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         return __bpf_try_make_writable(skb, write_len);
1867 }
1868
1869 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1870 {
1871         /* Idea is the following: should the needed direct read/write
1872          * test fail during runtime, we can pull in more data and redo
1873          * again, since implicitly, we invalidate previous checks here.
1874          *
1875          * Or, since we know how much we need to make read/writeable,
1876          * this can be done once at the program beginning for direct
1877          * access case. By this we overcome limitations of only current
1878          * headroom being accessible.
1879          */
1880         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1881 }
1882
1883 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1884         .func           = sk_skb_pull_data,
1885         .gpl_only       = false,
1886         .ret_type       = RET_INTEGER,
1887         .arg1_type      = ARG_PTR_TO_CTX,
1888         .arg2_type      = ARG_ANYTHING,
1889 };
1890
1891 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1892            u64, from, u64, to, u64, flags)
1893 {
1894         __sum16 *ptr;
1895
1896         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1897                 return -EINVAL;
1898         if (unlikely(offset > 0xffff || offset & 1))
1899                 return -EFAULT;
1900         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1901                 return -EFAULT;
1902
1903         ptr = (__sum16 *)(skb->data + offset);
1904         switch (flags & BPF_F_HDR_FIELD_MASK) {
1905         case 0:
1906                 if (unlikely(from != 0))
1907                         return -EINVAL;
1908
1909                 csum_replace_by_diff(ptr, to);
1910                 break;
1911         case 2:
1912                 csum_replace2(ptr, from, to);
1913                 break;
1914         case 4:
1915                 csum_replace4(ptr, from, to);
1916                 break;
1917         default:
1918                 return -EINVAL;
1919         }
1920
1921         return 0;
1922 }
1923
1924 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1925         .func           = bpf_l3_csum_replace,
1926         .gpl_only       = false,
1927         .ret_type       = RET_INTEGER,
1928         .arg1_type      = ARG_PTR_TO_CTX,
1929         .arg2_type      = ARG_ANYTHING,
1930         .arg3_type      = ARG_ANYTHING,
1931         .arg4_type      = ARG_ANYTHING,
1932         .arg5_type      = ARG_ANYTHING,
1933 };
1934
1935 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1936            u64, from, u64, to, u64, flags)
1937 {
1938         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1939         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1940         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1941         __sum16 *ptr;
1942
1943         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1944                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1945                 return -EINVAL;
1946         if (unlikely(offset > 0xffff || offset & 1))
1947                 return -EFAULT;
1948         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1949                 return -EFAULT;
1950
1951         ptr = (__sum16 *)(skb->data + offset);
1952         if (is_mmzero && !do_mforce && !*ptr)
1953                 return 0;
1954
1955         switch (flags & BPF_F_HDR_FIELD_MASK) {
1956         case 0:
1957                 if (unlikely(from != 0))
1958                         return -EINVAL;
1959
1960                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1961                 break;
1962         case 2:
1963                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1964                 break;
1965         case 4:
1966                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1967                 break;
1968         default:
1969                 return -EINVAL;
1970         }
1971
1972         if (is_mmzero && !*ptr)
1973                 *ptr = CSUM_MANGLED_0;
1974         return 0;
1975 }
1976
1977 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1978         .func           = bpf_l4_csum_replace,
1979         .gpl_only       = false,
1980         .ret_type       = RET_INTEGER,
1981         .arg1_type      = ARG_PTR_TO_CTX,
1982         .arg2_type      = ARG_ANYTHING,
1983         .arg3_type      = ARG_ANYTHING,
1984         .arg4_type      = ARG_ANYTHING,
1985         .arg5_type      = ARG_ANYTHING,
1986 };
1987
1988 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1989            __be32 *, to, u32, to_size, __wsum, seed)
1990 {
1991         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1992         u32 diff_size = from_size + to_size;
1993         int i, j = 0;
1994
1995         /* This is quite flexible, some examples:
1996          *
1997          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1998          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1999          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2000          *
2001          * Even for diffing, from_size and to_size don't need to be equal.
2002          */
2003         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2004                      diff_size > sizeof(sp->diff)))
2005                 return -EINVAL;
2006
2007         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2008                 sp->diff[j] = ~from[i];
2009         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2010                 sp->diff[j] = to[i];
2011
2012         return csum_partial(sp->diff, diff_size, seed);
2013 }
2014
2015 static const struct bpf_func_proto bpf_csum_diff_proto = {
2016         .func           = bpf_csum_diff,
2017         .gpl_only       = false,
2018         .pkt_access     = true,
2019         .ret_type       = RET_INTEGER,
2020         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
2021         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2022         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
2023         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2024         .arg5_type      = ARG_ANYTHING,
2025 };
2026
2027 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2028 {
2029         /* The interface is to be used in combination with bpf_csum_diff()
2030          * for direct packet writes. csum rotation for alignment as well
2031          * as emulating csum_sub() can be done from the eBPF program.
2032          */
2033         if (skb->ip_summed == CHECKSUM_COMPLETE)
2034                 return (skb->csum = csum_add(skb->csum, csum));
2035
2036         return -ENOTSUPP;
2037 }
2038
2039 static const struct bpf_func_proto bpf_csum_update_proto = {
2040         .func           = bpf_csum_update,
2041         .gpl_only       = false,
2042         .ret_type       = RET_INTEGER,
2043         .arg1_type      = ARG_PTR_TO_CTX,
2044         .arg2_type      = ARG_ANYTHING,
2045 };
2046
2047 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2048 {
2049         /* The interface is to be used in combination with bpf_skb_adjust_room()
2050          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2051          * is passed as flags, for example.
2052          */
2053         switch (level) {
2054         case BPF_CSUM_LEVEL_INC:
2055                 __skb_incr_checksum_unnecessary(skb);
2056                 break;
2057         case BPF_CSUM_LEVEL_DEC:
2058                 __skb_decr_checksum_unnecessary(skb);
2059                 break;
2060         case BPF_CSUM_LEVEL_RESET:
2061                 __skb_reset_checksum_unnecessary(skb);
2062                 break;
2063         case BPF_CSUM_LEVEL_QUERY:
2064                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2065                        skb->csum_level : -EACCES;
2066         default:
2067                 return -EINVAL;
2068         }
2069
2070         return 0;
2071 }
2072
2073 static const struct bpf_func_proto bpf_csum_level_proto = {
2074         .func           = bpf_csum_level,
2075         .gpl_only       = false,
2076         .ret_type       = RET_INTEGER,
2077         .arg1_type      = ARG_PTR_TO_CTX,
2078         .arg2_type      = ARG_ANYTHING,
2079 };
2080
2081 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2082 {
2083         return dev_forward_skb_nomtu(dev, skb);
2084 }
2085
2086 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2087                                       struct sk_buff *skb)
2088 {
2089         int ret = ____dev_forward_skb(dev, skb, false);
2090
2091         if (likely(!ret)) {
2092                 skb->dev = dev;
2093                 ret = netif_rx(skb);
2094         }
2095
2096         return ret;
2097 }
2098
2099 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2100 {
2101         int ret;
2102
2103         if (dev_xmit_recursion()) {
2104                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2105                 kfree_skb(skb);
2106                 return -ENETDOWN;
2107         }
2108
2109         skb->dev = dev;
2110         skb->tstamp = 0;
2111
2112         dev_xmit_recursion_inc();
2113         ret = dev_queue_xmit(skb);
2114         dev_xmit_recursion_dec();
2115
2116         return ret;
2117 }
2118
2119 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2120                                  u32 flags)
2121 {
2122         unsigned int mlen = skb_network_offset(skb);
2123
2124         if (mlen) {
2125                 __skb_pull(skb, mlen);
2126
2127                 /* At ingress, the mac header has already been pulled once.
2128                  * At egress, skb_pospull_rcsum has to be done in case that
2129                  * the skb is originated from ingress (i.e. a forwarded skb)
2130                  * to ensure that rcsum starts at net header.
2131                  */
2132                 if (!skb_at_tc_ingress(skb))
2133                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2134         }
2135         skb_pop_mac_header(skb);
2136         skb_reset_mac_len(skb);
2137         return flags & BPF_F_INGRESS ?
2138                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2139 }
2140
2141 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2142                                  u32 flags)
2143 {
2144         /* Verify that a link layer header is carried */
2145         if (unlikely(skb->mac_header >= skb->network_header)) {
2146                 kfree_skb(skb);
2147                 return -ERANGE;
2148         }
2149
2150         bpf_push_mac_rcsum(skb);
2151         return flags & BPF_F_INGRESS ?
