_update:
7274                 return &bpf_csum_update_proto;
7275         case BPF_FUNC_csum_level:
7276                 return &bpf_csum_level_proto;
7277         case BPF_FUNC_l3_csum_replace:
7278                 return &bpf_l3_csum_replace_proto;
7279         case BPF_FUNC_l4_csum_replace:
7280                 return &bpf_l4_csum_replace_proto;
7281         case BPF_FUNC_clone_redirect:
7282                 return &bpf_clone_redirect_proto;
7283         case BPF_FUNC_get_cgroup_classid:
7284                 return &bpf_get_cgroup_classid_proto;
7285         case BPF_FUNC_skb_vlan_push:
7286                 return &bpf_skb_vlan_push_proto;
7287         case BPF_FUNC_skb_vlan_pop:
7288                 return &bpf_skb_vlan_pop_proto;
7289         case BPF_FUNC_skb_change_proto:
7290                 return &bpf_skb_change_proto_proto;
7291         case BPF_FUNC_skb_change_type:
7292                 return &bpf_skb_change_type_proto;
7293         case BPF_FUNC_skb_adjust_room:
7294                 return &bpf_skb_adjust_room_proto;
7295         case BPF_FUNC_skb_change_tail:
7296                 return &bpf_skb_change_tail_proto;
7297         case BPF_FUNC_skb_change_head:
7298                 return &bpf_skb_change_head_proto;
7299         case BPF_FUNC_skb_get_tunnel_key:
7300                 return &bpf_skb_get_tunnel_key_proto;
7301         case BPF_FUNC_skb_set_tunnel_key:
7302                 return bpf_get_skb_set_tunnel_proto(func_id);
7303         case BPF_FUNC_skb_get_tunnel_opt:
7304                 return &bpf_skb_get_tunnel_opt_proto;
7305         case BPF_FUNC_skb_set_tunnel_opt:
7306                 return bpf_get_skb_set_tunnel_proto(func_id);
7307         case BPF_FUNC_redirect:
7308                 return &bpf_redirect_proto;
7309         case BPF_FUNC_redirect_neigh:
7310                 return &bpf_redirect_neigh_proto;
7311         case BPF_FUNC_redirect_peer:
7312                 return &bpf_redirect_peer_proto;
7313         case BPF_FUNC_get_route_realm:
7314                 return &bpf_get_route_realm_proto;
7315         case BPF_FUNC_get_hash_recalc:
7316                 return &bpf_get_hash_recalc_proto;
7317         case BPF_FUNC_set_hash_invalid:
7318                 return &bpf_set_hash_invalid_proto;
7319         case BPF_FUNC_set_hash:
7320                 return &bpf_set_hash_proto;
7321         case BPF_FUNC_perf_event_output:
7322                 return &bpf_skb_event_output_proto;
7323         case BPF_FUNC_get_smp_processor_id:
7324                 return &bpf_get_smp_processor_id_proto;
7325         case BPF_FUNC_skb_under_cgroup:
7326                 return &bpf_skb_under_cgroup_proto;
7327         case BPF_FUNC_get_socket_cookie:
7328                 return &bpf_get_socket_cookie_proto;
7329         case BPF_FUNC_get_socket_uid:
7330                 return &bpf_get_socket_uid_proto;
7331         case BPF_FUNC_fib_lookup:
7332                 return &bpf_skb_fib_lookup_proto;
7333         case BPF_FUNC_check_mtu:
7334                 return &bpf_skb_check_mtu_proto;
7335         case BPF_FUNC_sk_fullsock:
7336                 return &bpf_sk_fullsock_proto;
7337         case BPF_FUNC_sk_storage_get:
7338                 return &bpf_sk_storage_get_proto;
7339         case BPF_FUNC_sk_storage_delete:
7340                 return &bpf_sk_storage_delete_proto;
7341 #ifdef CONFIG_XFRM
7342         case BPF_FUNC_skb_get_xfrm_state:
7343                 return &bpf_skb_get_xfrm_state_proto;
7344 #endif
7345 #ifdef CONFIG_CGROUP_NET_CLASSID
7346         case BPF_FUNC_skb_cgroup_classid:
7347                 return &bpf_skb_cgroup_classid_proto;
7348 #endif
7349 #ifdef CONFIG_SOCK_CGROUP_DATA
7350         case BPF_FUNC_skb_cgroup_id:
7351                 return &bpf_skb_cgroup_id_proto;
7352         case BPF_FUNC_skb_ancestor_cgroup_id:
7353                 return &bpf_skb_ancestor_cgroup_id_proto;
7354 #endif
7355 #ifdef CONFIG_INET
7356         case BPF_FUNC_sk_lookup_tcp:
7357                 return &bpf_sk_lookup_tcp_proto;
7358         case BPF_FUNC_sk_lookup_udp:
7359                 return &bpf_sk_lookup_udp_proto;
7360         case BPF_FUNC_sk_release:
7361                 return &bpf_sk_release_proto;
7362         case BPF_FUNC_tcp_sock:
7363                 return &bpf_tcp_sock_proto;
7364         case BPF_FUNC_get_listener_sock:
7365                 return &bpf_get_listener_sock_proto;
7366         case BPF_FUNC_skc_lookup_tcp:
7367                 return &bpf_skc_lookup_tcp_proto;
7368         case BPF_FUNC_tcp_check_syncookie:
7369                 return &bpf_tcp_check_syncookie_proto;
7370         case BPF_FUNC_skb_ecn_set_ce:
7371                 return &bpf_skb_ecn_set_ce_proto;
7372         case BPF_FUNC_tcp_gen_syncookie:
7373                 return &bpf_tcp_gen_syncookie_proto;
7374         case BPF_FUNC_sk_assign:
7375                 return &bpf_sk_assign_proto;
7376 #endif
7377         default:
7378                 return bpf_sk_base_func_proto(func_id);
7379         }
7380 }
7381
7382 static const struct bpf_func_proto *
7383 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7384 {
7385         switch (func_id) {
7386         case BPF_FUNC_perf_event_output:
7387                 return &bpf_xdp_event_output_proto;
7388         case BPF_FUNC_get_smp_processor_id:
7389                 return &bpf_get_smp_processor_id_proto;
7390         case BPF_FUNC_csum_diff:
7391                 return &bpf_csum_diff_proto;
7392         case BPF_FUNC_xdp_adjust_head:
7393                 return &bpf_xdp_adjust_head_proto;
7394         case BPF_FUNC_xdp_adjust_meta:
7395                 return &bpf_xdp_adjust_meta_proto;
7396         case BPF_FUNC_redirect:
7397                 return &bpf_xdp_redirect_proto;
7398         case BPF_FUNC_redirect_map:
7399                 return &bpf_xdp_redirect_map_proto;
7400         case BPF_FUNC_xdp_adjust_tail:
7401                 return &bpf_xdp_adjust_tail_proto;
7402         case BPF_FUNC_fib_lookup:
7403                 return &bpf_xdp_fib_lookup_proto;
7404         case BPF_FUNC_check_mtu:
7405                 return &bpf_xdp_check_mtu_proto;
7406 #ifdef CONFIG_INET
7407         case BPF_FUNC_sk_lookup_udp:
7408                 return &bpf_xdp_sk_lookup_udp_proto;
7409         case BPF_FUNC_sk_lookup_tcp:
7410                 return &bpf_xdp_sk_lookup_tcp_proto;
7411         case BPF_FUNC_sk_release:
7412                 return &bpf_sk_release_proto;
7413         case BPF_FUNC_skc_lookup_tcp:
7414                 return &bpf_xdp_skc_lookup_tcp_proto;
7415         case BPF_FUNC_tcp_check_syncookie:
7416                 return &bpf_tcp_check_syncookie_proto;
7417         case BPF_FUNC_tcp_gen_syncookie:
7418                 return &bpf_tcp_gen_syncookie_proto;
7419 #endif
7420         default:
7421                 return bpf_sk_base_func_proto(func_id);
7422         }
7423 }
7424
7425 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7426 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7427
7428 static const struct bpf_func_proto *
7429 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7430 {
7431         switch (func_id) {
7432         case BPF_FUNC_setsockopt:
7433                 return &bpf_sock_ops_setsockopt_proto;
7434         case BPF_FUNC_getsockopt:
7435                 return &bpf_sock_ops_getsockopt_proto;
7436         case BPF_FUNC_sock_ops_cb_flags_set:
7437                 return &bpf_sock_ops_cb_flags_set_proto;
7438         case BPF_FUNC_sock_map_update:
7439                 return &bpf_sock_map_update_proto;
7440         case BPF_FUNC_sock_hash_update:
7441                 return &bpf_sock_hash_update_proto;
7442         case BPF_FUNC_get_socket_cookie:
7443                 return &bpf_get_socket_cookie_sock_ops_proto;
7444         case BPF_FUNC_get_local_storage:
7445                 return &bpf_get_local_storage_proto;
7446         case BPF_FUNC_perf_event_output:
7447                 return &bpf_event_output_data_proto;
7448         case BPF_FUNC_sk_storage_get:
7449                 return &bpf_sk_storage_get_proto;
7450         case BPF_FUNC_sk_storage_delete:
7451                 return &bpf_sk_storage_delete_proto;
7452 #ifdef CONFIG_INET
7453         case BPF_FUNC_load_hdr_opt:
7454                 return &bpf_sock_ops_load_hdr_opt_proto;
7455         case BPF_FUNC_store_hdr_opt:
7456                 return &bpf_sock_ops_store_hdr_opt_proto;
7457         case BPF_FUNC_reserve_hdr_opt:
7458                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7459         case BPF_FUNC_tcp_sock:
7460                 return &bpf_tcp_sock_proto;
7461 #endif /* CONFIG_INET */
7462         default:
7463                 return bpf_sk_base_func_proto(func_id);
7464         }
7465 }
7466
7467 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7468 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7469
7470 static const struct bpf_func_proto *
7471 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7472 {
7473         switch (func_id) {
7474         case BPF_FUNC_msg_redirect_map:
7475                 return &bpf_msg_redirect_map_proto;
7476         case BPF_FUNC_msg_redirect_hash:
7477                 return &bpf_msg_redirect_hash_proto;
7478         case BPF_FUNC_msg_apply_bytes:
7479                 return &bpf_msg_apply_bytes_proto;
7480         case BPF_FUNC_msg_cork_bytes:
7481                 return &bpf_msg_cork_bytes_proto;
7482         case BPF_FUNC_msg_pull_data:
7483                 return &bpf_msg_pull_data_proto;
7484         case BPF_FUNC_msg_push_data:
7485                 return &bpf_msg_push_data_proto;
7486         case BPF_FUNC_msg_pop_data:
7487                 return &bpf_msg_pop_data_proto;
7488         case BPF_FUNC_perf_event_output:
7489                 return &bpf_event_output_data_proto;
7490         case BPF_FUNC_get_current_uid_gid:
7491                 return &bpf_get_current_uid_gid_proto;
7492         case BPF_FUNC_get_current_pid_tgid:
7493                 return &bpf_get_current_pid_tgid_proto;
7494         case BPF_FUNC_sk_storage_get:
7495                 return &bpf_sk_storage_get_proto;
7496         case BPF_FUNC_sk_storage_delete:
7497                 return &bpf_sk_storage_delete_proto;
7498 #ifdef CONFIG_CGROUPS
7499         case BPF_FUNC_get_current_cgroup_id:
7500                 return &bpf_get_current_cgroup_id_proto;
7501         case BPF_FUNC_get_current_ancestor_cgroup_id:
7502                 return &bpf_get_current_ancestor_cgroup_id_proto;
7503 #endif
7504 #ifdef CONFIG_CGROUP_NET_CLASSID
7505         case BPF_FUNC_get_cgroup_classid:
7506                 return &bpf_get_cgroup_classid_curr_proto;
7507 #endif
7508         default:
7509                 return bpf_sk_base_func_proto(func_id);
7510         }
7511 }
7512
7513 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7514 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7515
7516 static const struct bpf_func_proto *
7517 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7518 {
7519         switch (func_id) {
7520         case BPF_FUNC_skb_store_bytes:
7521                 return &bpf_skb_store_bytes_proto;
7522         case BPF_FUNC_skb_load_bytes:
7523                 return &bpf_skb_load_bytes_proto;
7524         case BPF_FUNC_skb_pull_data:
7525                 return &sk_skb_pull_data_proto;
7526         case BPF_FUNC_skb_change_tail:
7527                 return &sk_skb_change_tail_proto;
7528         case BPF_FUNC_skb_change_head:
7529                 return &sk_skb_change_head_proto;
7530         case BPF_FUNC_skb_adjust_room:
7531                 return &sk_skb_adjust_room_proto;
7532         case BPF_FUNC_get_socket_cookie:
7533                 return &bpf_get_socket_cookie_proto;
7534         case BPF_FUNC_get_socket_uid:
7535                 return &bpf_get_socket_uid_proto;
7536         case BPF_FUNC_sk_redirect_map:
7537                 return &bpf_sk_redirect_map_proto;
7538         case BPF_FUNC_sk_redirect_hash:
7539                 return &bpf_sk_redirect_hash_proto;
7540         case BPF_FUNC_perf_event_output:
7541                 return &bpf_skb_event_output_proto;
7542 #ifdef CONFIG_INET
7543         case BPF_FUNC_sk_lookup_tcp:
7544                 return &bpf_sk_lookup_tcp_proto;
7545         case BPF_FUNC_sk_lookup_udp:
7546                 return &bpf_sk_lookup_udp_proto;
7547         case BPF_FUNC_sk_release:
7548                 return &bpf_sk_release_proto;
7549         case BPF_FUNC_skc_lookup_tcp:
7550                 return &bpf_skc_lookup_tcp_proto;
7551 #endif
7552         default:
7553                 return bpf_sk_base_func_proto(func_id);
7554         }
7555 }
7556
7557 static const struct bpf_func_proto *
7558 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7559 {
7560         switch (func_id) {
7561         case BPF_FUNC_skb_load_bytes:
7562                 return &bpf_flow_dissector_load_bytes_proto;
7563         default:
7564                 return bpf_sk_base_func_proto(func_id);
7565         }
7566 }
7567
7568 static const struct bpf_func_proto *
7569 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7570 {
7571         switch (func_id) {
7572         case BPF_FUNC_skb_load_bytes:
7573                 return &bpf_skb_load_bytes_proto;
7574         case BPF_FUNC_skb_pull_data:
7575                 return &bpf_skb_pull_data_proto;
7576         case BPF_FUNC_csum_diff:
7577                 return &bpf_csum_diff_proto;
7578         case BPF_FUNC_get_cgroup_classid:
7579                 return &bpf_get_cgroup_classid_proto;
7580         case BPF_FUNC_get_route_realm:
7581                 return &bpf_get_route_realm_proto;
7582         case BPF_FUNC_get_hash_recalc:
7583                 return &bpf_get_hash_recalc_proto;
7584         case BPF_FUNC_perf_event_output:
7585                 return &bpf_skb_event_output_proto;
7586         case BPF_FUNC_get_smp_processor_id:
7587                 return &bpf_get_smp_processor_id_proto;
7588         case BPF_FUNC_skb_under_cgroup:
7589                 return &bpf_skb_under_cgroup_proto;
7590         default:
7591                 return bpf_sk_base_func_proto(func_id);
7592         }
7593 }
7594
7595 static const struct bpf_func_proto *
7596 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7597 {
7598         switch (func_id) {
7599         case BPF_FUNC_lwt_push_encap:
7600                 return &bpf_lwt_in_push_encap_proto;
7601         default:
7602                 return lwt_out_func_proto(func_id, prog);
7603         }
7604 }
7605
7606 static const struct bpf_func_proto *
7607 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7608 {
7609         switch (func_id) {
7610         case BPF_FUNC_skb_get_tunnel_key:
7611                 return &bpf_skb_get_tunnel_key_proto;
7612         case BPF_FUNC_skb_set_tunnel_key:
7613                 return bpf_get_skb_set_tunnel_proto(func_id);
7614         case BPF_FUNC_skb_get_tunnel_opt:
7615                 return &bpf_skb_get_tunnel_opt_proto;
7616         case BPF_FUNC_skb_set_tunnel_opt:
7617                 return bpf_get_skb_set_tunnel_proto(func_id);
7618         case BPF_FUNC_redirect:
7619                 return &bpf_redirect_proto;
7620         case BPF_FUNC_clone_redirect:
7621                 return &bpf_clone_redirect_proto;
7622         case BPF_FUNC_skb_change_tail:
7623                 return &bpf_skb_change_tail_proto;
7624         case BPF_FUNC_skb_change_head:
7625                 return &bpf_skb_change_head_proto;
7626         case BPF_FUNC_skb_store_bytes:
7627                 return &bpf_skb_store_bytes_proto;
7628         case BPF_FUNC_csum_update:
7629                 return &bpf_csum_update_proto;
7630         case BPF_FUNC_csum_level:
7631                 return &bpf_csum_level_proto;
7632         case BPF_FUNC_l3_csum_replace:
7633                 return &bpf_l3_csum_replace_proto;
7634         case BPF_FUNC_l4_csum_replace:
7635                 return &bpf_l4_csum_replace_proto;
7636         case BPF_FUNC_set_hash_invalid:
7637                 return &bpf_set_hash_invalid_proto;
7638         case BPF_FUNC_lwt_push_encap:
7639                 return &bpf_lwt_xmit_push_encap_proto;
7640         default:
7641                 return lwt_out_func_proto(func_id, prog);
7642         }
7643 }
7644
7645 static const struct bpf_func_proto *
7646 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7647 {
7648         switch (func_id) {
7649 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7650         case BPF_FUNC_lwt_seg6_store_bytes:
7651                 return &bpf_lwt_seg6_store_bytes_proto;
7652         case BPF_FUNC_lwt_seg6_action:
7653                 return &bpf_lwt_seg6_action_proto;
7654         case BPF_FUNC_lwt_seg6_adjust_srh:
7655                 return &bpf_lwt_seg6_adjust_srh_proto;
7656 #endif
7657         default:
7658                 return lwt_out_func_proto(func_id, prog);
7659         }
7660 }
7661
7662 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7663                                     const struct bpf_prog *prog,
7664                                     struct bpf_insn_access_aux *info)
7665 {
7666         const int size_default = sizeof(__u32);
7667
7668         if (off < 0 || off >= sizeof(struct __sk_buff))
7669                 return false;
7670
7671         /* The verifier guarantees that size > 0. */
7672         if (off % size != 0)
7673                 return false;
7674
7675         switch (off) {
7676         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7677                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7678                         return false;
7679                 break;
7680         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7681         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7682         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7683         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7684         case bpf_ctx_range(struct __sk_buff, data):
7685         case bpf_ctx_range(struct __sk_buff, data_meta):
7686         case bpf_ctx_range(struct __sk_buff, data_end):
7687                 if (size != size_default)
7688                         return false;
7689                 break;
7690         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7691                 return false;
7692         case bpf_ctx_range(struct __sk_buff, tstamp):
7693                 if (size != sizeof(__u64))
7694                         return false;
7695                 break;
7696         case offsetof(struct __sk_buff, sk):
7697                 if (type == BPF_WRITE || size != sizeof(__u64))
7698                         return false;
7699                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7700                 break;
7701         default:
7702                 /* Only narrow read access allowed for now. */
7703                 if (type == BPF_WRITE) {
7704                         if (size != size_default)
7705                                 return false;
7706                 } else {
7707                         bpf_ctx_record_field_size(info, size_default);
7708                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7709                                 return false;
7710                 }
7711         }
7712
7713         return true;
7714 }
7715
7716 static bool sk_filter_is_valid_access(int off, int size,
7717                                       enum bpf_access_type type,
7718                                       const struct bpf_prog *prog,
7719                                       struct bpf_insn_access_aux *info)
7720 {
7721         switch (off) {
7722         case bpf_ctx_range(struct __sk_buff, tc_classid):
7723         case bpf_ctx_range(struct __sk_buff, data):
7724         case bpf_ctx_range(struct __sk_buff, data_meta):
7725         case bpf_ctx_range(struct __sk_buff, data_end):
7726         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7727         case bpf_ctx_range(struct __sk_buff, tstamp):
7728         case bpf_ctx_range(struct __sk_buff, wire_len):
7729                 return false;
7730         }
7731
7732         if (type == BPF_WRITE) {
7733                 switch (off) {
7734                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7735                         break;
7736                 default:
7737                         return false;
7738                 }
7739         }
7740
7741         return bpf_skb_is_valid_access(off, size, type, prog, info);
7742 }
7743
7744 static bool cg_skb_is_valid_access(int off, int size,
7745                                    enum bpf_access_type type,
7746                                    const struct bpf_prog *prog,
7747                                    struct bpf_insn_access_aux *info)
7748 {
7749         switch (off) {
7750         case bpf_ctx_range(struct __sk_buff, tc_classid):
7751         case bpf_ctx_range(struct __sk_buff, data_meta):
7752         case bpf_ctx_range(struct __sk_buff, wire_len):
7753                 return false;
7754         case bpf_ctx_range(struct __sk_buff, data):
7755         case bpf_ctx_range(struct __sk_buff, data_end):
7756                 if (!bpf_capable())
7757                         return false;
7758                 break;
7759         }
7760
7761         if (type == BPF_WRITE) {
7762                 switch (off) {
7763                 case bpf_ctx_range(struct __sk_buff, mark):
7764                 case bpf_ctx_range(struct __sk_buff, priority):
7765                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7766                         break;
7767                 case bpf_ctx_range(struct __sk_buff, tstamp):
7768                         if (!bpf_capable())
7769                                 return false;
7770                         break;
7771                 default:
7772                         return false;
7773                 }
7774         }
7775
7776         switch (off) {
7777         case bpf_ctx_range(struct __sk_buff, data):
7778                 info->reg_type = PTR_TO_PACKET;
7779                 break;
7780         case bpf_ctx_range(struct __sk_buff, data_end):
7781                 info->reg_type = PTR_TO_PACKET_END;
7782                 break;
7783         }
7784
7785         return bpf_skb_is_valid_access(off, size, type, prog, info);
7786 }
7787
7788 static bool lwt_is_valid_access(int off, int size,
7789                                 enum bpf_access_type type,
7790                                 const struct bpf_prog *prog,
7791                                 struct bpf_insn_access_aux *info)
7792 {
7793         switch (off) {
7794         case bpf_ctx_range(struct __sk_buff, tc_classid):
7795         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7796         case bpf_ctx_range(struct __sk_buff, data_meta):
7797         case bpf_ctx_range(struct __sk_buff, tstamp):
7798         case bpf_ctx_range(struct __sk_buff, wire_len):
7799                 return false;
7800         }
7801
7802         if (type == BPF_WRITE) {
7803                 switch (off) {
7804                 case bpf_ctx_range(struct __sk_buff, mark):
7805                 case bpf_ctx_range(struct __sk_buff, priority):
7806                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7807                         break;
7808                 default:
7809                         return false;
7810                 }
7811         }
7812
7813         switch (off) {
7814         case bpf_ctx_range(struct __sk_buff, data):
7815                 info->reg_type = PTR_TO_PACKET;
7816                 break;
7817         case bpf_ctx_range(struct __sk_buff, data_end):
7818                 info->reg_type = PTR_TO_PACKET_END;
7819                 break;
7820         }
7821
7822         return bpf_skb_is_valid_access(off, size, type, prog, info);
7823 }
7824
7825 /* Attach type specific accesses */
7826 static bool __sock_filter_check_attach_type(int off,
7827                                             enum bpf_access_type access_type,
7828                                             enum bpf_attach_type attach_type)
7829 {
7830         switch (off) {
7831         case offsetof(struct bpf_sock, bound_dev_if):
7832         case offsetof(struct bpf_sock, mark):
7833         case offsetof(struct bpf_sock, priority):
7834                 switch (attach_type) {
7835                 case BPF_CGROUP_INET_SOCK_CREATE:
7836                 case BPF_CGROUP_INET_SOCK_RELEASE:
7837                         goto full_access;
7838                 default:
7839                         return false;
7840                 }
7841         case bpf_ctx_range(struct bpf_sock, src_ip4):
7842                 switch (attach_type) {
7843                 case BPF_CGROUP_INET4_POST_BIND:
7844                         goto read_only;
7845                 default:
7846                         return false;
7847                 }
7848         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7849                 switch (attach_type) {
7850                 case BPF_CGROUP_INET6_POST_BIND:
7851                         goto read_only;
7852                 default:
7853                         return false;
7854                 }
7855         case bpf_ctx_range(struct bpf_sock, src_port):
7856                 switch (attach_type) {
7857                 case BPF_CGROUP_INET4_POST_BIND:
7858                 case BPF_CGROUP_INET6_POST_BIND:
7859                         goto read_only;
7860                 default:
7861                         return false;
7862                 }
7863         }
7864 read_only:
7865         return access_type == BPF_READ;
7866 full_access:
7867         return true;
7868 }
7869
7870 bool bpf_sock_common_is_valid_access(int off, int size,
7871                                      enum bpf_access_type type,
7872                                      struct bpf_insn_access_aux *info)
7873 {
7874         switch (off) {
7875         case bpf_ctx_range_till(struct bpf_sock, type, priority):
7876                 return false;
7877         default:
7878                 return bpf_sock_is_valid_access(off, size, type, info);
7879         }
7880 }
7881
7882 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7883                               struct bpf_insn_access_aux *info)
7884 {
7885         const int size_default = sizeof(__u32);
7886
7887         if (off < 0 || off >= sizeof(struct bpf_sock))
7888                 return false;
7889         if (off % size != 0)
7890                 return false;
7891
7892         switch (off) {
7893         case offsetof(struct bpf_sock, state):
7894         case offsetof(struct bpf_sock, family):
7895         case offsetof(struct bpf_sock, type):
7896         case offsetof(struct bpf_sock, protocol):
7897         case offsetof(struct bpf_sock, dst_port):
7898         case offsetof(struct bpf_sock, src_port):
7899         case offsetof(struct bpf_sock, rx_queue_mapping):
7900         case bpf_ctx_range(struct bpf_sock, src_ip4):
7901         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7902         case bpf_ctx_range(struct bpf_sock, dst_ip4):
7903         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7904                 bpf_ctx_record_field_size(info, size_default);
7905                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7906         }
7907
7908         return size == size_default;
7909 }
7910
7911 static bool sock_filter_is_valid_access(int off, int size,
7912                                         enum bpf_access_type type,
7913                                         const struct bpf_prog *prog,
7914                                         struct bpf_insn_access_aux *info)
7915 {
7916         if (!bpf_sock_is_valid_access(off, size, type, info))
7917                 return false;
7918         return __sock_filter_check_attach_type(off, type,
7919                                                prog->expected_attach_type);
7920 }
7921
7922 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7923                              const struct bpf_prog *prog)
7924 {
7925         /* Neither direct read nor direct write requires any preliminary
7926          * action.
7927          */
7928         return 0;
7929 }
7930
7931 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7932                                 const struct bpf_prog *prog, int drop_verdict)
7933 {
7934         struct bpf_insn *insn = insn_buf;
7935
7936         if (!direct_write)
7937                 return 0;
7938
7939         /* if (!skb->cloned)
7940          *       goto start;
7941          *
7942          * (Fast-path, otherwise approximation that we might be
7943          *  a clone, do the rest in helper.)