2152                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2153 }
2154
2155 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2156                           u32 flags)
2157 {
2158         if (dev_is_mac_header_xmit(dev))
2159                 return __bpf_redirect_common(skb, dev, flags);
2160         else
2161                 return __bpf_redirect_no_mac(skb, dev, flags);
2162 }
2163
2164 #if IS_ENABLED(CONFIG_IPV6)
2165 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2166                             struct net_device *dev, struct bpf_nh_params *nh)
2167 {
2168         u32 hh_len = LL_RESERVED_SPACE(dev);
2169         const struct in6_addr *nexthop;
2170         struct dst_entry *dst = NULL;
2171         struct neighbour *neigh;
2172
2173         if (dev_xmit_recursion()) {
2174                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2175                 goto out_drop;
2176         }
2177
2178         skb->dev = dev;
2179         skb->tstamp = 0;
2180
2181         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2182                 struct sk_buff *skb2;
2183
2184                 skb2 = skb_realloc_headroom(skb, hh_len);
2185                 if (unlikely(!skb2)) {
2186                         kfree_skb(skb);
2187                         return -ENOMEM;
2188                 }
2189                 if (skb->sk)
2190                         skb_set_owner_w(skb2, skb->sk);
2191                 consume_skb(skb);
2192                 skb = skb2;
2193         }
2194
2195         rcu_read_lock_bh();
2196         if (!nh) {
2197                 dst = skb_dst(skb);
2198                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2199                                       &ipv6_hdr(skb)->daddr);
2200         } else {
2201                 nexthop = &nh->ipv6_nh;
2202         }
2203         neigh = ip_neigh_gw6(dev, nexthop);
2204         if (likely(!IS_ERR(neigh))) {
2205                 int ret;
2206
2207                 sock_confirm_neigh(skb, neigh);
2208                 dev_xmit_recursion_inc();
2209                 ret = neigh_output(neigh, skb, false);
2210                 dev_xmit_recursion_dec();
2211                 rcu_read_unlock_bh();
2212                 return ret;
2213         }
2214         rcu_read_unlock_bh();
2215         if (dst)
2216                 IP6_INC_STATS(dev_net(dst->dev),
2217                               ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2218 out_drop:
2219         kfree_skb(skb);
2220         return -ENETDOWN;
2221 }
2222
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224                                    struct bpf_nh_params *nh)
2225 {
2226         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227         struct net *net = dev_net(dev);
2228         int err, ret = NET_XMIT_DROP;
2229
2230         if (!nh) {
2231                 struct dst_entry *dst;
2232                 struct flowi6 fl6 = {
2233                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234                         .flowi6_mark  = skb->mark,
2235                         .flowlabel    = ip6_flowinfo(ip6h),
2236                         .flowi6_oif   = dev->ifindex,
2237                         .flowi6_proto = ip6h->nexthdr,
2238                         .daddr        = ip6h->daddr,
2239                         .saddr        = ip6h->saddr,
2240                 };
2241
2242                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2243                 if (IS_ERR(dst))
2244                         goto out_drop;
2245
2246                 skb_dst_set(skb, dst);
2247         } else if (nh->nh_family != AF_INET6) {
2248                 goto out_drop;
2249         }
2250
2251         err = bpf_out_neigh_v6(net, skb, dev, nh);
2252         if (unlikely(net_xmit_eval(err)))
2253                 dev->stats.tx_errors++;
2254         else
2255                 ret = NET_XMIT_SUCCESS;
2256         goto out_xmit;
2257 out_drop:
2258         dev->stats.tx_errors++;
2259         kfree_skb(skb);
2260 out_xmit:
2261         return ret;
2262 }
2263 #else
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265                                    struct bpf_nh_params *nh)
2266 {
2267         kfree_skb(skb);
2268         return NET_XMIT_DROP;
2269 }
2270 #endif /* CONFIG_IPV6 */
2271
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274                             struct net_device *dev, struct bpf_nh_params *nh)
2275 {
2276         u32 hh_len = LL_RESERVED_SPACE(dev);
2277         struct neighbour *neigh;
2278         bool is_v6gw = false;
2279
2280         if (dev_xmit_recursion()) {
2281                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2282                 goto out_drop;
2283         }
2284
2285         skb->dev = dev;
2286         skb->tstamp = 0;
2287
2288         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289                 struct sk_buff *skb2;
2290
2291                 skb2 = skb_realloc_headroom(skb, hh_len);
2292                 if (unlikely(!skb2)) {
2293                         kfree_skb(skb);
2294                         return -ENOMEM;
2295                 }
2296                 if (skb->sk)
2297                         skb_set_owner_w(skb2, skb->sk);
2298                 consume_skb(skb);
2299                 skb = skb2;
2300         }
2301
2302         rcu_read_lock_bh();
2303         if (!nh) {
2304                 struct dst_entry *dst = skb_dst(skb);
2305                 struct rtable *rt = container_of(dst, struct rtable, dst);
2306
2307                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2308         } else if (nh->nh_family == AF_INET6) {
2309                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2310                 is_v6gw = true;
2311         } else if (nh->nh_family == AF_INET) {
2312                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2313         } else {
2314                 rcu_read_unlock_bh();
2315                 goto out_drop;
2316         }
2317
2318         if (likely(!IS_ERR(neigh))) {
2319                 int ret;
2320
2321                 sock_confirm_neigh(skb, neigh);
2322                 dev_xmit_recursion_inc();
2323                 ret = neigh_output(neigh, skb, is_v6gw);
2324                 dev_xmit_recursion_dec();
2325                 rcu_read_unlock_bh();
2326                 return ret;
2327         }
2328         rcu_read_unlock_bh();
2329 out_drop:
2330         kfree_skb(skb);
2331         return -ENETDOWN;
2332 }
2333
2334 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2335                                    struct bpf_nh_params *nh)
2336 {
2337         const struct iphdr *ip4h = ip_hdr(skb);
2338         struct net *net = dev_net(dev);
2339         int err, ret = NET_XMIT_DROP;
2340
2341         if (!nh) {
2342                 struct flowi4 fl4 = {
2343                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2344                         .flowi4_mark  = skb->mark,
2345                         .flowi4_tos   = RT_TOS(ip4h->tos),
2346                         .flowi4_oif   = dev->ifindex,
2347                         .flowi4_proto = ip4h->protocol,
2348                         .daddr        = ip4h->daddr,
2349                         .saddr        = ip4h->saddr,
2350                 };
2351                 struct rtable *rt;
2352
2353                 rt = ip_route_output_flow(net, &fl4, NULL);
2354                 if (IS_ERR(rt))
2355                         goto out_drop;
2356                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2357                         ip_rt_put(rt);
2358                         goto out_drop;
2359                 }
2360
2361                 skb_dst_set(skb, &rt->dst);
2362         }
2363
2364         err = bpf_out_neigh_v4(net, skb, dev, nh);
2365         if (unlikely(net_xmit_eval(err)))
2366                 dev->stats.tx_errors++;
2367         else
2368                 ret = NET_XMIT_SUCCESS;
2369         goto out_xmit;
2370 out_drop:
2371         dev->stats.tx_errors++;
2372         kfree_skb(skb);
2373 out_xmit:
2374         return ret;
2375 }
2376 #else
2377 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2378                                    struct bpf_nh_params *nh)
2379 {
2380         kfree_skb(skb);
2381         return NET_XMIT_DROP;
2382 }
2383 #endif /* CONFIG_INET */
2384
2385 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2386                                 struct bpf_nh_params *nh)
2387 {
2388         struct ethhdr *ethh = eth_hdr(skb);
2389
2390         if (unlikely(skb->mac_header >= skb->network_header))
2391                 goto out;
2392         bpf_push_mac_rcsum(skb);
2393         if (is_multicast_ether_addr(ethh->h_dest))
2394                 goto out;
2395
2396         skb_pull(skb, sizeof(*ethh));
2397         skb_unset_mac_header(skb);
2398         skb_reset_network_header(skb);
2399
2400         if (skb->protocol == htons(ETH_P_IP))
2401                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2402         else if (skb->protocol == htons(ETH_P_IPV6))
2403                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2404 out:
2405         kfree_skb(skb);
2406         return -ENOTSUPP;
2407 }
2408
2409 /* Internal, non-exposed redirect flags. */
2410 enum {
2411         BPF_F_NEIGH     = (1ULL << 1),
2412         BPF_F_PEER      = (1ULL << 2),
2413         BPF_F_NEXTHOP   = (1ULL << 3),
2414 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2415 };
2416
2417 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2418 {
2419         struct net_device *dev;
2420         struct sk_buff *clone;
2421         int ret;
2422
2423         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2424                 return -EINVAL;
2425
2426         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2427         if (unlikely(!dev))
2428                 return -EINVAL;
2429
2430         clone = skb_clone(skb, GFP_ATOMIC);
2431         if (unlikely(!clone))
2432                 return -ENOMEM;
2433
2434         /* For direct write, we need to keep the invariant that the skbs
2435          * we're dealing with need to be uncloned. Should uncloning fail
2436          * here, we need to free the just generated clone to unclone once
2437          * again.
2438          */
2439         ret = bpf_try_make_head_writable(skb);
2440         if (unlikely(ret)) {
2441                 kfree_skb(clone);
2442                 return -ENOMEM;
2443         }
2444
2445         return __bpf_redirect(clone, dev, flags);
2446 }
2447
2448 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2449         .func           = bpf_clone_redirect,
2450         .gpl_only       = false,
2451         .ret_type       = RET_INTEGER,
2452         .arg1_type      = ARG_PTR_TO_CTX,
2453         .arg2_type      = ARG_ANYTHING,
2454         .arg3_type      = ARG_ANYTHING,
2455 };
2456
2457 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2458 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2459
2460 int skb_do_redirect(struct sk_buff *skb)
2461 {
2462         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2463         struct net *net = dev_net(skb->dev);
2464         struct net_device *dev;
2465         u32 flags = ri->flags;
2466
2467         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2468         ri->tgt_index = 0;
2469         ri->flags = 0;
2470         if (unlikely(!dev))
2471                 goto out_drop;
2472         if (flags & BPF_F_PEER) {
2473                 const struct net_device_ops *ops = dev->netdev_ops;
2474
2475                 if (unlikely(!ops->ndo_get_peer_dev ||
2476                              !skb_at_tc_ingress(skb)))
2477                         goto out_drop;
2478                 dev = ops->ndo_get_peer_dev(dev);
2479                 if (unlikely(!dev ||
2480                              !(dev->flags & IFF_UP) ||
2481                              net_eq(net, dev_net(dev))))
2482                         goto out_drop;
2483                 skb->dev = dev;
2484                 return -EAGAIN;
2485         }
2486         return flags & BPF_F_NEIGH ?