7944          */
7945         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7946         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7947         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7948
7949         /* ret = bpf_skb_pull_data(skb, 0); */
7950         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7951         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7952         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7953                                BPF_FUNC_skb_pull_data);
7954         /* if (!ret)
7955          *      goto restore;
7956          * return TC_ACT_SHOT;
7957          */
7958         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7959         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7960         *insn++ = BPF_EXIT_INSN();
7961
7962         /* restore: */
7963         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7964         /* start: */
7965         *insn++ = prog->insnsi[0];
7966
7967         return insn - insn_buf;
7968 }
7969
7970 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7971                           struct bpf_insn *insn_buf)
7972 {
7973         bool indirect = BPF_MODE(orig->code) == BPF_IND;
7974         struct bpf_insn *insn = insn_buf;
7975
7976         if (!indirect) {
7977                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7978         } else {
7979                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7980                 if (orig->imm)
7981                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7982         }
7983         /* We're guaranteed here that CTX is in R6. */
7984         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7985
7986         switch (BPF_SIZE(orig->code)) {
7987         case BPF_B:
7988                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7989                 break;
7990         case BPF_H:
7991                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7992                 break;
7993         case BPF_W:
7994                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7995                 break;
7996         }
7997
7998         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7999         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8000         *insn++ = BPF_EXIT_INSN();
8001
8002         return insn - insn_buf;
8003 }
8004
8005 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8006                                const struct bpf_prog *prog)
8007 {
8008         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8009 }
8010
8011 static bool tc_cls_act_is_valid_access(int off, int size,
8012                                        enum bpf_access_type type,
8013                                        const struct bpf_prog *prog,
8014                                        struct bpf_insn_access_aux *info)
8015 {
8016         if (type == BPF_WRITE) {
8017                 switch (off) {
8018                 case bpf_ctx_range(struct __sk_buff, mark):
8019                 case bpf_ctx_range(struct __sk_buff, tc_index):
8020                 case bpf_ctx_range(struct __sk_buff, priority):
8021                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8022                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8023                 case bpf_ctx_range(struct __sk_buff, tstamp):
8024                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8025                         break;
8026                 default:
8027                         return false;
8028                 }
8029         }
8030
8031         switch (off) {
8032         case bpf_ctx_range(struct __sk_buff, data):
8033                 info->reg_type = PTR_TO_PACKET;
8034                 break;
8035         case bpf_ctx_range(struct __sk_buff, data_meta):
8036                 info->reg_type = PTR_TO_PACKET_META;
8037                 break;
8038         case bpf_ctx_range(struct __sk_buff, data_end):
8039                 info->reg_type = PTR_TO_PACKET_END;
8040                 break;
8041         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8042                 return false;
8043         }
8044
8045         return bpf_skb_is_valid_access(off, size, type, prog, info);
8046 }
8047
8048 static bool __is_valid_xdp_access(int off, int size)
8049 {
8050         if (off < 0 || off >= sizeof(struct xdp_md))
8051                 return false;
8052         if (off % size != 0)
8053                 return false;
8054         if (size != sizeof(__u32))
8055                 return false;
8056
8057         return true;
8058 }
8059
8060 static bool xdp_is_valid_access(int off, int size,
8061                                 enum bpf_access_type type,
8062                                 const struct bpf_prog *prog,
8063                                 struct bpf_insn_access_aux *info)
8064 {
8065         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8066                 switch (off) {
8067                 case offsetof(struct xdp_md, egress_ifindex):
8068                         return false;
8069                 }
8070         }
8071
8072         if (type == BPF_WRITE) {
8073                 if (bpf_prog_is_dev_bound(prog->aux)) {
8074                         switch (off) {
8075                         case offsetof(struct xdp_md, rx_queue_index):
8076                                 return __is_valid_xdp_access(off, size);
8077                         }
8078                 }
8079                 return false;
8080         }
8081
8082         switch (off) {
8083         case offsetof(struct xdp_md, data):
8084                 info->reg_type = PTR_TO_PACKET;
8085                 break;
8086         case offsetof(struct xdp_md, data_meta):
8087                 info->reg_type = PTR_TO_PACKET_META;
8088                 break;
8089         case offsetof(struct xdp_md, data_end):
8090                 info->reg_type = PTR_TO_PACKET_END;
8091                 break;
8092         }
8093
8094         return __is_valid_xdp_access(off, size);
8095 }
8096
8097 void bpf_warn_invalid_xdp_action(u32 act)
8098 {
8099         const u32 act_max = XDP_REDIRECT;
8100
8101         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8102                   act > act_max ? "Illegal" : "Driver unsupported",
8103                   act);
8104 }
8105 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8106
8107 static bool sock_addr_is_valid_access(int off, int size,
8108                                       enum bpf_access_type type,
8109                                       const struct bpf_prog *prog,
8110                                       struct bpf_insn_access_aux *info)
8111 {
8112         const int size_default = sizeof(__u32);
8113
8114         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8115                 return false;
8116         if (off % size != 0)
8117                 return false;
8118
8119         /* Disallow access to IPv6 fields from IPv4 contex and vise
8120          * versa.
8121          */
8122         switch (off) {
8123         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8124                 switch (prog->expected_attach_type) {
8125                 case BPF_CGROUP_INET4_BIND:
8126                 case BPF_CGROUP_INET4_CONNECT:
8127                 case BPF_CGROUP_INET4_GETPEERNAME:
8128                 case BPF_CGROUP_INET4_GETSOCKNAME:
8129                 case BPF_CGROUP_UDP4_SENDMSG:
8130                 case BPF_CGROUP_UDP4_RECVMSG:
8131                         break;
8132                 default:
8133                         return false;
8134                 }
8135                 break;
8136         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8137                 switch (prog->expected_attach_type) {
8138                 case BPF_CGROUP_INET6_BIND:
8139                 case BPF_CGROUP_INET6_CONNECT:
8140                 case BPF_CGROUP_INET6_GETPEERNAME:
8141                 case BPF_CGROUP_INET6_GETSOCKNAME:
8142                 case BPF_CGROUP_UDP6_SENDMSG:
8143                 case BPF_CGROUP_UDP6_RECVMSG:
8144                         break;
8145                 default:
8146                         return false;
8147                 }
8148                 break;
8149         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8150                 switch (prog->expected_attach_type) {
8151                 case BPF_CGROUP_UDP4_SENDMSG:
8152                         break;
8153                 default:
8154                         return false;
8155                 }
8156                 break;
8157         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8158                                 msg_src_ip6[3]):
8159                 switch (prog->expected_attach_type) {
8160                 case BPF_CGROUP_UDP6_SENDMSG:
8161                         break;
8162                 default:
8163                         return false;
8164                 }
8165                 break;
8166         }
8167
8168         switch (off) {
8169         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8170         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8171         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8172         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8173                                 msg_src_ip6[3]):
8174         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8175                 if (type == BPF_READ) {
8176                         bpf_ctx_record_field_size(info, size_default);
8177
8178                         if (bpf_ctx_wide_access_ok(off, size,
8179                                                    struct bpf_sock_addr,
8180                                                    user_ip6))
8181                                 return true;
8182
8183                         if (bpf_ctx_wide_access_ok(off, size,
8184                                                    struct bpf_sock_addr,
8185                                                    msg_src_ip6))
8186                                 return true;
8187
8188                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8189                                 return false;
8190                 } else {
8191                         if (bpf_ctx_wide_access_ok(off, size,
8192                                                    struct bpf_sock_addr,
8193                                                    user_ip6))
8194                                 return true;
8195
8196                         if (bpf_ctx_wide_access_ok(off, size,
8197                                                    struct bpf_sock_addr,
8198                                                    msg_src_ip6))
8199                                 return true;
8200
8201                         if (size != size_default)
8202                                 return false;
8203                 }
8204                 break;
8205         case offsetof(struct bpf_sock_addr, sk):
8206                 if (type != BPF_READ)
8207                         return false;
8208                 if (size != sizeof(__u64))
8209                         return false;
8210                 info->reg_type = PTR_TO_SOCKET;
8211                 break;
8212         default:
8213                 if (type == BPF_READ) {
8214                         if (size != size_default)
8215                                 return false;
8216                 } else {
8217                         return false;
8218                 }
8219         }
8220
8221         return true;
8222 }
8223
8224 static bool sock_ops_is_valid_access(int off, int size,
8225                                      enum bpf_access_type type,
8226                                      const struct bpf_prog *prog,
8227                                      struct bpf_insn_access_aux *info)
8228 {
8229         const int size_default = sizeof(__u32);
8230
8231         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8232                 return false;
8233
8234         /* The verifier guarantees that size > 0. */
8235         if (off % size != 0)
8236                 return false;
8237
8238         if (type == BPF_WRITE) {
8239                 switch (off) {
8240                 case offsetof(struct bpf_sock_ops, reply):
8241                 case offsetof(struct bpf_sock_ops, sk_txhash):
8242                         if (size != size_default)
8243                                 return false;
8244                         break;
8245                 default:
8246                         return false;
8247                 }
8248         } else {
8249                 switch (off) {
8250                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8251                                         bytes_acked):
8252                         if (size != sizeof(__u64))
8253                                 return false;
8254                         break;
8255                 case offsetof(struct bpf_sock_ops, sk):
8256                         if (size != sizeof(__u64))
8257                                 return false;
8258                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8259                         break;
8260                 case offsetof(struct bpf_sock_ops, skb_data):
8261                         if (size != sizeof(__u64))
8262                                 return false;
8263                         info->reg_type = PTR_TO_PACKET;
8264                         break;
8265                 case offsetof(struct bpf_sock_ops, skb_data_end):
8266                         if (size != sizeof(__u64))
8267                                 return false;
8268                         info->reg_type = PTR_TO_PACKET_END;
8269                         break;
8270                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8271                         bpf_ctx_record_field_size(info, size_default);
8272                         return bpf_ctx_narrow_access_ok(off, size,
8273                                                         size_default);
8274                 default:
8275                         if (size != size_default)
8276                                 return false;
8277                         break;
8278                 }
8279         }
8280
8281         return true;
8282 }
8283
8284 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8285                            const struct bpf_prog *prog)
8286 {
8287         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8288 }
8289
8290 static bool sk_skb_is_valid_access(int off, int size,
8291                                    enum bpf_access_type type,
8292                                    const struct bpf_prog *prog,
8293                                    struct bpf_insn_access_aux *info)
8294 {
8295         switch (off) {
8296         case bpf_ctx_range(struct __sk_buff, tc_classid):
8297         case bpf_ctx_range(struct __sk_buff, data_meta):
8298         case bpf_ctx_range(struct __sk_buff, tstamp):
8299         case bpf_ctx_range(struct __sk_buff, wire_len):
8300                 return false;
8301         }
8302
8303         if (type == BPF_WRITE) {
8304                 switch (off) {
8305                 case bpf_ctx_range(struct __sk_buff, tc_index):
8306                 case bpf_ctx_range(struct __sk_buff, priority):
8307                         break;
8308                 default:
8309                         return false;
8310                 }
8311         }
8312
8313         switch (off) {
8314         case bpf_ctx_range(struct __sk_buff, mark):
8315                 return false;
8316         case bpf_ctx_range(struct __sk_buff, data):
8317                 info->reg_type = PTR_TO_PACKET;
8318                 break;
8319         case bpf_ctx_range(struct __sk_buff, data_end):
8320                 info->reg_type = PTR_TO_PACKET_END;
8321                 break;
8322         }
8323
8324         return bpf_skb_is_valid_access(off, size, type, prog, info);
8325 }
8326
8327 static bool sk_msg_is_valid_access(int off, int size,
8328                                    enum bpf_access_type type,
8329                                    const struct bpf_prog *prog,
8330                                    struct bpf_insn_access_aux *info)
8331 {
8332         if (type == BPF_WRITE)
8333                 return false;
8334
8335         if (off % size != 0)
8336                 return false;
8337
8338         switch (off) {
8339         case offsetof(struct sk_msg_md, data):
8340                 info->reg_type = PTR_TO_PACKET;
8341                 if (size != sizeof(__u64))
8342                         return false;
8343                 break;
8344         case offsetof(struct sk_msg_md, data_end):
8345                 info->reg_type = PTR_TO_PACKET_END;
8346                 if (size != sizeof(__u64))
8347                         return false;
8348                 break;
8349         case offsetof(struct sk_msg_md, sk):
8350                 if (size != sizeof(__u64))
8351                         return false;
8352                 info->reg_type = PTR_TO_SOCKET;
8353                 break;
8354         case bpf_ctx_range(struct sk_msg_md, family):
8355         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8356         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8357         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8358         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8359         case bpf_ctx_range(struct sk_msg_md, remote_port):
8360         case bpf_ctx_range(struct sk_msg_md, local_port):
8361         case bpf_ctx_range(struct sk_msg_md, size):
8362                 if (size != sizeof(__u32))
8363                         return false;
8364                 break;
8365         default:
8366                 return false;
8367         }
8368         return true;
8369 }
8370
8371 static bool flow_dissector_is_valid_access(int off, int size,
8372                                            enum bpf_access_type type,
8373                                            const struct bpf_prog *prog,
8374                                            struct bpf_insn_access_aux *info)
8375 {
8376         const int size_default = sizeof(__u32);
8377
8378         if (off < 0 || off >= sizeof(struct __sk_buff))
8379                 return false;
8380
8381         if (type == BPF_WRITE)
8382                 return false;
8383
8384         switch (off) {
8385         case bpf_ctx_range(struct __sk_buff, data):
8386                 if (size != size_default)
8387                         return false;
8388                 info->reg_type = PTR_TO_PACKET;
8389                 return true;
8390         case bpf_ctx_range(struct __sk_buff, data_end):
8391                 if (size != size_default)
8392                         return false;
8393                 info->reg_type = PTR_TO_PACKET_END;
8394                 return true;
8395         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8396                 if (size != sizeof(__u64))
8397                         return false;
8398                 info->reg_type = PTR_TO_FLOW_KEYS;
8399                 return true;
8400         default:
8401                 return false;
8402         }
8403 }
8404
8405 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8406                                              const struct bpf_insn *si,
8407                                              struct bpf_insn *insn_buf,
8408                                              struct bpf_prog *prog,
8409                                              u32 *target_size)
8410
8411 {
8412         struct bpf_insn *insn = insn_buf;
8413
8414         switch (si->off) {
8415         case offsetof(struct __sk_buff, data):
8416                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8417                                       si->dst_reg, si->src_reg,
8418                                       offsetof(struct bpf_flow_dissector, data));
8419                 break;
8420
8421         case offsetof(struct __sk_buff, data_end):
8422                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8423                                       si->dst_reg, si->src_reg,
8424                                       offsetof(struct bpf_flow_dissector, data_end));
8425                 break;
8426
8427         case offsetof(struct __sk_buff, flow_keys):
8428                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8429                                       si->dst_reg, si->src_reg,
8430                                       offsetof(struct bpf_flow_dissector, flow_keys));
8431                 break;
8432         }
8433
8434         return insn - insn_buf;
8435 }
8436
8437 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8438                                                   struct bpf_insn *insn)
8439 {
8440         /* si->dst_reg = skb_shinfo(SKB); */
8441 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8442         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8443                               BPF_REG_AX, si->src_reg,
8444                               offsetof(struct sk_buff, end));
8445         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8446                               si->dst_reg, si->src_reg,
8447                               offsetof(struct sk_buff, head));
8448         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8449 #else
8450         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8451                               si->dst_reg, si->src_reg,
8452                               offsetof(struct sk_buff, end));
8453 #endif
8454
8455         return insn;
8456 }
8457
8458 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8459                                   const struct bpf_insn *si,
8460                                   struct bpf_insn *insn_buf,
8461                                   struct bpf_prog *prog, u32 *target_size)
8462 {
8463         struct bpf_insn *insn = insn_buf;
8464         int off;
8465
8466         switch (si->off) {
8467         case offsetof(struct __sk_buff, len):
8468                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8469                                       bpf_target_off(struct sk_buff, len, 4,
8470                                                      target_size));
8471                 break;
8472
8473         case offsetof(struct __sk_buff, protocol):
8474                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8475                                       bpf_target_off(struct sk_buff, protocol, 2,
8476                                                      target_size));
8477                 break;
8478
8479         case offsetof(struct __sk_buff, vlan_proto):
8480                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8481                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8482                                                      target_size));
8483                 break;
8484
8485         case offsetof(struct __sk_buff, priority):
8486                 if (type == BPF_WRITE)
8487                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8488                                               bpf_target_off(struct sk_buff, priority, 4,
8489                                                              target_size));
8490                 else
8491                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8492                                               bpf_target_off(struct sk_buff, priority, 4,
8493                                                              target_size));
8494                 break;
8495
8496         case offsetof(struct __sk_buff, ingress_ifindex):
8497                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8498                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8499                                                      target_size));
8500                 break;
8501
8502         case offsetof(struct __sk_buff, ifindex):
8503                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8504                                       si->dst_reg, si->src_reg,
8505                                       offsetof(struct sk_buff, dev));
8506                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8507                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8508                                       bpf_target_off(struct net_device, ifindex, 4,
8509                                                      target_size));
8510                 break;
8511
8512         case offsetof(struct __sk_buff, hash):
8513                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8514                                       bpf_target_off(struct sk_buff, hash, 4,
8515                                                      target_size));
8516                 break;
8517
8518         case offsetof(struct __sk_buff, mark):
8519                 if (type == BPF_WRITE)
8520                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8521                                               bpf_target_off(struct sk_buff, mark, 4,
8522                                                              target_size));
8523                 else
8524                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8525                                               bpf_target_off(struct sk_buff, mark, 4,
8526                                                              target_size));
8527                 break;
8528
8529         case offsetof(struct __sk_buff, pkt_type):
8530                 *target_size = 1;
8531                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8532                                       PKT_TYPE_OFFSET());
8533                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8534 #ifdef __BIG_ENDIAN_BITFIELD
8535                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8536 #endif
8537                 break;
8538
8539         case offsetof(struct __sk_buff, queue_mapping):
8540                 if (type == BPF_WRITE) {
8541                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8542                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8543                                               bpf_target_off(struct sk_buff,
8544                                                              queue_mapping,
8545                                                              2, target_size));
8546                 } else {
8547                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8548                                               bpf_target_off(struct sk_buff,
8549                                                              queue_mapping,
8550                                                              2, target_size));
8551                 }
8552                 break;
8553
8554         case offsetof(struct __sk_buff, vlan_present):
8555                 *target_size = 1;
8556                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8557                                       PKT_VLAN_PRESENT_OFFSET());
8558                 if (PKT_VLAN_PRESENT_BIT)
8559                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8560                 if (PKT_VLAN_PRESENT_BIT < 7)
8561                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8562                 break;
8563
8564         case offsetof(struct __sk_buff, vlan_tci):
8565                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8566                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8567                                                      target_size));
8568                 break;
8569
8570         case offsetof(struct __sk_buff, cb[0]) ...
8571              offsetofend(struct __sk_buff, cb[4]) - 1:
8572                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8573                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8574                               offsetof(struct qdisc_skb_cb, data)) %
8575                              sizeof(__u64));
8576
8577                 prog->cb_access = 1;
8578                 off  = si->off;
8579                 off -= offsetof(struct __sk_buff, cb[0]);
8580                 off += offsetof(struct sk_buff, cb);
8581                 off += offsetof(struct qdisc_skb_cb, data);
8582                 if (type == BPF_WRITE)
8583                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8584                                               si->src_reg, off);
8585                 else
8586                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8587                                               si->src_reg, off);
8588                 break;
8589
8590         case offsetof(struct __sk_buff, tc_classid):
8591                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8592
8593                 off  = si->off;
8594                 off -= offsetof(struct __sk_buff, tc_classid);
8595                 off += offsetof(struct sk_buff, cb);
8596                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8597                 *target_size = 2;
8598                 if (type == BPF_WRITE)
8599                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8600                                               si->src_reg, off);
8601                 else
8602                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8603                                               si->src_reg, off);
8604                 break;
8605
8606         case offsetof(struct __sk_buff, data):
8607                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8608                                       si->dst_reg, si->src_reg,
8609                                       offsetof(struct sk_buff, data));
8610                 break;
8611
8612         case offsetof(struct __sk_buff, data_meta):
8613                 off  = si->off;
8614                 off -= offsetof(struct __sk_buff, data_meta);
8615                 off += offsetof(struct sk_buff, cb);
8616                 off += offsetof(struct bpf_skb_data_end, data_meta);
8617                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8618                                       si->src_reg, off);
8619                 break;
8620
8621         case offsetof(struct __sk_buff, data_end):
8622                 off  = si->off;
8623                 off -= offsetof(struct __sk_buff, data_end);
8624                 off += offsetof(struct sk_buff, cb);
8625                 off += offsetof(struct bpf_skb_data_end, data_end);
8626                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8627                                       si->src_reg, off);
8628                 break;
8629
8630         case offsetof(struct __sk_buff, tc_index):
8631 #ifdef CONFIG_NET_SCHED
8632                 if (type == BPF_WRITE)
8633                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8634                                               bpf_target_off(struct sk_buff, tc_index, 2,
8635                                                              target_size));
8636                 else
8637                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8638                                               bpf_target_off(struct sk_buff, tc_index, 2,
8639                                                              target_size));
8640 #else
8641                 *target_size = 2;
8642                 if (type == BPF_WRITE)
8643                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8644                 else
8645                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8646 #endif
8647                 break;
8648
8649         case offsetof(struct __sk_buff, napi_id):
8650 #if defined(CONFIG_NET_RX_BUSY_POLL)
8651                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8652                                       bpf_target_off(struct sk_buff, napi_id, 4,
8653                                                      target_size));
8654                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8655                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8656 #else
8657                 *target_size = 4;
8658                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8659 #endif
8660                 break;
8661         case offsetof(struct __sk_buff, family):
8662                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8663
8664                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8665                                       si->dst_reg, si->src_reg,
8666                                       offsetof(struct sk_buff, sk));
8667                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8668                                       bpf_target_off(struct sock_common,
8669                                                      skc_family,
8670                                                      2, target_size));
8671                 break;
8672         case offsetof(struct __sk_buff, remote_ip4):
8673                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8674
8675                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8676                                       si->dst_reg, si->src_reg,
8677                                       offsetof(struct sk_buff, sk));
8678                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8679                                       bpf_target_off(struct sock_common,
8680                                                      skc_daddr,
8681                                                      4, target_size));
8682                 break;
8683         case offsetof(struct __sk_buff, local_ip4):
8684                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8685                                           skc_rcv_saddr) != 4);
8686
8687                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8688                                       si->dst_reg, si->src_reg,
8689                                       offsetof(struct sk_buff, sk));
8690                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8691                                       bpf_target_off(struct sock_common,
8692                                                      skc_rcv_saddr,
8693                                                      4, target_size));
8694                 break;
8695         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8696              offsetof(struct __sk_buff, remote_ip6[3]):
8697 #if IS_ENABLED(CONFIG_IPV6)
8698                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8699                                           skc_v6_daddr.s6_addr32[0]) != 4);
8700
8701                 off = si->off;
8702                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8703
8704                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8705                                       si->dst_reg, si->src_reg,
8706                                       offsetof(struct sk_buff, sk));
8707                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8708                                       offsetof(struct sock_common,
8709                                                skc_v6_daddr.s6_addr32[0]) +
8710                                       off);
8711 #else
8712                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8713 #endif
8714                 break;
8715         case offsetof(struct __sk_buff, local_ip6[0]) ...