2487                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2488                                     &ri->nh : NULL) :
2489                __bpf_redirect(skb, dev, flags);
2490 out_drop:
2491         kfree_skb(skb);
2492         return -EINVAL;
2493 }
2494
2495 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2496 {
2497         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2498
2499         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2500                 return TC_ACT_SHOT;
2501
2502         ri->flags = flags;
2503         ri->tgt_index = ifindex;
2504
2505         return TC_ACT_REDIRECT;
2506 }
2507
2508 static const struct bpf_func_proto bpf_redirect_proto = {
2509         .func           = bpf_redirect,
2510         .gpl_only       = false,
2511         .ret_type       = RET_INTEGER,
2512         .arg1_type      = ARG_ANYTHING,
2513         .arg2_type      = ARG_ANYTHING,
2514 };
2515
2516 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2517 {
2518         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2519
2520         if (unlikely(flags))
2521                 return TC_ACT_SHOT;
2522
2523         ri->flags = BPF_F_PEER;
2524         ri->tgt_index = ifindex;
2525
2526         return TC_ACT_REDIRECT;
2527 }
2528
2529 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2530         .func           = bpf_redirect_peer,
2531         .gpl_only       = false,
2532         .ret_type       = RET_INTEGER,
2533         .arg1_type      = ARG_ANYTHING,
2534         .arg2_type      = ARG_ANYTHING,
2535 };
2536
2537 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2538            int, plen, u64, flags)
2539 {
2540         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2541
2542         if (unlikely((plen && plen < sizeof(*params)) || flags))
2543                 return TC_ACT_SHOT;
2544
2545         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2546         ri->tgt_index = ifindex;
2547
2548         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2549         if (plen)
2550                 memcpy(&ri->nh, params, sizeof(ri->nh));
2551
2552         return TC_ACT_REDIRECT;
2553 }
2554
2555 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2556         .func           = bpf_redirect_neigh,
2557         .gpl_only       = false,
2558         .ret_type       = RET_INTEGER,
2559         .arg1_type      = ARG_ANYTHING,
2560         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
2561         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2562         .arg4_type      = ARG_ANYTHING,
2563 };
2564
2565 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2566 {
2567         msg->apply_bytes = bytes;
2568         return 0;
2569 }
2570
2571 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2572         .func           = bpf_msg_apply_bytes,
2573         .gpl_only       = false,
2574         .ret_type       = RET_INTEGER,
2575         .arg1_type      = ARG_PTR_TO_CTX,
2576         .arg2_type      = ARG_ANYTHING,
2577 };
2578
2579 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2580 {
2581         msg->cork_bytes = bytes;
2582         return 0;
2583 }
2584
2585 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2586         .func           = bpf_msg_cork_bytes,
2587         .gpl_only       = false,
2588         .ret_type       = RET_INTEGER,
2589         .arg1_type      = ARG_PTR_TO_CTX,
2590         .arg2_type      = ARG_ANYTHING,
2591 };
2592
2593 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2594            u32, end, u64, flags)
2595 {
2596         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2597         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2598         struct scatterlist *sge;
2599         u8 *raw, *to, *from;
2600         struct page *page;
2601
2602         if (unlikely(flags || end <= start))
2603                 return -EINVAL;
2604
2605         /* First find the starting scatterlist element */
2606         i = msg->sg.start;
2607         do {
2608                 offset += len;
2609                 len = sk_msg_elem(msg, i)->length;
2610                 if (start < offset + len)
2611                         break;
2612                 sk_msg_iter_var_next(i);
2613         } while (i != msg->sg.end);
2614
2615         if (unlikely(start >= offset + len))
2616                 return -EINVAL;
2617
2618         first_sge = i;
2619         /* The start may point into the sg element so we need to also
2620          * account for the headroom.
2621          */
2622         bytes_sg_total = start - offset + bytes;
2623         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2624                 goto out;
2625
2626         /* At this point we need to linearize multiple scatterlist
2627          * elements or a single shared page. Either way we need to
2628          * copy into a linear buffer exclusively owned by BPF. Then
2629          * place the buffer in the scatterlist and fixup the original
2630          * entries by removing the entries now in the linear buffer
2631          * and shifting the remaining entries. For now we do not try
2632          * to copy partial entries to avoid complexity of running out
2633          * of sg_entry slots. The downside is reading a single byte
2634          * will copy the entire sg entry.
2635          */
2636         do {
2637                 copy += sk_msg_elem(msg, i)->length;
2638                 sk_msg_iter_var_next(i);
2639                 if (bytes_sg_total <= copy)
2640                         break;
2641         } while (i != msg->sg.end);
2642         last_sge = i;
2643
2644         if (unlikely(bytes_sg_total > copy))
2645                 return -EINVAL;
2646
2647         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2648                            get_order(copy));
2649         if (unlikely(!page))
2650                 return -ENOMEM;
2651
2652         raw = page_address(page);
2653         i = first_sge;
2654         do {
2655                 sge = sk_msg_elem(msg, i);
2656                 from = sg_virt(sge);
2657                 len = sge->length;
2658                 to = raw + poffset;
2659
2660                 memcpy(to, from, len);
2661                 poffset += len;
2662                 sge->length = 0;
2663                 put_page(sg_page(sge));
2664
2665                 sk_msg_iter_var_next(i);
2666         } while (i != last_sge);
2667
2668         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2669
2670         /* To repair sg ring we need to shift entries. If we only
2671          * had a single entry though we can just replace it and
2672          * be done. Otherwise walk the ring and shift the entries.
2673          */
2674         WARN_ON_ONCE(last_sge == first_sge);
2675         shift = last_sge > first_sge ?
2676                 last_sge - first_sge - 1 :
2677                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2678         if (!shift)
2679                 goto out;
2680
2681         i = first_sge;
2682         sk_msg_iter_var_next(i);
2683         do {
2684                 u32 move_from;
2685
2686                 if (i + shift >= NR_MSG_FRAG_IDS)
2687                         move_from = i + shift - NR_MSG_FRAG_IDS;
2688                 else
2689                         move_from = i + shift;
2690                 if (move_from == msg->sg.end)
2691                         break;
2692
2693                 msg->sg.data[i] = msg->sg.data[move_from];
2694                 msg->sg.data[move_from].length = 0;
2695                 msg->sg.data[move_from].page_link = 0;
2696                 msg->sg.data[move_from].offset = 0;
2697                 sk_msg_iter_var_next(i);
2698         } while (1);
2699
2700         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2701                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2702                       msg->sg.end - shift;
2703 out:
2704         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2705         msg->data_end = msg->data + bytes;
2706         return 0;
2707 }
2708
2709 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2710         .func           = bpf_msg_pull_data,
2711         .gpl_only       = false,
2712         .ret_type       = RET_INTEGER,
2713         .arg1_type      = ARG_PTR_TO_CTX,
2714         .arg2_type      = ARG_ANYTHING,
2715         .arg3_type      = ARG_ANYTHING,
2716         .arg4_type      = ARG_ANYTHING,
2717 };
2718
2719 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2720            u32, len, u64, flags)
2721 {
2722         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2723         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2724         u8 *raw, *to, *from;
2725         struct page *page;
2726
2727         if (unlikely(flags))
2728                 return -EINVAL;
2729
2730         /* First find the starting scatterlist element */
2731         i = msg->sg.start;
2732         do {
2733                 offset += l;
2734                 l = sk_msg_elem(msg, i)->length;
2735
2736                 if (start < offset + l)
2737                         break;
2738                 sk_msg_iter_var_next(i);
2739         } while (i != msg->sg.end);
2740
2741         if (start >= offset + l)
2742                 return -EINVAL;
2743
2744         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2745
2746         /* If no space available will fallback to copy, we need at
2747          * least one scatterlist elem available to push data into
2748          * when start aligns to the beginning of an element or two
2749          * when it falls inside an element. We handle the start equals
2750          * offset case because its the common case for inserting a
2751          * header.
2752          */
2753         if (!space || (space == 1 && start != offset))
2754                 copy = msg->sg.data[i].length;
2755
2756         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2757                            get_order(copy + len));
2758         if (unlikely(!page))
2759                 return -ENOMEM;
2760
2761         if (copy) {
2762                 int front, back;
2763
2764                 raw = page_address(page);
2765
2766                 psge = sk_msg_elem(msg, i);
2767                 front = start - offset;
2768                 back = psge->length - front;
2769                 from = sg_virt(psge);
2770
2771                 if (front)
2772                         memcpy(raw, from, front);
2773
2774                 if (back) {
2775                         from += front;
2776                         to = raw + front + len;
2777
2778                         memcpy(to, from, back);
2779                 }
2780
2781                 put_page(sg_page(psge));
2782         } else if (start - offset) {
2783                 psge = sk_msg_elem(msg, i);
2784                 rsge = sk_msg_elem_cpy(msg, i);
2785
2786                 psge->length = start - offset;
2787                 rsge.length -= psge->length;
2788                 rsge.offset += start;
2789
2790                 sk_msg_iter_var_next(i);
2791                 sg_unmark_end(psge);
2792                 sg_unmark_end(&rsge);
2793                 sk_msg_iter_next(msg, end);
2794         }
2795
2796         /* Slot(s) to place newly allocated data */
2797         new = i;
2798
2799         /* Shift one or two slots as needed */
2800         if (!copy) {
2801                 sge = sk_msg_elem_cpy(msg, i);
2802
2803                 sk_msg_iter_var_next(i);
2804                 sg_unmark_end(&sge);
2805                 sk_msg_iter_next(msg, end);
2806
2807                 nsge = sk_msg_elem_cpy(msg, i);
2808                 if (rsge.length) {
2809                         sk_msg_iter_var_next(i);
2810                         nnsge = sk_msg_elem_cpy(msg, i);
2811                 }
2812
2813                 while (i != msg->sg.end) {
2814                         msg->sg.data[i] = sge;
2815                         sge = nsge;
2816                         sk_msg_iter_var_next(i);
2817                         if (rsge.length) {
2818                                 nsge = nnsge;
2819                                 nnsge = sk_msg_elem_cpy(msg, i);
2820                         } else {
2821                                 nsge = sk_msg_elem_cpy(msg, i);
2822                         }
2823                 }
2824         }
2825
2826         /* Place newly allocated data buffer */
2827         sk_mem_charge(msg->sk, len);
2828         msg->sg.size += len;
2829         __clear_bit(new, &msg->sg.copy);
2830         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2831         if (rsge.length) {
2832                 get_page(sg_page(&rsge));
2833                 sk_msg_iter_var_next(new);
2834                 msg->sg.data[new] = rsge;
2835         }
2836
2837         sk_msg_compute_data_pointers(msg);
2838         return 0;
2839 }
2840
2841 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2842         .func           = bpf_msg_push_data,
2843         .gpl_only       = false,
2844         .ret_type       = RET_INTEGER,
2845         .arg1_type      = ARG_PTR_TO_CTX,
2846         .arg2_type      = ARG_ANYTHING,
2847         .arg3_type      = ARG_ANYTHING,
2848         .arg4_type      = ARG_ANYTHING,
2849 };
2850
2851 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2852 {
2853         int prev;
2854
2855         do {
2856                 prev = i;
2857                 sk_msg_iter_var_next(i);
2858                 msg->sg.data[prev] = msg->sg.data[i];
2859         } while (i != msg->sg.end);
2860
2861         sk_msg_iter_prev(msg, end);
2862 }
2863
2864 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2865 {
2866         struct scatterlist tmp, sge;
2867
2868         sk_msg_iter_next(msg, end);
2869         sge = sk_msg_elem_cpy(msg, i);
2870         sk_msg_iter_var_next(i);
2871         tmp = sk_msg_elem_cpy(msg, i);
2872
2873         while (i != msg->sg.end) {
2874                 msg->sg.data[i] = sge;
2875                 sk_msg_iter_var_next(i);
2876                 sge = tmp;
2877                 tmp = sk_msg_elem_cpy(msg, i);
2878         }
2879 }
2880
2881 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2882            u32, len, u64, flags)
2883 {
2884         u32 i = 0, l = 0, space, offset = 0;
2885         u64 last = start + len;
2886         int pop;
2887
2888         if (unlikely(flags))
2889                 return -EINVAL;
2890
2891         /* First find the starting scatterlist element */
2892         i = msg->sg.start;
2893         do {
2894                 offset += l;
2895                 l = sk_msg_elem(msg, i)->length;
2896
2897                 if (start < offset + l)
2898                         break;
2899                 sk_msg_iter_var_next(i);
2900         } while (i != msg->sg.end);
2901
2902         /* Bounds checks: start and pop must be inside message */
2903         if (start >= offset + l || last >= msg->sg.size)
2904                 return -EINVAL;
2905
2906         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2907
2908         pop = len;
2909         /* --------------| offset
2910          * -| start      |-------- len -------|
2911          *
2912          *  |----- a ----|-------- pop -------|----- b ----|
2913          *  |______________________________________________| length
2914          *
2915          *
2916          * a:   region at front of scatter element to save
2917          * b:   region at back of scatter element to save when length > A + pop
2918          * pop: region to pop from element, same as input 'pop' here will be
2919          *      decremented below per iteration.
2920          *
2921          * Two top-level cases to handle when start != offset, first B is non
2922          * zero and second B is zero corresponding to when a pop includes more
2923          * than one element.
2924          *
2925          * Then if B is non-zero AND there is no space allocate space and
2926          * compact A, B regions into page. If there is space shift ring to
2927          * the rigth free'ing the next element in ring to place B, leaving
2928          * A untouched except to reduce length.
2929          */
2930         if (start != offset) {
2931                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2932                 int a = start;
2933                 int b = sge->length - pop - a;
2934
2935                 sk_msg_iter_var_next(i);
2936
2937                 if (pop < sge->length - a) {
2938                         if (space) {
2939                                 sge->length = a;
2940                                 sk_msg_shift_right(msg, i);
2941                                 nsge = sk_msg_elem(msg, i);
2942                                 get_page(sg_page(sge));
2943                                 sg_set_page(nsge,
2944                                             sg_page(sge),
2945                                             b, sge->offset + pop + a);
2946                         } else {
2947                                 struct page *page, *orig;
2948                                 u8 *to, *from;
2949
2950                                 page = alloc_pages(__GFP_NOWARN |
2951                                                    __GFP_COMP   | GFP_ATOMIC,
2952                                                    get_order(a + b));
2953                                 if (unlikely(!page))
2954                                         return -ENOMEM;
2955
2956                                 sge->length = a;
2957                                 orig = sg_page(sge);
2958                                 from = sg_virt(sge);
2959                                 to = page_address(page);
2960                                 memcpy(to, from, a);
2961                                 memcpy(to + a, from + a + pop, b);
2962                                 sg_set_page(sge, page, a + b, 0);
2963                                 put_page(orig);
2964                         }
2965                         pop = 0;
2966                 } else if (pop >= sge->length - a) {
2967                         pop -= (sge->length - a);
2968                         sge->length = a;
2969                 }
2970         }
2971
2972         /* From above the current layout _must_ be as follows,
2973          *
2974          * -| offset
2975          * -| start
2976          *
2977          *  |---- pop ---|---------------- b ------------|
2978          *  |____________________________________________| length
2979          *
2980          * Offset and start of the current msg elem are equal because in the
2981          * previous case we handled offset != start and either consumed the
2982          * entire element and advanced to the next element OR pop == 0.
2983          *
2984          * Two cases to handle here are first pop is less than the length
2985          * leaving some remainder b above. Simply adjust the element's layout
2986          * in this case. Or pop >= length of the element so that b = 0. In this
2987          * case advance to next element decrementing pop.
2988          */
2989         while (pop) {
2990                 struct scatterlist *sge = sk_msg_elem(msg, i);
2991
2992                 if (pop < sge->length) {
2993                         sge->length -= pop;
2994                         sge->offset += pop;
2995                         pop = 0;
2996                 } else {
2997                         pop -= sge->length;
2998                         sk_msg_shift_left(msg, i);
2999                 }
3000                 sk_msg_iter_var_next(i);
3001         }
3002
3003         sk_mem_uncharge(msg->sk, len - pop);
3004         msg->sg.size -= (len - pop);
3005         sk_msg_compute_data_pointers(msg);
3006         return 0;
3007 }
3008
3009 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3010         .func           = bpf_msg_pop_data,
3011         .gpl_only       = false,
3012         .ret_type       = RET_INTEGER,
3013         .arg1_type      = ARG_PTR_TO_CTX,
3014         .arg2_type      = ARG_ANYTHING,
3015         .arg3_type      = ARG_ANYTHING,
3016         .arg4_type      = ARG_ANYTHING,
3017 };
3018
3019 #ifdef CONFIG_CGROUP_NET_CLASSID
3020 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3021 {
3022         return __task_get_classid(current);
3023 }
3024
3025 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3026         .func           = bpf_get_cgroup_classid_curr,
3027         .gpl_only       = false,
3028         .ret_type       = RET_INTEGER,
3029 };
3030
3031 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3032 {
3033         struct sock *sk = skb_to_full_sk(skb);
3034
3035         if (!sk || !sk_fullsock(sk))
3036                 return 0;
3037
3038         return sock_cgroup_classid(&sk->sk_cgrp_data);
3039 }
3040
3041 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3042         .func           = bpf_skb_cgroup_classid,
3043         .gpl_only       = false,
3044         .ret_type       = RET_INTEGER,
3045         .arg1_type      = ARG_PTR_TO_CTX,
3046 };
3047 #endif
3048
3049 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3050 {
3051         return task_get_classid(skb);
3052 }
3053
3054 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3055         .func           = bpf_get_cgroup_classid,
3056         .gpl_only       = false,
3057         .ret_type       = RET_INTEGER,
3058         .arg1_type      = ARG_PTR_TO_CTX,
3059 };
3060
3061 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3062 {
3063         return dst_tclassid(skb);
3064 }
3065
3066 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3067         .func           = bpf_get_route_realm,
3068         .gpl_only       = false,
3069         .ret_type       = RET_INTEGER,
3070         .arg1_type      = ARG_PTR_TO_CTX,
3071 };
3072
3073 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3074 {
3075         /* If skb_clear_hash() was called due to mangling, we can
3076          * trigger SW recalculation here. Later access to hash
3077          * can then use the inline skb->hash via context directly
3078          * instead of calling this helper again.
3079          */
3080         return skb_get_hash(skb);
3081 }
3082
3083 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3084         .func           = bpf_get_hash_recalc,
3085         .gpl_only       = false,
3086         .ret_type       = RET_INTEGER,
3087         .arg1_type      = ARG_PTR_TO_CTX,
3088 };
3089
3090 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3091 {
3092         /* After all direct packet write, this can be used once for
3093          * triggering a lazy recalc on next skb_get_hash() invocation.