8716              offsetof(struct __sk_buff, local_ip6[3]):
8717 #if IS_ENABLED(CONFIG_IPV6)
8718                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8719                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8720
8721                 off = si->off;
8722                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8723
8724                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8725                                       si->dst_reg, si->src_reg,
8726                                       offsetof(struct sk_buff, sk));
8727                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8728                                       offsetof(struct sock_common,
8729                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8730                                       off);
8731 #else
8732                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8733 #endif
8734                 break;
8735
8736         case offsetof(struct __sk_buff, remote_port):
8737                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8738
8739                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8740                                       si->dst_reg, si->src_reg,
8741                                       offsetof(struct sk_buff, sk));
8742                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8743                                       bpf_target_off(struct sock_common,
8744                                                      skc_dport,
8745                                                      2, target_size));
8746 #ifndef __BIG_ENDIAN_BITFIELD
8747                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8748 #endif
8749                 break;
8750
8751         case offsetof(struct __sk_buff, local_port):
8752                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8753
8754                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8755                                       si->dst_reg, si->src_reg,
8756                                       offsetof(struct sk_buff, sk));
8757                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8758                                       bpf_target_off(struct sock_common,
8759                                                      skc_num, 2, target_size));
8760                 break;
8761
8762         case offsetof(struct __sk_buff, tstamp):
8763                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8764
8765                 if (type == BPF_WRITE)
8766                         *insn++ = BPF_STX_MEM(BPF_DW,
8767                                               si->dst_reg, si->src_reg,
8768                                               bpf_target_off(struct sk_buff,
8769                                                              tstamp, 8,
8770                                                              target_size));
8771                 else
8772                         *insn++ = BPF_LDX_MEM(BPF_DW,
8773                                               si->dst_reg, si->src_reg,
8774                                               bpf_target_off(struct sk_buff,
8775                                                              tstamp, 8,
8776                                                              target_size));
8777                 break;
8778
8779         case offsetof(struct __sk_buff, gso_segs):
8780                 insn = bpf_convert_shinfo_access(si, insn);
8781                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8782                                       si->dst_reg, si->dst_reg,
8783                                       bpf_target_off(struct skb_shared_info,
8784                                                      gso_segs, 2,
8785                                                      target_size));
8786                 break;
8787         case offsetof(struct __sk_buff, gso_size):
8788                 insn = bpf_convert_shinfo_access(si, insn);
8789                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8790                                       si->dst_reg, si->dst_reg,
8791                                       bpf_target_off(struct skb_shared_info,
8792                                                      gso_size, 2,
8793                                                      target_size));
8794                 break;
8795         case offsetof(struct __sk_buff, wire_len):
8796                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8797
8798                 off = si->off;
8799                 off -= offsetof(struct __sk_buff, wire_len);
8800                 off += offsetof(struct sk_buff, cb);
8801                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8802                 *target_size = 4;
8803                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8804                 break;
8805
8806         case offsetof(struct __sk_buff, sk):
8807                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8808                                       si->dst_reg, si->src_reg,
8809                                       offsetof(struct sk_buff, sk));
8810                 break;
8811         }
8812
8813         return insn - insn_buf;
8814 }
8815
8816 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8817                                 const struct bpf_insn *si,
8818                                 struct bpf_insn *insn_buf,
8819                                 struct bpf_prog *prog, u32 *target_size)
8820 {
8821         struct bpf_insn *insn = insn_buf;
8822         int off;
8823
8824         switch (si->off) {
8825         case offsetof(struct bpf_sock, bound_dev_if):
8826                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8827
8828                 if (type == BPF_WRITE)
8829                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8830                                         offsetof(struct sock, sk_bound_dev_if));
8831                 else
8832                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8833                                       offsetof(struct sock, sk_bound_dev_if));
8834                 break;
8835
8836         case offsetof(struct bpf_sock, mark):
8837                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8838
8839                 if (type == BPF_WRITE)
8840                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8841                                         offsetof(struct sock, sk_mark));
8842                 else
8843                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8844                                       offsetof(struct sock, sk_mark));
8845                 break;
8846
8847         case offsetof(struct bpf_sock, priority):
8848                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8849
8850                 if (type == BPF_WRITE)
8851                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8852                                         offsetof(struct sock, sk_priority));
8853                 else
8854                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8855                                       offsetof(struct sock, sk_priority));
8856                 break;
8857
8858         case offsetof(struct bpf_sock, family):
8859                 *insn++ = BPF_LDX_MEM(
8860                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8861                         si->dst_reg, si->src_reg,
8862                         bpf_target_off(struct sock_common,
8863                                        skc_family,
8864                                        sizeof_field(struct sock_common,
8865                                                     skc_family),
8866                                        target_size));
8867                 break;
8868
8869         case offsetof(struct bpf_sock, type):
8870                 *insn++ = BPF_LDX_MEM(
8871                         BPF_FIELD_SIZEOF(struct sock, sk_type),
8872                         si->dst_reg, si->src_reg,
8873                         bpf_target_off(struct sock, sk_type,
8874                                        sizeof_field(struct sock, sk_type),
8875                                        target_size));
8876                 break;
8877
8878         case offsetof(struct bpf_sock, protocol):
8879                 *insn++ = BPF_LDX_MEM(
8880                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8881                         si->dst_reg, si->src_reg,
8882                         bpf_target_off(struct sock, sk_protocol,
8883                                        sizeof_field(struct sock, sk_protocol),
8884                                        target_size));
8885                 break;
8886
8887         case offsetof(struct bpf_sock, src_ip4):
8888                 *insn++ = BPF_LDX_MEM(
8889                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8890                         bpf_target_off(struct sock_common, skc_rcv_saddr,
8891                                        sizeof_field(struct sock_common,
8892                                                     skc_rcv_saddr),
8893                                        target_size));
8894                 break;
8895
8896         case offsetof(struct bpf_sock, dst_ip4):
8897                 *insn++ = BPF_LDX_MEM(
8898                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8899                         bpf_target_off(struct sock_common, skc_daddr,
8900                                        sizeof_field(struct sock_common,
8901                                                     skc_daddr),
8902                                        target_size));
8903                 break;
8904
8905         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8906 #if IS_ENABLED(CONFIG_IPV6)
8907                 off = si->off;
8908                 off -= offsetof(struct bpf_sock, src_ip6[0]);
8909                 *insn++ = BPF_LDX_MEM(
8910                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8911                         bpf_target_off(
8912                                 struct sock_common,
8913                                 skc_v6_rcv_saddr.s6_addr32[0],
8914                                 sizeof_field(struct sock_common,
8915                                              skc_v6_rcv_saddr.s6_addr32[0]),
8916                                 target_size) + off);
8917 #else
8918                 (void)off;
8919                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8920 #endif
8921                 break;
8922
8923         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8924 #if IS_ENABLED(CONFIG_IPV6)
8925                 off = si->off;
8926                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8927                 *insn++ = BPF_LDX_MEM(
8928                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8929                         bpf_target_off(struct sock_common,
8930                                        skc_v6_daddr.s6_addr32[0],
8931                                        sizeof_field(struct sock_common,
8932                                                     skc_v6_daddr.s6_addr32[0]),
8933                                        target_size) + off);
8934 #else
8935                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8936                 *target_size = 4;
8937 #endif
8938                 break;
8939
8940         case offsetof(struct bpf_sock, src_port):
8941                 *insn++ = BPF_LDX_MEM(
8942                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8943                         si->dst_reg, si->src_reg,
8944                         bpf_target_off(struct sock_common, skc_num,
8945                                        sizeof_field(struct sock_common,
8946                                                     skc_num),
8947                                        target_size));
8948                 break;
8949
8950         case offsetof(struct bpf_sock, dst_port):
8951                 *insn++ = BPF_LDX_MEM(
8952                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8953                         si->dst_reg, si->src_reg,
8954                         bpf_target_off(struct sock_common, skc_dport,
8955                                        sizeof_field(struct sock_common,
8956                                                     skc_dport),
8957                                        target_size));
8958                 break;
8959
8960         case offsetof(struct bpf_sock, state):
8961                 *insn++ = BPF_LDX_MEM(
8962                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8963                         si->dst_reg, si->src_reg,
8964                         bpf_target_off(struct sock_common, skc_state,
8965                                        sizeof_field(struct sock_common,
8966                                                     skc_state),
8967                                        target_size));
8968                 break;
8969         case offsetof(struct bpf_sock, rx_queue_mapping):
8970 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8971                 *insn++ = BPF_LDX_MEM(
8972                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8973                         si->dst_reg, si->src_reg,
8974                         bpf_target_off(struct sock, sk_rx_queue_mapping,
8975                                        sizeof_field(struct sock,
8976                                                     sk_rx_queue_mapping),
8977                                        target_size));
8978                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8979                                       1);
8980                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8981 #else
8982                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8983                 *target_size = 2;
8984 #endif
8985                 break;
8986         }
8987
8988         return insn - insn_buf;
8989 }
8990
8991 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8992                                          const struct bpf_insn *si,
8993                                          struct bpf_insn *insn_buf,
8994                                          struct bpf_prog *prog, u32 *target_size)
8995 {
8996         struct bpf_insn *insn = insn_buf;
8997
8998         switch (si->off) {
8999         case offsetof(struct __sk_buff, ifindex):
9000                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9001                                       si->dst_reg, si->src_reg,
9002                                       offsetof(struct sk_buff, dev));
9003                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9004                                       bpf_target_off(struct net_device, ifindex, 4,
9005                                                      target_size));
9006                 break;
9007         default:
9008                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9009                                               target_size);
9010         }
9011
9012         return insn - insn_buf;
9013 }
9014
9015 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9016                                   const struct bpf_insn *si,
9017                                   struct bpf_insn *insn_buf,
9018                                   struct bpf_prog *prog, u32 *target_size)
9019 {
9020         struct bpf_insn *insn = insn_buf;
9021
9022         switch (si->off) {
9023         case offsetof(struct xdp_md, data):
9024                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9025                                       si->dst_reg, si->src_reg,
9026                                       offsetof(struct xdp_buff, data));
9027                 break;
9028         case offsetof(struct xdp_md, data_meta):
9029                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9030                                       si->dst_reg, si->src_reg,
9031                                       offsetof(struct xdp_buff, data_meta));
9032                 break;
9033         case offsetof(struct xdp_md, data_end):
9034                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9035                                       si->dst_reg, si->src_reg,
9036                                       offsetof(struct xdp_buff, data_end));
9037                 break;
9038         case offsetof(struct xdp_md, ingress_ifindex):
9039                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9040                                       si->dst_reg, si->src_reg,
9041                                       offsetof(struct xdp_buff, rxq));
9042                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9043                                       si->dst_reg, si->dst_reg,
9044                                       offsetof(struct xdp_rxq_info, dev));
9045                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9046                                       offsetof(struct net_device, ifindex));
9047                 break;
9048         case offsetof(struct xdp_md, rx_queue_index):
9049                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9050                                       si->dst_reg, si->src_reg,
9051                                       offsetof(struct xdp_buff, rxq));
9052                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9053                                       offsetof(struct xdp_rxq_info,
9054                                                queue_index));
9055                 break;
9056         case offsetof(struct xdp_md, egress_ifindex):
9057                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9058                                       si->dst_reg, si->src_reg,
9059                                       offsetof(struct xdp_buff, txq));
9060                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9061                                       si->dst_reg, si->dst_reg,
9062                                       offsetof(struct xdp_txq_info, dev));
9063                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9064                                       offsetof(struct net_device, ifindex));
9065                 break;
9066         }
9067
9068         return insn - insn_buf;
9069 }
9070
9071 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9072  * context Structure, F is Field in context structure that contains a pointer
9073  * to Nested Structure of type NS that has the field NF.
9074  *
9075  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9076  * sure that SIZE is not greater than actual size of S.F.NF.
9077  *
9078  * If offset OFF is provided, the load happens from that offset relative to
9079  * offset of NF.
9080  */
9081 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9082         do {                                                                   \
9083                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9084                                       si->src_reg, offsetof(S, F));            \
9085                 *insn++ = BPF_LDX_MEM(                                         \
9086                         SIZE, si->dst_reg, si->dst_reg,                        \
9087                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9088                                        target_size)                            \
9089                                 + OFF);                                        \
9090         } while (0)
9091
9092 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9093         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9094                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9095
9096 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9097  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9098  *
9099  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9100  * "register" since two registers available in convert_ctx_access are not
9101  * enough: we can't override neither SRC, since it contains value to store, nor
9102  * DST since it contains pointer to context that may be used by later
9103  * instructions. But we need a temporary place to save pointer to nested
9104  * structure whose field we want to store to.
9105  */
9106 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9107         do {                                                                   \
9108                 int tmp_reg = BPF_REG_9;                                       \
9109                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9110                         --tmp_reg;                                             \
9111                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9112                         --tmp_reg;                                             \
9113                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9114                                       offsetof(S, TF));                        \
9115                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9116                                       si->dst_reg, offsetof(S, F));            \
9117                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9118                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9119                                        target_size)                            \
9120                                 + OFF);                                        \
9121                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9122                                       offsetof(S, TF));                        \
9123         } while (0)
9124
9125 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9126                                                       TF)                      \
9127         do {                                                                   \
9128                 if (type == BPF_WRITE) {                                       \
9129                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9130                                                          OFF, TF);             \
9131                 } else {                                                       \
9132                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9133                                 S, NS, F, NF, SIZE, OFF);  \
9134                 }                                                              \
9135         } while (0)
9136
9137 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9138         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9139                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9140
9141 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9142                                         const struct bpf_insn *si,
9143                                         struct bpf_insn *insn_buf,
9144                                         struct bpf_prog *prog, u32 *target_size)
9145 {
9146         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9147         struct bpf_insn *insn = insn_buf;
9148
9149         switch (si->off) {
9150         case offsetof(struct bpf_sock_addr, user_family):
9151                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9152                                             struct sockaddr, uaddr, sa_family);
9153                 break;
9154
9155         case offsetof(struct bpf_sock_addr, user_ip4):
9156                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9157                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9158                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9159                 break;
9160
9161         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9162                 off = si->off;
9163                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9164                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9165                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9166                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9167                         tmp_reg);
9168                 break;
9169
9170         case offsetof(struct bpf_sock_addr, user_port):
9171                 /* To get port we need to know sa_family first and then treat
9172                  * sockaddr as either sockaddr_in or sockaddr_in6.
9173                  * Though we can simplify since port field has same offset and
9174                  * size in both structures.
9175                  * Here we check this invariant and use just one of the
9176                  * structures if it's true.
9177                  */
9178                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9179                              offsetof(struct sockaddr_in6, sin6_port));
9180                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9181                              sizeof_field(struct sockaddr_in6, sin6_port));
9182                 /* Account for sin6_port being smaller than user_port. */
9183                 port_size = min(port_size, BPF_LDST_BYTES(si));
9184                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9185                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9186                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9187                 break;
9188
9189         case offsetof(struct bpf_sock_addr, family):
9190                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9191                                             struct sock, sk, sk_family);
9192                 break;
9193
9194         case offsetof(struct bpf_sock_addr, type):
9195                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9196                                             struct sock, sk, sk_type);
9197                 break;
9198
9199         case offsetof(struct bpf_sock_addr, protocol):
9200                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9201                                             struct sock, sk, sk_protocol);
9202                 break;
9203
9204         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9205                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9206                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9207                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9208                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9209                 break;
9210
9211         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9212                                 msg_src_ip6[3]):
9213                 off = si->off;
9214                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9215                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9216                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9217                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9218                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9219                 break;
9220         case offsetof(struct bpf_sock_addr, sk):
9221                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9222                                       si->dst_reg, si->src_reg,
9223                                       offsetof(struct bpf_sock_addr_kern, sk));
9224                 break;
9225         }
9226
9227         return insn - insn_buf;
9228 }
9229
9230 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9231                                        const struct bpf_insn *si,
9232                                        struct bpf_insn *insn_buf,
9233                                        struct bpf_prog *prog,
9234                                        u32 *target_size)
9235 {
9236         struct bpf_insn *insn = insn_buf;
9237         int off;
9238
9239 /* Helper macro for adding read access to tcp_sock or sock fields. */
9240 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9241         do {                                                                  \
9242                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9243                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9244                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9245                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9246                         reg--;                                                \
9247                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9248                         reg--;                                                \
9249                 if (si->dst_reg == si->src_reg) {                             \
9250                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9251                                           offsetof(struct bpf_sock_ops_kern,  \
9252                                           temp));                             \
9253                         fullsock_reg = reg;                                   \
9254                         jmp += 2;                                             \
9255                 }                                                             \
9256                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9257                                                 struct bpf_sock_ops_kern,     \
9258                                                 is_fullsock),                 \
9259                                       fullsock_reg, si->src_reg,              \
9260                                       offsetof(struct bpf_sock_ops_kern,      \
9261                                                is_fullsock));                 \
9262                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9263                 if (si->dst_reg == si->src_reg)                               \
9264                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9265                                       offsetof(struct bpf_sock_ops_kern,      \
9266                                       temp));                                 \
9267                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9268                                                 struct bpf_sock_ops_kern, sk),\
9269                                       si->dst_reg, si->src_reg,               \
9270                                       offsetof(struct bpf_sock_ops_kern, sk));\
9271                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9272                                                        OBJ_FIELD),            \
9273                                       si->dst_reg, si->dst_reg,               \
9274                                       offsetof(OBJ, OBJ_FIELD));              \
9275                 if (si->dst_reg == si->src_reg) {                             \
9276                         *insn++ = BPF_JMP_A(1);                               \
9277                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9278                                       offsetof(struct bpf_sock_ops_kern,      \
9279                                       temp));                                 \
9280                 }                                                             \
9281         } while (0)
9282
9283 #define SOCK_OPS_GET_SK()                                                             \
9284         do {                                                                  \
9285                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9286                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9287                         reg--;                                                \
9288                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9289                         reg--;                                                \
9290                 if (si->dst_reg == si->src_reg) {                             \
9291                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9292                                           offsetof(struct bpf_sock_ops_kern,  \
9293                                           temp));                             \
9294                         fullsock_reg = reg;                                   \
9295                         jmp += 2;                                             \
9296                 }                                                             \
9297                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9298                                                 struct bpf_sock_ops_kern,     \
9299                                                 is_fullsock),                 \
9300                                       fullsock_reg, si->src_reg,              \
9301                                       offsetof(struct bpf_sock_ops_kern,      \
9302                                                is_fullsock));                 \
9303                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9304                 if (si->dst_reg == si->src_reg)                               \
9305                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9306                                       offsetof(struct bpf_sock_ops_kern,      \
9307                                       temp));                                 \
9308                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9309                                                 struct bpf_sock_ops_kern, sk),\
9310                                       si->dst_reg, si->src_reg,               \
9311                                       offsetof(struct bpf_sock_ops_kern, sk));\
9312                 if (si->dst_reg == si->src_reg) {                             \
9313                         *insn++ = BPF_JMP_A(1);                               \
9314                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9315                                       offsetof(struct bpf_sock_ops_kern,      \
9316                                       temp));                                 \
9317                 }                                                             \
9318         } while (0)
9319
9320 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9321                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9322
9323 /* Helper macro for adding write access to tcp_sock or sock fields.