3094          */
3095         skb_clear_hash(skb);
3096         return 0;
3097 }
3098
3099 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3100         .func           = bpf_set_hash_invalid,
3101         .gpl_only       = false,
3102         .ret_type       = RET_INTEGER,
3103         .arg1_type      = ARG_PTR_TO_CTX,
3104 };
3105
3106 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3107 {
3108         /* Set user specified hash as L4(+), so that it gets returned
3109          * on skb_get_hash() call unless BPF prog later on triggers a
3110          * skb_clear_hash().
3111          */
3112         __skb_set_sw_hash(skb, hash, true);
3113         return 0;
3114 }
3115
3116 static const struct bpf_func_proto bpf_set_hash_proto = {
3117         .func           = bpf_set_hash,
3118         .gpl_only       = false,
3119         .ret_type       = RET_INTEGER,
3120         .arg1_type      = ARG_PTR_TO_CTX,
3121         .arg2_type      = ARG_ANYTHING,
3122 };
3123
3124 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3125            u16, vlan_tci)
3126 {
3127         int ret;
3128
3129         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3130                      vlan_proto != htons(ETH_P_8021AD)))
3131                 vlan_proto = htons(ETH_P_8021Q);
3132
3133         bpf_push_mac_rcsum(skb);
3134         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3135         bpf_pull_mac_rcsum(skb);
3136
3137         bpf_compute_data_pointers(skb);
3138         return ret;
3139 }
3140
3141 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3142         .func           = bpf_skb_vlan_push,
3143         .gpl_only       = false,
3144         .ret_type       = RET_INTEGER,
3145         .arg1_type      = ARG_PTR_TO_CTX,
3146         .arg2_type      = ARG_ANYTHING,
3147         .arg3_type      = ARG_ANYTHING,
3148 };
3149
3150 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3151 {
3152         int ret;
3153
3154         bpf_push_mac_rcsum(skb);
3155         ret = skb_vlan_pop(skb);
3156         bpf_pull_mac_rcsum(skb);
3157
3158         bpf_compute_data_pointers(skb);
3159         return ret;
3160 }
3161
3162 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3163         .func           = bpf_skb_vlan_pop,
3164         .gpl_only       = false,
3165         .ret_type       = RET_INTEGER,
3166         .arg1_type      = ARG_PTR_TO_CTX,
3167 };
3168
3169 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3170 {
3171         /* Caller already did skb_cow() with len as headroom,
3172          * so no need to do it here.
3173          */
3174         skb_push(skb, len);
3175         memmove(skb->data, skb->data + len, off);
3176         memset(skb->data + off, 0, len);
3177
3178         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3179          * needed here as it does not change the skb->csum
3180          * result for checksum complete when summing over
3181          * zeroed blocks.
3182          */
3183         return 0;
3184 }
3185
3186 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3187 {
3188         /* skb_ensure_writable() is not needed here, as we're
3189          * already working on an uncloned skb.
3190          */
3191         if (unlikely(!pskb_may_pull(skb, off + len)))
3192                 return -ENOMEM;
3193
3194         skb_postpull_rcsum(skb, skb->data + off, len);
3195         memmove(skb->data + len, skb->data, off);
3196         __skb_pull(skb, len);
3197
3198         return 0;
3199 }
3200
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3202 {
3203         bool trans_same = skb->transport_header == skb->network_header;
3204         int ret;
3205
3206         /* There's no need for __skb_push()/__skb_pull() pair to
3207          * get to the start of the mac header as we're guaranteed
3208          * to always start from here under eBPF.
3209          */
3210         ret = bpf_skb_generic_push(skb, off, len);
3211         if (likely(!ret)) {
3212                 skb->mac_header -= len;
3213                 skb->network_header -= len;
3214                 if (trans_same)
3215                         skb->transport_header = skb->network_header;
3216         }
3217
3218         return ret;
3219 }
3220
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3222 {
3223         bool trans_same = skb->transport_header == skb->network_header;
3224         int ret;
3225
3226         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227         ret = bpf_skb_generic_pop(skb, off, len);
3228         if (likely(!ret)) {
3229                 skb->mac_header += len;
3230                 skb->network_header += len;
3231                 if (trans_same)
3232                         skb->transport_header = skb->network_header;
3233         }
3234
3235         return ret;
3236 }
3237
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3239 {
3240         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241         u32 off = skb_mac_header_len(skb);
3242         int ret;
3243
3244         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3245                 return -ENOTSUPP;
3246
3247         ret = skb_cow(skb, len_diff);
3248         if (unlikely(ret < 0))
3249                 return ret;
3250
3251         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3252         if (unlikely(ret < 0))
3253                 return ret;
3254
3255         if (skb_is_gso(skb)) {
3256                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3257
3258                 /* SKB_GSO_TCPV4 needs to be changed into
3259                  * SKB_GSO_TCPV6.
3260                  */
3261                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3262                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3263                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3264                 }
3265
3266                 /* Due to IPv6 header, MSS needs to be downgraded. */
3267                 skb_decrease_gso_size(shinfo, len_diff);
3268                 /* Header must be checked, and gso_segs recomputed. */
3269                 shinfo->gso_type |= SKB_GSO_DODGY;
3270                 shinfo->gso_segs = 0;
3271         }
3272
3273         skb->protocol = htons(ETH_P_IPV6);
3274         skb_clear_hash(skb);
3275
3276         return 0;
3277 }
3278
3279 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3280 {
3281         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3282         u32 off = skb_mac_header_len(skb);
3283         int ret;
3284
3285         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3286                 return -ENOTSUPP;
3287
3288         ret = skb_unclone(skb, GFP_ATOMIC);
3289         if (unlikely(ret < 0))
3290                 return ret;
3291
3292         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3293         if (unlikely(ret < 0))
3294                 return ret;
3295
3296         if (skb_is_gso(skb)) {
3297                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3298
3299                 /* SKB_GSO_TCPV6 needs to be changed into
3300                  * SKB_GSO_TCPV4.
3301                  */
3302                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3303                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3304                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3305                 }
3306
3307                 /* Due to IPv4 header, MSS can be upgraded. */
3308                 skb_increase_gso_size(shinfo, len_diff);
3309                 /* Header must be checked, and gso_segs recomputed. */
3310                 shinfo->gso_type |= SKB_GSO_DODGY;
3311                 shinfo->gso_segs = 0;
3312         }
3313
3314         skb->protocol = htons(ETH_P_IP);
3315         skb_clear_hash(skb);
3316
3317         return 0;
3318 }
3319
3320 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3321 {
3322         __be16 from_proto = skb->protocol;
3323
3324         if (from_proto == htons(ETH_P_IP) &&
3325               to_proto == htons(ETH_P_IPV6))
3326                 return bpf_skb_proto_4_to_6(skb);
3327
3328         if (from_proto == htons(ETH_P_IPV6) &&
3329               to_proto == htons(ETH_P_IP))
3330                 return bpf_skb_proto_6_to_4(skb);
3331
3332         return -ENOTSUPP;
3333 }
3334
3335 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3336            u64, flags)
3337 {
3338         int ret;
3339
3340         if (unlikely(flags))
3341                 return -EINVAL;
3342
3343         /* General idea is that this helper does the basic groundwork
3344          * needed for changing the protocol, and eBPF program fills the
3345          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3346          * and other helpers, rather than passing a raw buffer here.
3347          *
3348          * The rationale is to keep this minimal and without a need to
3349          * deal with raw packet data. F.e. even if we would pass buffers
3350          * here, the program still needs to call the bpf_lX_csum_replace()
3351          * helpers anyway. Plus, this way we keep also separation of
3352          * concerns, since f.e. bpf_skb_store_bytes() should only take
3353          * care of stores.
3354          *
3355          * Currently, additional options and extension header space are
3356          * not supported, but flags register is reserved so we can adapt
3357          * that. For offloads, we mark packet as dodgy, so that headers
3358          * need to be verified first.
3359          */
3360         ret = bpf_skb_proto_xlat(skb, proto);
3361         bpf_compute_data_pointers(skb);
3362         return ret;
3363 }
3364
3365 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3366         .func           = bpf_skb_change_proto,
3367         .gpl_only       = false,
3368         .ret_type       = RET_INTEGER,
3369         .arg1_type      = ARG_PTR_TO_CTX,
3370         .arg2_type      = ARG_ANYTHING,
3371         .arg3_type      = ARG_ANYTHING,
3372 };
3373
3374 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3375 {
3376         /* We only allow a restricted subset to be changed for now. */
3377         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3378                      !skb_pkt_type_ok(pkt_type)))
3379                 return -EINVAL;
3380
3381         skb->pkt_type = pkt_type;
3382         return 0;
3383 }
3384
3385 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3386         .func           = bpf_skb_change_type,
3387         .gpl_only       = false,
3388         .ret_type       = RET_INTEGER,
3389         .arg1_type      = ARG_PTR_TO_CTX,
3390         .arg2_type      = ARG_ANYTHING,
3391 };
3392
3393 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3394 {
3395         switch (skb->protocol) {
3396         case htons(ETH_P_IP):
3397                 return sizeof(struct iphdr);
3398         case htons(ETH_P_IPV6):
3399                 return sizeof(struct ipv6hdr);
3400         default:
3401                 return ~0U;
3402         }
3403 }
3404
3405 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3406                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3407
3408 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3409                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3410                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3411                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3412                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3413                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3414                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3415
3416 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3417                             u64 flags)
3418 {
3419         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3420         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3421         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3422         unsigned int gso_type = SKB_GSO_DODGY;
3423         int ret;
3424
3425         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3426                 /* udp gso_size delineates datagrams, only allow if fixed */
3427                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3428                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3429                         return -ENOTSUPP;
3430         }
3431
3432         ret = skb_cow_head(skb, len_diff);
3433         if (unlikely(ret < 0))
3434                 return ret;
3435
3436         if (encap) {
3437                 if (skb->protocol != htons(ETH_P_IP) &&
3438                     skb->protocol != htons(ETH_P_IPV6))
3439                         return -ENOTSUPP;
3440
3441                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3442                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3443                         return -EINVAL;
3444
3445                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3446                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3447                         return -EINVAL;
3448
3449                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3450                     inner_mac_len < ETH_HLEN)
3451                         return -EINVAL;
3452
3453                 if (skb->encapsulation)
3454                         return -EALREADY;
3455
3456                 mac_len = skb->network_header - skb->mac_header;
3457                 inner_net = skb->network_header;
3458                 if (inner_mac_len > len_diff)
3459                         return -EINVAL;
3460                 inner_trans = skb->transport_header;
3461         }
3462
3463         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3464         if (unlikely(ret < 0))
3465                 return ret;
3466
3467         if (encap) {
3468                 skb->inner_mac_header = inner_net - inner_mac_len;
3469                 skb->inner_network_header = inner_net;
3470                 skb->inner_transport_header = inner_trans;
3471
3472                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3473                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3474                 else
3475                         skb_set_inner_protocol(skb, skb->protocol);
3476
3477                 skb->encapsulation = 1;
3478                 skb_set_network_header(skb, mac_len);
3479
3480                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3481                         gso_type |= SKB_GSO_UDP_TUNNEL;
3482                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3483                         gso_type |= SKB_GSO_GRE;
3484                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3485                         gso_type |= SKB_GSO_IPXIP6;
3486                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3487                         gso_type |= SKB_GSO_IPXIP4;
3488
3489                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3490                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3491                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3492                                         sizeof(struct ipv6hdr) :
3493                                         sizeof(struct iphdr);
3494
3495                         skb_set_transport_header(skb, mac_len + nh_len);
3496                 }
3497
3498                 /* Match skb->protocol to new outer l3 protocol */
3499                 if (skb->protocol == htons(ETH_P_IP) &&
3500                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3501                         skb->protocol = htons(ETH_P_IPV6);
3502                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3503                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3504                         skb->protocol = htons(ETH_P_IP);
3505         }
3506
3507         if (skb_is_gso(skb)) {
3508                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3509
3510                 /* Due to header grow, MSS needs to be downgraded. */
3511                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3512                         skb_decrease_gso_size(shinfo, len_diff);
3513
3514                 /* Header must be checked, and gso_segs recomputed. */
3515                 shinfo->gso_type |= gso_type;
3516                 shinfo->gso_segs = 0;
3517         }
3518
3519         return 0;
3520 }
3521
3522 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3523                               u64 flags)
3524 {
3525         int ret;
3526
3527         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3528                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3529                 return -EINVAL;
3530
3531         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3532                 /* udp gso_size delineates datagrams, only allow if fixed */
3533                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3534                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3535                         return -ENOTSUPP;
3536         }
3537
3538         ret = skb_unclone(skb, GFP_ATOMIC);
3539         if (unlikely(ret < 0))
3540                 return ret;
3541
3542         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3543         if (unlikely(ret < 0))
3544                 return ret;
3545
3546         if (skb_is_gso(skb)) {
3547                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3548
3549                 /* Due to header shrink, MSS can be upgraded. */
3550                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3551                         skb_increase_gso_size(shinfo, len_diff);
3552
3553                 /* Header must be checked, and gso_segs recomputed. */
3554                 shinfo->gso_type |= SKB_GSO_DODGY;
3555                 shinfo->gso_segs = 0;
3556         }
3557
3558         return 0;
3559 }
3560
3561 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3562
3563 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3564            u32, mode, u64, flags)
3565 {
3566         u32 len_diff_abs = abs(len_diff);
3567         bool shrink = len_diff < 0;
3568         int ret = 0;
3569
3570         if (unlikely(flags || mode))
3571                 return -EINVAL;
3572         if (unlikely(len_diff_abs > 0xfffU))
3573                 return -EFAULT;
3574
3575         if (!shrink) {
3576                 ret = skb_cow(skb, len_diff);
3577                 if (unlikely(ret < 0))
3578                         return ret;
3579                 __skb_push(skb, len_diff_abs);
3580                 memset(skb->data, 0, len_diff_abs);
3581         } else {
3582                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3583                         return -ENOMEM;
3584                 __skb_pull(skb, len_diff_abs);
3585         }
3586         if (tls_sw_has_ctx_rx(skb->sk)) {
3587                 struct strp_msg *rxm = strp_msg(skb);
3588
3589                 rxm->full_len += len_diff;
3590         }
3591         return ret;
3592 }
3593
3594 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3595         .func           = sk_skb_adjust_room,
3596         .gpl_only       = false,
3597         .ret_type       = RET_INTEGER,
3598         .arg1_type      = ARG_PTR_TO_CTX,
3599         .arg2_type      = ARG_ANYTHING,
3600         .arg3_type      = ARG_ANYTHING,
3601         .arg4_type      = ARG_ANYTHING,
3602 };
3603
3604 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3605            u32, mode, u64, flags)
3606 {
3607         u32 len_cur, len_diff_abs = abs(len_diff);
3608         u32 len_min = bpf_skb_net_base_len(skb);
3609         u32 len_max = BPF_SKB_MAX_LEN;
3610         __be16 proto = skb->protocol;
3611         bool shrink = len_diff < 0;
3612         u32 off;
3613         int ret;
3614
3615         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3616                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3617                 return -EINVAL;
3618         if (unlikely(len_diff_abs > 0xfffU))
3619                 return -EFAULT;
3620         if (unlikely(proto != htons(ETH_P_IP) &&
3621                      proto != htons(ETH_P_IPV6)))
3622                 return -ENOTSUPP;
3623
3624         off = skb_mac_header_len(skb);
3625         switch (mode) {
3626         case BPF_ADJ_ROOM_NET:
3627                 off += bpf_skb_net_base_len(skb);
3628                 break;
3629         case BPF_ADJ_ROOM_MAC:
3630                 break;
3631         default:
3632                 return -ENOTSUPP;
3633         }
3634
3635         len_cur = skb->len - skb_network_offset(skb);
3636         if ((shrink && (len_diff_abs >= len_cur ||
3637                         len_cur - len_diff_abs < len_min)) ||
3638             (!shrink && (skb->len + len_diff_abs > len_max &&
3639                          !skb_is_gso(skb))))
3640                 return -ENOTSUPP;
3641
3642         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3643                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3644         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3645                 __skb_reset_checksum_unnecessary(skb);
3646
3647         bpf_compute_data_pointers(skb);
3648         return ret;
3649 }
3650
3651 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3652         .func           = bpf_skb_adjust_room,
3653         .gpl_only       = false,
3654         .ret_type       = RET_INTEGER,
3655         .arg1_type      = ARG_PTR_TO_CTX,
3656         .arg2_type      = ARG_ANYTHING,
3657         .arg3_type      = ARG_ANYTHING,
3658         .arg4_type      = ARG_ANYTHING,
3659 };
3660
3661 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3662 {
3663         u32 min_len = skb_network_offset(skb);
3664
3665         if (skb_transport_header_was_set(skb))
3666                 min_len = skb_transport_offset(skb);
3667         if (skb->ip_summed == CHECKSUM_PARTIAL)
3668                 min_len = skb_checksum_start_offset(skb) +
3669                           skb->csum_offset + sizeof(__sum16);
3670         return min_len;
3671 }
3672
3673 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3674 {
3675         unsigned int old_len = skb->len;
3676         int ret;
3677
3678         ret = __skb_grow_rcsum(skb, new_len);
3679         if (!ret)
3680                 memset(skb->data + old_len, 0, new_len - old_len);
3681         return ret;
3682 }
3683
3684 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3685 {
3686         return __skb_trim_rcsum(skb, new_len);
3687 }
3688
3689 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3690                                         u64 flags)
3691 {
3692         u32 max_len = BPF_SKB_MAX_LEN;
3693         u32 min_len = __bpf_skb_min_len(skb);
3694         int ret;
3695
3696         if (unlikely(flags || new_len > max_len || new_len < min_len))
3697                 return -EINVAL;
3698         if (skb->encapsulation)
3699                 return -ENOTSUPP;
3700
3701         /* The basic idea of this helper is that it's performing the
3702          * needed work to either grow or trim an skb, and eBPF program
3703          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3704          * bpf_lX_csum_replace() and others rather than passing a raw
3705          * buffer here. This one is a slow path helper and intended
3706          * for replies with control messages.
3707          *
3708          * Like in bpf_skb_change_proto(), we want to keep this rather
3709          * minimal and without protocol specifics so that we are able
3710          * to separate concerns as in bpf_skb_store_bytes() should only
3711          * be the one responsible for writing buffers.
3712          *
3713          * It's really expected to be a slow path operation here for
3714          * control message replies, so we're implicitly linearizing,
3715          * uncloning and drop offloads from the skb by this.