9324  * The macro is called with two registers, dst_reg which contains a pointer
9325  * to ctx (context) and src_reg which contains the value that should be
9326  * stored. However, we need an additional register since we cannot overwrite
9327  * dst_reg because it may be used later in the program.
9328  * Instead we "borrow" one of the other register. We first save its value
9329  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9330  * it at the end of the macro.
9331  */
9332 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9333         do {                                                                  \
9334                 int reg = BPF_REG_9;                                          \
9335                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9336                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9337                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9338                         reg--;                                                \
9339                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9340                         reg--;                                                \
9341                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9342                                       offsetof(struct bpf_sock_ops_kern,      \
9343                                                temp));                        \
9344                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9345                                                 struct bpf_sock_ops_kern,     \
9346                                                 is_fullsock),                 \
9347                                       reg, si->dst_reg,                       \
9348                                       offsetof(struct bpf_sock_ops_kern,      \
9349                                                is_fullsock));                 \
9350                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9351                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9352                                                 struct bpf_sock_ops_kern, sk),\
9353                                       reg, si->dst_reg,                       \
9354                                       offsetof(struct bpf_sock_ops_kern, sk));\
9355                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9356                                       reg, si->src_reg,                       \
9357                                       offsetof(OBJ, OBJ_FIELD));              \
9358                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9359                                       offsetof(struct bpf_sock_ops_kern,      \
9360                                                temp));                        \
9361         } while (0)
9362
9363 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9364         do {                                                                  \
9365                 if (TYPE == BPF_WRITE)                                        \
9366                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9367                 else                                                          \
9368                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9369         } while (0)
9370
9371         if (insn > insn_buf)
9372                 return insn - insn_buf;
9373
9374         switch (si->off) {
9375         case offsetof(struct bpf_sock_ops, op):
9376                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9377                                                        op),
9378                                       si->dst_reg, si->src_reg,
9379                                       offsetof(struct bpf_sock_ops_kern, op));
9380                 break;
9381
9382         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9383              offsetof(struct bpf_sock_ops, replylong[3]):
9384                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9385                              sizeof_field(struct bpf_sock_ops_kern, reply));
9386                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9387                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9388                 off = si->off;
9389                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9390                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9391                 if (type == BPF_WRITE)
9392                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9393                                               off);
9394                 else
9395                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9396                                               off);
9397                 break;
9398
9399         case offsetof(struct bpf_sock_ops, family):
9400                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9401
9402                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9403                                               struct bpf_sock_ops_kern, sk),
9404                                       si->dst_reg, si->src_reg,
9405                                       offsetof(struct bpf_sock_ops_kern, sk));
9406                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9407                                       offsetof(struct sock_common, skc_family));
9408                 break;
9409
9410         case offsetof(struct bpf_sock_ops, remote_ip4):
9411                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9412
9413                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9414                                                 struct bpf_sock_ops_kern, sk),
9415                                       si->dst_reg, si->src_reg,
9416                                       offsetof(struct bpf_sock_ops_kern, sk));
9417                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9418                                       offsetof(struct sock_common, skc_daddr));
9419                 break;
9420
9421         case offsetof(struct bpf_sock_ops, local_ip4):
9422                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9423                                           skc_rcv_saddr) != 4);
9424
9425                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9426                                               struct bpf_sock_ops_kern, sk),
9427                                       si->dst_reg, si->src_reg,
9428                                       offsetof(struct bpf_sock_ops_kern, sk));
9429                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9430                                       offsetof(struct sock_common,
9431                                                skc_rcv_saddr));
9432                 break;
9433
9434         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9435              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9436 #if IS_ENABLED(CONFIG_IPV6)
9437                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9438                                           skc_v6_daddr.s6_addr32[0]) != 4);
9439
9440                 off = si->off;
9441                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9442                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9443                                                 struct bpf_sock_ops_kern, sk),
9444                                       si->dst_reg, si->src_reg,
9445                                       offsetof(struct bpf_sock_ops_kern, sk));
9446                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9447                                       offsetof(struct sock_common,
9448                                                skc_v6_daddr.s6_addr32[0]) +
9449                                       off);
9450 #else
9451                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9452 #endif
9453                 break;
9454
9455         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9456              offsetof(struct bpf_sock_ops, local_ip6[3]):
9457 #if IS_ENABLED(CONFIG_IPV6)
9458                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9459                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9460
9461                 off = si->off;
9462                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9463                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9464                                                 struct bpf_sock_ops_kern, sk),
9465                                       si->dst_reg, si->src_reg,
9466                                       offsetof(struct bpf_sock_ops_kern, sk));
9467                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9468                                       offsetof(struct sock_common,
9469                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9470                                       off);
9471 #else
9472                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9473 #endif
9474                 break;
9475
9476         case offsetof(struct bpf_sock_ops, remote_port):
9477                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9478
9479                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9480                                                 struct bpf_sock_ops_kern, sk),
9481                                       si->dst_reg, si->src_reg,
9482                                       offsetof(struct bpf_sock_ops_kern, sk));
9483                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9484                                       offsetof(struct sock_common, skc_dport));
9485 #ifndef __BIG_ENDIAN_BITFIELD
9486                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9487 #endif
9488                 break;
9489
9490         case offsetof(struct bpf_sock_ops, local_port):
9491                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9492
9493                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9494                                                 struct bpf_sock_ops_kern, sk),
9495                                       si->dst_reg, si->src_reg,
9496                                       offsetof(struct bpf_sock_ops_kern, sk));
9497                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9498                                       offsetof(struct sock_common, skc_num));
9499                 break;
9500
9501         case offsetof(struct bpf_sock_ops, is_fullsock):
9502                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9503                                                 struct bpf_sock_ops_kern,
9504                                                 is_fullsock),
9505                                       si->dst_reg, si->src_reg,
9506                                       offsetof(struct bpf_sock_ops_kern,
9507                                                is_fullsock));
9508                 break;
9509
9510         case offsetof(struct bpf_sock_ops, state):
9511                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9512
9513                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9514                                                 struct bpf_sock_ops_kern, sk),
9515                                       si->dst_reg, si->src_reg,
9516                                       offsetof(struct bpf_sock_ops_kern, sk));
9517                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9518                                       offsetof(struct sock_common, skc_state));
9519                 break;
9520
9521         case offsetof(struct bpf_sock_ops, rtt_min):
9522                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9523                              sizeof(struct minmax));
9524                 BUILD_BUG_ON(sizeof(struct minmax) <
9525                              sizeof(struct minmax_sample));
9526
9527                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9528                                                 struct bpf_sock_ops_kern, sk),
9529                                       si->dst_reg, si->src_reg,
9530                                       offsetof(struct bpf_sock_ops_kern, sk));
9531                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9532                                       offsetof(struct tcp_sock, rtt_min) +
9533                                       sizeof_field(struct minmax_sample, t));
9534                 break;
9535
9536         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9537                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9538                                    struct tcp_sock);
9539                 break;
9540
9541         case offsetof(struct bpf_sock_ops, sk_txhash):
9542                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9543                                           struct sock, type);
9544                 break;
9545         case offsetof(struct bpf_sock_ops, snd_cwnd):
9546                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9547                 break;
9548         case offsetof(struct bpf_sock_ops, srtt_us):
9549                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9550                 break;
9551         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9552                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9553                 break;
9554         case offsetof(struct bpf_sock_ops, rcv_nxt):
9555                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9556                 break;
9557         case offsetof(struct bpf_sock_ops, snd_nxt):
9558                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9559                 break;
9560         case offsetof(struct bpf_sock_ops, snd_una):
9561                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9562                 break;
9563         case offsetof(struct bpf_sock_ops, mss_cache):
9564                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9565                 break;
9566         case offsetof(struct bpf_sock_ops, ecn_flags):
9567                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9568                 break;
9569         case offsetof(struct bpf_sock_ops, rate_delivered):
9570                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9571                 break;
9572         case offsetof(struct bpf_sock_ops, rate_interval_us):
9573                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9574                 break;
9575         case offsetof(struct bpf_sock_ops, packets_out):
9576                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9577                 break;
9578         case offsetof(struct bpf_sock_ops, retrans_out):
9579                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9580                 break;
9581         case offsetof(struct bpf_sock_ops, total_retrans):
9582                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9583                 break;
9584         case offsetof(struct bpf_sock_ops, segs_in):
9585                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9586                 break;
9587         case offsetof(struct bpf_sock_ops, data_segs_in):
9588                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9589                 break;
9590         case offsetof(struct bpf_sock_ops, segs_out):
9591                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9592                 break;
9593         case offsetof(struct bpf_sock_ops, data_segs_out):
9594                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9595                 break;
9596         case offsetof(struct bpf_sock_ops, lost_out):
9597                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9598                 break;
9599         case offsetof(struct bpf_sock_ops, sacked_out):
9600                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9601                 break;
9602         case offsetof(struct bpf_sock_ops, bytes_received):
9603                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9604                 break;
9605         case offsetof(struct bpf_sock_ops, bytes_acked):
9606                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9607                 break;
9608         case offsetof(struct bpf_sock_ops, sk):
9609                 SOCK_OPS_GET_SK();
9610                 break;
9611         case offsetof(struct bpf_sock_ops, skb_data_end):
9612                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9613                                                        skb_data_end),
9614                                       si->dst_reg, si->src_reg,
9615                                       offsetof(struct bpf_sock_ops_kern,
9616                                                skb_data_end));
9617                 break;
9618         case offsetof(struct bpf_sock_ops, skb_data):
9619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9620                                                        skb),
9621                                       si->dst_reg, si->src_reg,
9622                                       offsetof(struct bpf_sock_ops_kern,
9623                                                skb));
9624                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9625                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9626                                       si->dst_reg, si->dst_reg,
9627                                       offsetof(struct sk_buff, data));
9628                 break;
9629         case offsetof(struct bpf_sock_ops, skb_len):
9630                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9631                                                        skb),
9632                                       si->dst_reg, si->src_reg,
9633                                       offsetof(struct bpf_sock_ops_kern,
9634                                                skb));
9635                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9636                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9637                                       si->dst_reg, si->dst_reg,
9638                                       offsetof(struct sk_buff, len));
9639                 break;
9640         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9641                 off = offsetof(struct sk_buff, cb);
9642                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9643                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9644                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9645                                                        skb),
9646                                       si->dst_reg, si->src_reg,
9647                                       offsetof(struct bpf_sock_ops_kern,