3716          */
3717         ret = __bpf_try_make_writable(skb, skb->len);
3718         if (!ret) {
3719                 if (new_len > skb->len)
3720                         ret = bpf_skb_grow_rcsum(skb, new_len);
3721                 else if (new_len < skb->len)
3722                         ret = bpf_skb_trim_rcsum(skb, new_len);
3723                 if (!ret && skb_is_gso(skb))
3724                         skb_gso_reset(skb);
3725         }
3726         return ret;
3727 }
3728
3729 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3730            u64, flags)
3731 {
3732         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3733
3734         bpf_compute_data_pointers(skb);
3735         return ret;
3736 }
3737
3738 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3739         .func           = bpf_skb_change_tail,
3740         .gpl_only       = false,
3741         .ret_type       = RET_INTEGER,
3742         .arg1_type      = ARG_PTR_TO_CTX,
3743         .arg2_type      = ARG_ANYTHING,
3744         .arg3_type      = ARG_ANYTHING,
3745 };
3746
3747 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3748            u64, flags)
3749 {
3750         return __bpf_skb_change_tail(skb, new_len, flags);
3751 }
3752
3753 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3754         .func           = sk_skb_change_tail,
3755         .gpl_only       = false,
3756         .ret_type       = RET_INTEGER,
3757         .arg1_type      = ARG_PTR_TO_CTX,
3758         .arg2_type      = ARG_ANYTHING,
3759         .arg3_type      = ARG_ANYTHING,
3760 };
3761
3762 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3763                                         u64 flags)
3764 {
3765         u32 max_len = BPF_SKB_MAX_LEN;
3766         u32 new_len = skb->len + head_room;
3767         int ret;
3768
3769         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3770                      new_len < skb->len))
3771                 return -EINVAL;
3772
3773         ret = skb_cow(skb, head_room);
3774         if (likely(!ret)) {
3775                 /* Idea for this helper is that we currently only
3776                  * allow to expand on mac header. This means that
3777                  * skb->protocol network header, etc, stay as is.
3778                  * Compared to bpf_skb_change_tail(), we're more
3779                  * flexible due to not needing to linearize or
3780                  * reset GSO. Intention for this helper is to be
3781                  * used by an L3 skb that needs to push mac header
3782                  * for redirection into L2 device.
3783                  */
3784                 __skb_push(skb, head_room);
3785                 memset(skb->data, 0, head_room);
3786                 skb_reset_mac_header(skb);
3787         }
3788
3789         return ret;
3790 }
3791
3792 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3793            u64, flags)
3794 {
3795         int ret = __bpf_skb_change_head(skb, head_room, flags);
3796
3797         bpf_compute_data_pointers(skb);
3798         return ret;
3799 }
3800
3801 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3802         .func           = bpf_skb_change_head,
3803         .gpl_only       = false,
3804         .ret_type       = RET_INTEGER,
3805         .arg1_type      = ARG_PTR_TO_CTX,
3806         .arg2_type      = ARG_ANYTHING,
3807         .arg3_type      = ARG_ANYTHING,
3808 };
3809
3810 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3811            u64, flags)
3812 {
3813         return __bpf_skb_change_head(skb, head_room, flags);
3814 }
3815
3816 static const struct bpf_func_proto sk_skb_change_head_proto = {
3817         .func           = sk_skb_change_head,
3818         .gpl_only       = false,
3819         .ret_type       = RET_INTEGER,
3820         .arg1_type      = ARG_PTR_TO_CTX,
3821         .arg2_type      = ARG_ANYTHING,
3822         .arg3_type      = ARG_ANYTHING,
3823 };
3824 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3825 {
3826         return xdp_data_meta_unsupported(xdp) ? 0 :
3827                xdp->data - xdp->data_meta;
3828 }
3829
3830 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3831 {
3832         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3833         unsigned long metalen = xdp_get_metalen(xdp);
3834         void *data_start = xdp_frame_end + metalen;
3835         void *data = xdp->data + offset;
3836
3837         if (unlikely(data < data_start ||
3838                      data > xdp->data_end - ETH_HLEN))
3839                 return -EINVAL;
3840
3841         if (metalen)
3842                 memmove(xdp->data_meta + offset,
3843                         xdp->data_meta, metalen);
3844         xdp->data_meta += offset;
3845         xdp->data = data;
3846
3847         return 0;
3848 }
3849
3850 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3851         .func           = bpf_xdp_adjust_head,
3852         .gpl_only       = false,
3853         .ret_type       = RET_INTEGER,
3854         .arg1_type      = ARG_PTR_TO_CTX,
3855         .arg2_type      = ARG_ANYTHING,
3856 };
3857
3858 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3859 {
3860         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3861         void *data_end = xdp->data_end + offset;
3862
3863         /* Notice that xdp_data_hard_end have reserved some tailroom */
3864         if (unlikely(data_end > data_hard_end))
3865                 return -EINVAL;
3866
3867         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3868         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3869                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3870                 return -EINVAL;
3871         }
3872
3873         if (unlikely(data_end < xdp->data + ETH_HLEN))
3874                 return -EINVAL;
3875
3876         /* Clear memory area on grow, can contain uninit kernel memory */
3877         if (offset > 0)
3878                 memset(xdp->data_end, 0, offset);
3879
3880         xdp->data_end = data_end;
3881
3882         return 0;
3883 }
3884
3885 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3886         .func           = bpf_xdp_adjust_tail,
3887         .gpl_only       = false,
3888         .ret_type       = RET_INTEGER,
3889         .arg1_type      = ARG_PTR_TO_CTX,
3890         .arg2_type      = ARG_ANYTHING,
3891 };
3892
3893 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3894 {
3895         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3896         void *meta = xdp->data_meta + offset;
3897         unsigned long metalen = xdp->data - meta;
3898
3899         if (xdp_data_meta_unsupported(xdp))
3900                 return -ENOTSUPP;
3901         if (unlikely(meta < xdp_frame_end ||
3902                      meta > xdp->data))
3903                 return -EINVAL;
3904         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3905                      (metalen > 32)))
3906                 return -EACCES;
3907
3908         xdp->data_meta = meta;
3909
3910         return 0;
3911 }
3912
3913 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3914         .func           = bpf_xdp_adjust_meta,
3915         .gpl_only       = false,
3916         .ret_type       = RET_INTEGER,
3917         .arg1_type      = ARG_PTR_TO_CTX,
3918         .arg2_type      = ARG_ANYTHING,
3919 };
3920
3921 void xdp_do_flush(void)
3922 {
3923         __dev_flush();
3924         __cpu_map_flush();
3925         __xsk_map_flush();
3926 }
3927 EXPORT_SYMBOL_GPL(xdp_do_flush);
3928
3929 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3930                     struct bpf_prog *xdp_prog)
3931 {
3932         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3933         enum bpf_map_type map_type = ri->map_type;
3934         void *fwd = ri->tgt_value;
3935         u32 map_id = ri->map_id;
3936         int err;
3937
3938         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3939         ri->map_type = BPF_MAP_TYPE_UNSPEC;
3940
3941         switch (map_type) {
3942         case BPF_MAP_TYPE_DEVMAP:
3943                 fallthrough;
3944         case BPF_MAP_TYPE_DEVMAP_HASH:
3945                 err = dev_map_enqueue(fwd, xdp, dev);
3946                 break;
3947         case BPF_MAP_TYPE_CPUMAP:
3948                 err = cpu_map_enqueue(fwd, xdp, dev);
3949                 break;
3950         case BPF_MAP_TYPE_XSKMAP:
3951                 err = __xsk_map_redirect(fwd, xdp);
3952                 break;
3953         case BPF_MAP_TYPE_UNSPEC:
3954                 if (map_id == INT_MAX) {
3955                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3956                         if (unlikely(!fwd)) {
3957                                 err = -EINVAL;
3958                                 break;
3959                         }
3960                         err = dev_xdp_enqueue(fwd, xdp, dev);
3961                         break;
3962                 }
3963                 fallthrough;
3964         default:
3965                 err = -EBADRQC;
3966         }
3967
3968         if (unlikely(err))
3969                 goto err;
3970
3971         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
3972         return 0;
3973 err:
3974         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
3975         return err;
3976 }
3977 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3978
3979 static int xdp_do_generic_redirect_map(struct net_device *dev,
3980                                        struct sk_buff *skb,
3981                                        struct xdp_buff *xdp,
3982                                        struct bpf_prog *xdp_prog,
3983                                        void *fwd,
3984                                        enum bpf_map_type map_type, u32 map_id)
3985 {
3986         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3987         int err;
3988
3989         switch (map_type) {
3990         case BPF_MAP_TYPE_DEVMAP:
3991                 fallthrough;
3992         case BPF_MAP_TYPE_DEVMAP_HASH:
3993                 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
3994                 if (unlikely(err))
3995                         goto err;
3996                 break;
3997         case BPF_MAP_TYPE_XSKMAP:
3998                 err = xsk_generic_rcv(fwd, xdp);
3999                 if (err)
4000                         goto err;
4001                 consume_skb(skb);
4002                 break;
4003         default:
4004                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4005                 err = -EBADRQC;
4006                 goto err;
4007         }
4008
4009         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4010         return 0;
4011 err:
4012         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4013         return err;
4014 }
4015
4016 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4017                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4018 {
4019         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4020         enum bpf_map_type map_type = ri->map_type;
4021         void *fwd = ri->tgt_value;
4022         u32 map_id = ri->map_id;
4023         int err;
4024
4025         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4026         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4027
4028         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4029                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4030                 if (unlikely(!fwd)) {
4031                         err = -EINVAL;
4032                         goto err;
4033                 }
4034
4035                 err = xdp_ok_fwd_dev(fwd, skb->len);
4036                 if (unlikely(err))
4037                         goto err;
4038
4039                 skb->dev = fwd;
4040                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4041                 generic_xdp_tx(skb, xdp_prog);
4042                 return 0;
4043         }
4044
4045         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4046 err:
4047         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4048         return err;
4049 }
4050
4051 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4052 {
4053         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4054
4055         if (unlikely(flags))
4056                 return XDP_ABORTED;
4057
4058         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4059          * by map_idr) is used for ifindex based XDP redirect.