9648                                                skb));
9649                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9650                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9651                                                        tcp_flags),
9652                                       si->dst_reg, si->dst_reg, off);
9653                 break;
9654         }
9655         return insn - insn_buf;
9656 }
9657
9658 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9659                                      const struct bpf_insn *si,
9660                                      struct bpf_insn *insn_buf,
9661                                      struct bpf_prog *prog, u32 *target_size)
9662 {
9663         struct bpf_insn *insn = insn_buf;
9664         int off;
9665
9666         switch (si->off) {
9667         case offsetof(struct __sk_buff, data_end):
9668                 off  = si->off;
9669                 off -= offsetof(struct __sk_buff, data_end);
9670                 off += offsetof(struct sk_buff, cb);
9671                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9672                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9673                                       si->src_reg, off);
9674                 break;
9675         default:
9676                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9677                                               target_size);
9678         }
9679
9680         return insn - insn_buf;
9681 }
9682
9683 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9684                                      const struct bpf_insn *si,
9685                                      struct bpf_insn *insn_buf,
9686                                      struct bpf_prog *prog, u32 *target_size)
9687 {
9688         struct bpf_insn *insn = insn_buf;
9689 #if IS_ENABLED(CONFIG_IPV6)
9690         int off;
9691 #endif
9692
9693         /* convert ctx uses the fact sg element is first in struct */
9694         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9695
9696         switch (si->off) {
9697         case offsetof(struct sk_msg_md, data):
9698                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9699                                       si->dst_reg, si->src_reg,
9700                                       offsetof(struct sk_msg, data));
9701                 break;
9702         case offsetof(struct sk_msg_md, data_end):
9703                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9704                                       si->dst_reg, si->src_reg,
9705                                       offsetof(struct sk_msg, data_end));
9706                 break;
9707         case offsetof(struct sk_msg_md, family):
9708                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9709
9710                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9711                                               struct sk_msg, sk),
9712                                       si->dst_reg, si->src_reg,
9713                                       offsetof(struct sk_msg, sk));
9714                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9715                                       offsetof(struct sock_common, skc_family));
9716                 break;
9717
9718         case offsetof(struct sk_msg_md, remote_ip4):
9719                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9720
9721                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9722                                                 struct sk_msg, sk),
9723                                       si->dst_reg, si->src_reg,
9724                                       offsetof(struct sk_msg, sk));
9725                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9726                                       offsetof(struct sock_common, skc_daddr));
9727                 break;
9728
9729         case offsetof(struct sk_msg_md, local_ip4):
9730                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9731                                           skc_rcv_saddr) != 4);
9732
9733                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9734                                               struct sk_msg, sk),
9735                                       si->dst_reg, si->src_reg,
9736                                       offsetof(struct sk_msg, sk));
9737                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9738                                       offsetof(struct sock_common,
9739                                                skc_rcv_saddr));
9740                 break;
9741
9742         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9743              offsetof(struct sk_msg_md, remote_ip6[3]):
9744 #if IS_ENABLED(CONFIG_IPV6)
9745                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9746                                           skc_v6_daddr.s6_addr32[0]) != 4);
9747
9748                 off = si->off;
9749                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9750                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9751                                                 struct sk_msg, sk),
9752                                       si->dst_reg, si->src_reg,
9753                                       offsetof(struct sk_msg, sk));
9754                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9755                                       offsetof(struct sock_common,
9756                                                skc_v6_daddr.s6_addr32[0]) +
9757                                       off);
9758 #else
9759                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9760 #endif
9761                 break;
9762
9763         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9764              offsetof(struct sk_msg_md, local_ip6[3]):
9765 #if IS_ENABLED(CONFIG_IPV6)
9766                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9767                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9768
9769                 off = si->off;
9770                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9771                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9772                                                 struct sk_msg, sk),
9773                                       si->dst_reg, si->src_reg,
9774                                       offsetof(struct sk_msg, sk));
9775                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9776                                       offsetof(struct sock_common,
9777                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9778                                       off);
9779 #else
9780                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9781 #endif
9782                 break;
9783
9784         case offsetof(struct sk_msg_md, remote_port):
9785                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9786
9787                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9788                                                 struct sk_msg, sk),
9789                                       si->dst_reg, si->src_reg,
9790                                       offsetof(struct sk_msg, sk));
9791                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9792                                       offsetof(struct sock_common, skc_dport));
9793 #ifndef __BIG_ENDIAN_BITFIELD
9794                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9795 #endif
9796                 break;
9797
9798         case offsetof(struct sk_msg_md, local_port):
9799                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9800
9801                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9802                                                 struct sk_msg, sk),
9803                                       si->dst_reg, si->src_reg,
9804                                       offsetof(struct sk_msg, sk));
9805                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9806                                       offsetof(struct sock_common, skc_num));
9807                 break;
9808
9809         case offsetof(struct sk_msg_md, size):
9810                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9811                                       si->dst_reg, si->src_reg,
9812                                       offsetof(struct sk_msg_sg, size));
9813                 break;
9814
9815         case offsetof(struct sk_msg_md, sk):
9816                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9817                                       si->dst_reg, si->src_reg,
9818                                       offsetof(struct sk_msg, sk));
9819                 break;
9820         }
9821
9822         return insn - insn_buf;
9823 }
9824
9825 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9826         .get_func_proto         = sk_filter_func_proto,
9827         .is_valid_access        = sk_filter_is_valid_access,
9828         .convert_ctx_access     = bpf_convert_ctx_access,
9829         .gen_ld_abs             = bpf_gen_ld_abs,
9830 };
9831
9832 const struct bpf_prog_ops sk_filter_prog_ops = {
9833         .test_run               = bpf_prog_test_run_skb,
9834 };
9835
9836 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9837         .get_func_proto         = tc_cls_act_func_proto,
9838         .is_valid_access        = tc_cls_act_is_valid_access,
9839         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
9840         .gen_prologue           = tc_cls_act_prologue,
9841         .gen_ld_abs             = bpf_gen_ld_abs,
9842 };
9843
9844 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9845         .test_run               = bpf_prog_test_run_skb,
9846 };
9847
9848 const struct bpf_verifier_ops xdp_verifier_ops = {
9849         .get_func_proto         = xdp_func_proto,
9850         .is_valid_access        = xdp_is_valid_access,
9851         .convert_ctx_access     = xdp_convert_ctx_access,
9852         .gen_prologue           = bpf_noop_prologue,
9853 };
9854
9855 const struct bpf_prog_ops xdp_prog_ops = {
9856         .test_run               = bpf_prog_test_run_xdp,
9857 };
9858
9859 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9860         .get_func_proto         = cg_skb_func_proto,
9861         .is_valid_access        = cg_skb_is_valid_access,
9862         .convert_ctx_access     = bpf_convert_ctx_access,
9863 };
9864
9865 const struct bpf_prog_ops cg_skb_prog_ops = {
9866         .test_run               = bpf_prog_test_run_skb,
9867 };
9868
9869 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9870         .get_func_proto         = lwt_in_func_proto,
9871         .is_valid_access        = lwt_is_valid_access,
9872         .convert_ctx_access     = bpf_convert_ctx_access,
9873 };
9874
9875 const struct bpf_prog_ops lwt_in_prog_ops = {
9876         .test_run               = bpf_prog_test_run_skb,
9877 };
9878
9879 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9880         .get_func_proto         = lwt_out_func_proto,
9881         .is_valid_access        = lwt_is_valid_access,
9882         .convert_ctx_access     = bpf_convert_ctx_access,
9883 };
9884
9885 const struct bpf_prog_ops lwt_out_prog_ops = {
9886         .test_run               = bpf_prog_test_run_skb,
9887 };
9888
9889 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9890         .get_func_proto         = lwt_xmit_func_proto,
9891         .is_valid_access        = lwt_is_valid_access,
9892         .convert_ctx_access     = bpf_convert_ctx_access,
9893         .gen_prologue           = tc_cls_act_prologue,
9894 };
9895
9896 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9897         .test_run               = bpf_prog_test_run_skb,
9898 };
9899
9900 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9901         .get_func_proto         = lwt_seg6local_func_proto,
9902         .is_valid_access        = lwt_is_valid_access,
9903         .convert_ctx_access     = bpf_convert_ctx_access,
9904 };
9905
9906 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9907         .test_run               = bpf_prog_test_run_skb,
9908 };
9909
9910 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9911         .get_func_proto         = sock_filter_func_proto,
9912         .is_valid_access        = sock_filter_is_valid_access,
9913         .convert_ctx_access     = bpf_sock_convert_ctx_access,
9914 };
9915
9916 const struct bpf_prog_ops cg_sock_prog_ops = {
9917 };
9918
9919 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9920         .get_func_proto         = sock_addr_func_proto,
9921         .is_valid_access        = sock_addr_is_valid_access,
9922         .convert_ctx_access     = sock_addr_convert_ctx_access,
9923 };
9924
9925 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9926 };
9927
9928 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9929         .get_func_proto         = sock_ops_func_proto,
9930         .is_valid_access        = sock_ops_is_valid_access,
9931         .convert_ctx_access     = sock_ops_convert_ctx_access,
9932 };
9933
9934 const struct bpf_prog_ops sock_ops_prog_ops = {
9935 };
9936
9937 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9938         .get_func_proto         = sk_skb_func_proto,
9939         .is_valid_access        = sk_skb_is_valid_access,
9940         .convert_ctx_access     = sk_skb_convert_ctx_access,
9941         .gen_prologue           = sk_skb_prologue,
9942 };
9943
9944 const struct bpf_prog_ops sk_skb_prog_ops = {
9945 };
9946
9947 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9948         .get_func_proto         = sk_msg_func_proto,
9949         .is_valid_access        = sk_msg_is_valid_access,
9950         .convert_ctx_access     = sk_msg_convert_ctx_access,
9951         .gen_prologue           = bpf_noop_prologue,
9952 };
9953
9954 const struct bpf_prog_ops sk_msg_prog_ops = {
9955 };
9956
9957 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9958         .get_func_proto         = flow_dissector_func_proto,
9959         .is_valid_access        = flow_dissector_is_valid_access,
9960         .convert_ctx_access     = flow_dissector_convert_ctx_access,
9961 };
9962
9963 const struct bpf_prog_ops flow_dissector_prog_ops = {
9964         .test_run               = bpf_prog_test_run_flow_dissector,
9965 };
9966
9967 int sk_detach_filter(struct sock *sk)
9968 {
9969         int ret = -ENOENT;
9970         struct sk_filter *filter;
9971
9972         if (sock_flag(sk, SOCK_FILTER_LOCKED))
9973                 return -EPERM;
9974
9975         filter = rcu_dereference_protected(sk->sk_filter,
9976                                            lockdep_sock_is_held(sk));
9977         if (filter) {
9978                 RCU_INIT_POINTER(sk->sk_filter, NULL);
9979                 sk_filter_uncharge(sk, filter);
9980                 ret = 0;
9981         }
9982
9983         return ret;
9984 }
9985 EXPORT_SYMBOL_GPL(sk_detach_filter);
9986
9987 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9988                   unsigned int len)
9989 {
9990         struct sock_fprog_kern *fprog;
9991         struct sk_filter *filter;
9992         int ret = 0;
9993
9994         lock_sock(sk);
9995         filter = rcu_dereference_protected(sk->sk_filter,
9996                                            lockdep_sock_is_held(sk));
9997         if (!filter)
9998                 goto out;
9999
10000         /* We're copying the filter that has been originally attached,
10001          * so no conversion/decode needed anymore. eBPF programs that
10002          * have no original program cannot be dumped through this.
10003          */
10004         ret = -EACCES;
10005         fprog = filter->prog->orig_prog;
10006         if (!fprog)
10007                 goto out;
10008
10009         ret = fprog->len;
10010         if (!len)
10011                 /* User space only enquires number of filter blocks. */
10012                 goto out;
10013
10014         ret = -EINVAL;
10015         if (len < fprog->len)
10016                 goto out;
10017
10018         ret = -EFAULT;
10019         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10020                 goto out;
10021
10022         /* Instead of bytes, the API requests to return the number
10023          * of filter blocks.
10024          */
10025         ret = fprog->len;
10026 out:
10027         release_sock(sk);
10028         return ret;
10029 }
10030
10031 #ifdef CONFIG_INET
10032 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10033                                     struct sock_reuseport *reuse,
10034                                     struct sock *sk, struct sk_buff *skb,
10035                                     u32 hash)
10036 {
10037         reuse_kern->skb = skb;
10038         reuse_kern->sk = sk;
10039         reuse_kern->selected_sk = NULL;
10040         reuse_kern->data_end = skb->data + skb_headlen(skb);
10041         reuse_kern->hash = hash;
10042         reuse_kern->reuseport_id = reuse->reuseport_id;
10043         reuse_kern->bind_inany = reuse->bind_inany;
10044 }
10045
10046 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10047                                   struct bpf_prog *prog, struct sk_buff *skb,
10048                                   u32 hash)
10049 {
10050         struct sk_reuseport_kern reuse_kern;
10051         enum sk_action action;
10052
10053         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
10054         action = BPF_PROG_RUN(prog, &reuse_kern);
10055
10056         if (action == SK_PASS)
10057                 return reuse_kern.selected_sk;
10058         else
10059                 return ERR_PTR(-ECONNREFUSED);
10060 }
10061
10062 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10063            struct bpf_map *, map, void *, key, u32, flags)
10064 {
10065         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10066         struct sock_reuseport *reuse;
10067         struct sock *selected_sk;
10068
10069         selected_sk = map->ops->map_lookup_elem(map, key);
10070         if (!selected_sk)
10071                 return -ENOENT;
10072
10073         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10074         if (!reuse) {
10075                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10076                 if (sk_is_refcounted(selected_sk))
10077                         sock_put(selected_sk);
10078
10079                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10080                  * The only (!reuse) case here is - the sk has already been
10081                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10082                  *
10083                  * Other maps (e.g. sock_map) do not provide this guarantee and
10084                  * the sk may never be in the reuseport group to begin with.
10085                  */
10086                 return is_sockarray ? -ENOENT : -EINVAL;
10087         }
10088
10089         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10090                 struct sock *sk = reuse_kern->sk;
10091
10092                 if (sk->sk_protocol != selected_sk->sk_protocol)
10093                         return -EPROTOTYPE;
10094                 else if (sk->sk_family != selected_sk->sk_family)
10095                         return -EAFNOSUPPORT;
10096
10097                 /* Catch all. Likely bound to a different sockaddr. */
10098                 return -EBADFD;
10099         }
10100
10101         reuse_kern->selected_sk = selected_sk;
10102
10103         return 0;
10104 }
10105
10106 static const struct bpf_func_proto sk_select_reuseport_proto = {
10107         .func           = sk_select_reuseport,
10108         .gpl_only       = false,
10109         .ret_type       = RET_INTEGER,
10110         .arg1_type      = ARG_PTR_TO_CTX,
10111         .arg2_type      = ARG_CONST_MAP_PTR,
10112         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10113         .arg4_type      = ARG_ANYTHING,
10114 };
10115
10116 BPF_CALL_4(sk_reuseport_load_bytes,
10117            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10118            void *, to, u32, len)
10119 {
10120         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10121 }
10122
10123 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10124         .func           = sk_reuseport_load_bytes,
10125         .gpl_only       = false,
10126         .ret_type       = RET_INTEGER,
10127         .arg1_type      = ARG_PTR_TO_CTX,
10128         .arg2_type      = ARG_ANYTHING,
10129         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10130         .arg4_type      = ARG_CONST_SIZE,
10131 };
10132
10133 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10134            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10135            void *, to, u32, len, u32, start_header)
10136 {
10137         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10138                                                len, start_header);
10139 }
10140
10141 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10142         .func           = sk_reuseport_load_bytes_relative,
10143         .gpl_only       = false,
10144         .ret_type       = RET_INTEGER,
10145         .arg1_type      = ARG_PTR_TO_CTX,
10146         .arg2_type      = ARG_ANYTHING,
10147         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10148         .arg4_type      = ARG_CONST_SIZE,
10149         .arg5_type      = ARG_ANYTHING,
10150 };
10151
10152 static const struct bpf_func_proto *
10153 sk_reuseport_func_proto(enum bpf_func_id func_id,
10154                         const struct bpf_prog *prog)
10155 {
10156         switch (func_id) {
10157         case BPF_FUNC_sk_select_reuseport:
10158                 return &sk_select_reuseport_proto;
10159         case BPF_FUNC_skb_load_bytes:
10160                 return &sk_reuseport_load_bytes_proto;
10161         case BPF_FUNC_skb_load_bytes_relative:
10162                 return &sk_reuseport_load_bytes_relative_proto;
10163         default:
10164                 return bpf_base_func_proto(func_id);
10165         }
10166 }
10167
10168 static bool
10169 sk_reuseport_is_valid_access(int off, int size,
10170                              enum bpf_access_type type,
10171                              const struct bpf_prog *prog,
10172                              struct bpf_insn_access_aux *info)
10173 {
10174         const u32 size_default = sizeof(__u32);
10175
10176         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10177             off % size || type != BPF_READ)
10178                 return false;
10179
10180         switch (off) {
10181         case offsetof(struct sk_reuseport_md, data):
10182                 info->reg_type = PTR_TO_PACKET;
10183                 return size == sizeof(__u64);
10184
10185         case offsetof(struct sk_reuseport_md, data_end):
10186                 info->reg_type = PTR_TO_PACKET_END;
10187                 return size == sizeof(__u64);
10188
10189         case offsetof(struct sk_reuseport_md, hash):
10190                 return size == size_default;
10191
10192         /* Fields that allow narrowing */
10193         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10194                 if (size < sizeof_field(struct sk_buff, protocol))
10195                         return false;
10196                 fallthrough;
10197         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10198         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10199         case bpf_ctx_range(struct sk_reuseport_md, len):
10200                 bpf_ctx_record_field_size(info, size_default);
10201                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10202
10203         default:
10204                 return false;
10205         }
10206 }
10207
10208 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10209         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10210                               si->dst_reg, si->src_reg,                 \
10211                               bpf_target_off(struct sk_reuseport_kern, F, \
10212                                              sizeof_field(struct sk_reuseport_kern, F), \
10213                                              target_size));             \
10214         })
10215
10216 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10217         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10218                                     struct sk_buff,                     \
10219                                     skb,                                \
10220                                     SKB_FIELD)
10221
10222 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10223         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10224                                     struct sock,                        \
10225                                     sk,                                 \
10226                                     SK_FIELD)
10227
10228 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10229                                            const struct bpf_insn *si,
10230                                            struct bpf_insn *insn_buf,
10231                                            struct bpf_prog *prog,
10232                                            u32 *target_size)
10233 {
10234         struct bpf_insn *insn = insn_buf;
10235
10236         switch (si->off) {
10237         case offsetof(struct sk_reuseport_md, data):
10238                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10239                 break;
10240
10241         case offsetof(struct sk_reuseport_md, len):
10242                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10243                 break;
10244
10245         case offsetof(struct sk_reuseport_md, eth_protocol):
10246                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10247                 break;
10248
10249         case offsetof(struct sk_reuseport_md, ip_protocol):
10250                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10251                 break;
10252
10253         case offsetof(struct sk_reuseport_md, data_end):
10254                 SK_REUSEPORT_LOAD_FIELD(data_end);
10255                 break;
10256
10257         case offsetof(struct sk_reuseport_md, hash):
10258                 SK_REUSEPORT_LOAD_FIELD(hash);
10259                 break;
10260
10261         case offsetof(struct sk_reuseport_md, bind_inany):
10262                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10263                 break;
10264         }
10265
10266         return insn - insn_buf;
10267 }
10268
10269 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10270         .get_func_proto         = sk_reuseport_func_proto,
10271         .is_valid_access        = sk_reuseport_is_valid_access,
10272         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10273 };
10274
10275 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10276 };
10277
10278 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10279 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10280
10281 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10282            struct sock *, sk, u64, flags)
10283 {
10284         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10285                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10286                 return -EINVAL;
10287         if (unlikely(sk && sk_is_refcounted(sk)))
10288                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10289         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10290                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10291
10292         /* Check if socket is suitable for packet L3/L4 protocol */
10293         if (sk && sk->sk_protocol != ctx->protocol)
10294                 return -EPROTOTYPE;
10295         if (sk && sk->sk_family != ctx->family &&
10296             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10297                 return -EAFNOSUPPORT;
10298
10299         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10300                 return -EEXIST;
10301
10302         /* Select socket as lookup result */
10303         ctx->selected_sk = sk;
10304         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10305         return 0;
10306 }
10307
10308 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10309         .func           = bpf_sk_lookup_assign,
10310         .gpl_only       = false,
10311         .ret_type       = RET_INTEGER,
10312         .arg1_type      = ARG_PTR_TO_CTX,
10313         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10314         .arg3_type      = ARG_ANYTHING,
10315 };
10316
10317 static const struct bpf_func_proto *
10318 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10319 {
10320         switch (func_id) {
10321         case BPF_FUNC_perf_event_output:
10322                 return &bpf_event_output_data_proto;
10323         case BPF_FUNC_sk_assign:
10324                 return &bpf_sk_lookup_assign_proto;
10325         case BPF_FUNC_sk_release:
10326                 return &bpf_sk_release_proto;
10327         default:
10328                 return bpf_sk_base_func_proto(func_id);
10329         }
10330 }
10331
10332 static bool sk_lookup_is_valid_access(int off, int size,
10333                                       enum bpf_access_type type,
10334                                       const struct bpf_prog *prog,
10335                                       struct bpf_insn_access_aux *info)
10336 {
10337         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10338                 return false;
10339         if (off % size != 0)
10340                 return false;
10341         if (type != BPF_READ)
10342                 return false;
10343
10344         switch (off) {
10345         case offsetof(struct bpf_sk_lookup, sk):
10346                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10347                 return size == sizeof(__u64);
10348
10349         case bpf_ctx_range(struct bpf_sk_lookup, family):
10350         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10351         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10352         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10353         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10354         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10355         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10356         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10357                 bpf_ctx_record_field_size(info, sizeof(__u32));
10358                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10359
10360         default:
10361                 return false;
10362         }
10363 }
10364
10365 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10366                                         const struct bpf_insn *si,
10367                                         struct bpf_insn *insn_buf,
10368                                         struct bpf_prog *prog,
10369                                         u32 *target_size)
10370 {
10371         struct bpf_insn *insn = insn_buf;
10372
10373         switch (si->off) {
10374         case offsetof(struct bpf_sk_lookup, sk):
10375                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10376                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10377                 break;
10378
10379         case offsetof(struct bpf_sk_lookup, family):
10380                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10381                                       bpf_target_off(struct bpf_sk_lookup_kern,
10382                                                      family, 2, target_size));
10383                 break;
10384
10385         case offsetof(struct bpf_sk_lookup, protocol):
10386                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10387                                       bpf_target_off(struct bpf_sk_lookup_kern,
10388                                                      protocol, 2, target_size));
10389                 break;
10390
10391         case offsetof(struct bpf_sk_lookup, remote_ip4):
10392                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10393                                       bpf_target_off(struct bpf_sk_lookup_kern,
10394                                                      v4.saddr, 4, target_size));
10395                 break;
10396
10397         case offsetof(struct bpf_sk_lookup, local_ip4):
10398                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10399                                       bpf_target_off(struct bpf_sk_lookup_kern,
10400                                                      v4.daddr, 4, target_size));
10401                 break;
10402
10403         case bpf_ctx_range_till(struct bpf_sk_lookup,
10404                                 remote_ip6[0], remote_ip6[3]): {
10405 #if IS_ENABLED(CONFIG_IPV6)
10406                 int off = si->off;
10407
10408                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10409                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10410                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10411                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10412                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10413                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10414 #else
10415                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10416 #endif
10417                 break;
10418         }
10419         case bpf_ctx_range_till(struct bpf_sk_lookup,
10420                                 local_ip6[0], local_ip6[3]): {
10421 #if IS_ENABLED(CONFIG_IPV6)
10422                 int off = si->off;
10423
10424                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10425                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10426                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10427                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10428                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10429                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10430 #else
10431                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10432 #endif
10433                 break;
10434         }
10435         case offsetof(struct bpf_sk_lookup, remote_port):
10436                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10437                                       bpf_target_off(struct bpf_sk_lookup_kern,
10438                                                      sport, 2, target_size));
10439                 break;
10440
10441         case offsetof(struct bpf_sk_lookup, local_port):
10442                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10443                                       bpf_target_off(struct bpf_sk_lookup_kern,
10444                                                      dport, 2, target_size));
10445                 break;
10446         }
10447
10448         return insn - insn_buf;
10449 }
10450
10451 const struct bpf_prog_ops sk_lookup_prog_ops = {
10452 };
10453
10454 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10455         .get_func_proto         = sk_lookup_func_proto,
10456         .is_valid_access        = sk_lookup_is_valid_access,
10457         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10458 };
10459
10460 #endif /* CONFIG_INET */
10461
10462 DEFINE_BPF_DISPATCHER(xdp)
10463
10464 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10465 {
10466         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10467 }
10468
10469 #ifdef CONFIG_DEBUG_INFO_BTF
10470 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10471 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10472 BTF_SOCK_TYPE_xxx
10473 #undef BTF_SOCK_TYPE
10474 #else
10475 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10476 #endif
10477
10478 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10479 {
10480         /* tcp6_sock type is not generated in dwarf and hence btf,
10481          * trigger an explicit type generation here.
10482          */
10483         BTF_TYPE_EMIT(struct tcp6_sock);
10484         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10485             sk->sk_family == AF_INET6)
10486                 return (unsigned long)sk;
10487
10488         return (unsigned long)NULL;
10489 }
10490
10491 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10492         .func                   = bpf_skc_to_tcp6_sock,
10493         .gpl_only               = false,
10494         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10495         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10496         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10497 };
10498
10499 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10500 {
10501         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10502                 return (unsigned long)sk;
10503
10504         return (unsigned long)NULL;
10505 }
10506
10507 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10508         .func                   = bpf_skc_to_tcp_sock,
10509         .gpl_only               = false,
10510         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10511         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10512         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10513 };
10514
10515 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10516 {
10517         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10518          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10519          */
10520         BTF_TYPE_EMIT(struct inet_timewait_sock);
10521         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10522
10523 #ifdef CONFIG_INET
10524         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10525                 return (unsigned long)sk;
10526 #endif
10527
10528 #if IS_BUILTIN(CONFIG_IPV6)
10529         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10530                 return (unsigned long)sk;
10531 #endif
10532
10533         return (unsigned long)NULL;
10534 }
10535
10536 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10537         .func                   = bpf_skc_to_tcp_timewait_sock,
10538         .gpl_only               = false,
10539         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10540         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10541         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10542 };
10543
10544 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10545 {
10546 #ifdef CONFIG_INET
10547         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10548                 return (unsigned long)sk;
10549 #endif
10550
10551 #if IS_BUILTIN(CONFIG_IPV6)
10552         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10553                 return (unsigned long)sk;
10554 #endif
10555
10556         return (unsigned long)NULL;
10557 }
10558
10559 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10560         .func                   = bpf_skc_to_tcp_request_sock,
10561         .gpl_only               = false,
10562         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10563         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10564         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10565 };
10566
10567 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10568 {
10569         /* udp6_sock type is not generated in dwarf and hence btf,
10570          * trigger an explicit type generation here.
10571          */
10572         BTF_TYPE_EMIT(struct udp6_sock);
10573         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10574             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10575                 return (unsigned long)sk;
10576
10577         return (unsigned long)NULL;
10578 }
10579
10580 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10581         .func                   = bpf_skc_to_udp6_sock,
10582         .gpl_only               = false,
10583         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10584         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10585         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10586 };
10587
10588 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10589 {
10590         return (unsigned long)sock_from_file(file);
10591 }
10592
10593 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10594 BTF_ID(struct, socket)
10595 BTF_ID(struct, file)
10596
10597 const struct bpf_func_proto bpf_sock_from_file_proto = {
10598         .func           = bpf_sock_from_file,
10599         .gpl_only       = false,
10600         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10601         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10602         .arg1_type      = ARG_PTR_TO_BTF_ID,
10603         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10604 };
10605
10606 static const struct bpf_func_proto *
10607 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10608 {
10609         const struct bpf_func_proto *func;
10610
10611         switch (func_id) {
10612         case BPF_FUNC_skc_to_tcp6_sock:
10613                 func = &bpf_skc_to_tcp6_sock_proto;
10614                 break;
10615         case BPF_FUNC_skc_to_tcp_sock:
10616                 func = &bpf_skc_to_tcp_sock_proto;
10617                 break;
10618         case BPF_FUNC_skc_to_tcp_timewait_sock:
10619                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10620                 break;
10621         case BPF_FUNC_skc_to_tcp_request_sock:
10622                 func = &bpf_skc_to_tcp_request_sock_proto;
10623                 break;
10624         case BPF_FUNC_skc_to_udp6_sock:
10625                 func = &bpf_skc_to_udp6_sock_proto;
10626                 break;
10627         default:
10628                 return bpf_base_func_proto(func_id);
10629         }
10630
10631         if (!perfmon_capable())
10632                 return NULL;
10633
10634         return func;
10635 }