4060          */
4061         ri->tgt_index = ifindex;
4062         ri->map_id = INT_MAX;
4063         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4064
4065         return XDP_REDIRECT;
4066 }
4067
4068 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4069         .func           = bpf_xdp_redirect,
4070         .gpl_only       = false,
4071         .ret_type       = RET_INTEGER,
4072         .arg1_type      = ARG_ANYTHING,
4073         .arg2_type      = ARG_ANYTHING,
4074 };
4075
4076 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4077            u64, flags)
4078 {
4079         return map->ops->map_redirect(map, ifindex, flags);
4080 }
4081
4082 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4083         .func           = bpf_xdp_redirect_map,
4084         .gpl_only       = false,
4085         .ret_type       = RET_INTEGER,
4086         .arg1_type      = ARG_CONST_MAP_PTR,
4087         .arg2_type      = ARG_ANYTHING,
4088         .arg3_type      = ARG_ANYTHING,
4089 };
4090
4091 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4092                                   unsigned long off, unsigned long len)
4093 {
4094         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4095
4096         if (unlikely(!ptr))
4097                 return len;
4098         if (ptr != dst_buff)
4099                 memcpy(dst_buff, ptr, len);
4100
4101         return 0;
4102 }
4103
4104 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4105            u64, flags, void *, meta, u64, meta_size)
4106 {
4107         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4108
4109         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4110                 return -EINVAL;
4111         if (unlikely(!skb || skb_size > skb->len))
4112                 return -EFAULT;
4113
4114         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4115                                 bpf_skb_copy);
4116 }
4117
4118 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4119         .func           = bpf_skb_event_output,
4120         .gpl_only       = true,
4121         .ret_type       = RET_INTEGER,
4122         .arg1_type      = ARG_PTR_TO_CTX,
4123         .arg2_type      = ARG_CONST_MAP_PTR,
4124         .arg3_type      = ARG_ANYTHING,
4125         .arg4_type      = ARG_PTR_TO_MEM,
4126         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4127 };
4128
4129 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4130
4131 const struct bpf_func_proto bpf_skb_output_proto = {
4132         .func           = bpf_skb_event_output,
4133         .gpl_only       = true,
4134         .ret_type       = RET_INTEGER,
4135         .arg1_type      = ARG_PTR_TO_BTF_ID,
4136         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4137         .arg2_type      = ARG_CONST_MAP_PTR,
4138         .arg3_type      = ARG_ANYTHING,
4139         .arg4_type      = ARG_PTR_TO_MEM,
4140         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4141 };
4142
4143 static unsigned short bpf_tunnel_key_af(u64 flags)
4144 {
4145         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4146 }
4147
4148 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4149            u32, size, u64, flags)
4150 {
4151         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4152         u8 compat[sizeof(struct bpf_tunnel_key)];
4153         void *to_orig = to;
4154         int err;
4155
4156         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4157                 err = -EINVAL;
4158                 goto err_clear;
4159         }
4160         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4161                 err = -EPROTO;
4162                 goto err_clear;
4163         }
4164         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4165                 err = -EINVAL;
4166                 switch (size) {
4167                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4168                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4169                         goto set_compat;
4170                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4171                         /* Fixup deprecated structure layouts here, so we have
4172                          * a common path later on.
4173                          */
4174                         if (ip_tunnel_info_af(info) != AF_INET)
4175                                 goto err_clear;
4176 set_compat:
4177                         to = (struct bpf_tunnel_key *)compat;
4178                         break;
4179                 default:
4180                         goto err_clear;
4181                 }
4182         }
4183
4184         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4185         to->tunnel_tos = info->key.tos;
4186         to->tunnel_ttl = info->key.ttl;
4187         to->tunnel_ext = 0;
4188
4189         if (flags & BPF_F_TUNINFO_IPV6) {
4190                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4191                        sizeof(to->remote_ipv6));
4192                 to->tunnel_label = be32_to_cpu(info->key.label);
4193         } else {
4194                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4195                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4196                 to->tunnel_label = 0;
4197         }
4198
4199         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4200                 memcpy(to_orig, to, size);
4201
4202         return 0;
4203 err_clear:
4204         memset(to_orig, 0, size);
4205         return err;
4206 }
4207
4208 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4209         .func           = bpf_skb_get_tunnel_key,
4210         .gpl_only       = false,
4211         .ret_type       = RET_INTEGER,
4212         .arg1_type      = ARG_PTR_TO_CTX,
4213         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4214         .arg3_type      = ARG_CONST_SIZE,
4215         .arg4_type      = ARG_ANYTHING,
4216 };
4217
4218 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4219 {
4220         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4221         int err;
4222
4223         if (unlikely(!info ||
4224                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4225                 err = -ENOENT;
4226                 goto err_clear;
4227         }
4228         if (unlikely(size < info->options_len)) {
4229                 err = -ENOMEM;
4230                 goto err_clear;
4231         }
4232
4233         ip_tunnel_info_opts_get(to, info);
4234         if (size > info->options_len)
4235                 memset(to + info->options_len, 0, size - info->options_len);
4236
4237         return info->options_len;
4238 err_clear:
4239         memset(to, 0, size);
4240         return err;
4241 }
4242
4243 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4244         .func           = bpf_skb_get_tunnel_opt,
4245         .gpl_only       = false,
4246         .ret_type       = RET_INTEGER,
4247         .arg1_type      = ARG_PTR_TO_CTX,
4248         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4249         .arg3_type      = ARG_CONST_SIZE,
4250 };
4251
4252 static struct metadata_dst __percpu *md_dst;
4253
4254 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4255            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4256 {
4257         struct metadata_dst *md = this_cpu_ptr(md_dst);
4258         u8 compat[sizeof(struct bpf_tunnel_key)];
4259         struct ip_tunnel_info *info;
4260
4261         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4262                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4263                 return -EINVAL;
4264         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4265                 switch (size) {
4266                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4267                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4268                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4269                         /* Fixup deprecated structure layouts here, so we have
4270                          * a common path later on.
4271                          */
4272                         memcpy(compat, from, size);
4273                         memset(compat + size, 0, sizeof(compat) - size);
4274                         from = (const struct bpf_tunnel_key *) compat;
4275                         break;
4276                 default:
4277                         return -EINVAL;
4278                 }
4279         }
4280         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4281                      from->tunnel_ext))
4282                 return -EINVAL;
4283
4284         skb_dst_drop(skb);
4285         dst_hold((struct dst_entry *) md);
4286         skb_dst_set(skb, (struct dst_entry *) md);
4287
4288         info = &md->u.tun_info;
4289         memset(info, 0, sizeof(*info));
4290         info->mode = IP_TUNNEL_INFO_TX;
4291
4292         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4293         if (flags & BPF_F_DONT_FRAGMENT)
4294                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4295         if (flags & BPF_F_ZERO_CSUM_TX)
4296                 info->key.tun_flags &= ~TUNNEL_CSUM;
4297         if (flags & BPF_F_SEQ_NUMBER)
4298                 info->key.tun_flags |= TUNNEL_SEQ;
4299
4300         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4301         info->key.tos = from->tunnel_tos;
4302         info->key.ttl = from->tunnel_ttl;
4303
4304         if (flags & BPF_F_TUNINFO_IPV6) {
4305                 info->mode |= IP_TUNNEL_INFO_IPV6;
4306                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4307                        sizeof(from->remote_ipv6));
4308                 info->key.label = cpu_to_be32(from->tunnel_label) &
4309                                   IPV6_FLOWLABEL_MASK;
4310         } else {
4311                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4312         }
4313
4314         return 0;
4315 }
4316
4317 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4318         .func           = bpf_skb_set_tunnel_key,
4319         .gpl_only       = false,
4320         .ret_type       = RET_INTEGER,
4321         .arg1_type      = ARG_PTR_TO_CTX,
4322         .arg2_type      = ARG_PTR_TO_MEM,
4323         .arg3_type      = ARG_CONST_SIZE,
4324         .arg4_type      = ARG_ANYTHING,
4325 };
4326
4327 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4328            const u8 *, from, u32, size)
4329 {
4330         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4331         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4332
4333         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4334                 return -EINVAL;
4335         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4336                 return -ENOMEM;
4337
4338         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4339
4340         return 0;
4341 }
4342
4343 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4344         .func           = bpf_skb_set_tunnel_opt,
4345         .gpl_only       = false,
4346         .ret_type       = RET_INTEGER,
4347         .arg1_type      = ARG_PTR_TO_CTX,
4348         .arg2_type      = ARG_PTR_TO_MEM,
4349         .arg3_type      = ARG_CONST_SIZE,
4350 };
4351
4352 static const struct bpf_func_proto *
4353 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4354 {
4355         if (!md_dst) {
4356                 struct metadata_dst __percpu *tmp;
4357
4358                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4359                                                 METADATA_IP_TUNNEL,
4360                                                 GFP_KERNEL);
4361                 if (!tmp)
4362                         return NULL;
4363                 if (cmpxchg(&md_dst, NULL, tmp))
4364                         metadata_dst_free_percpu(tmp);
4365         }
4366
4367         switch (which) {
4368         case BPF_FUNC_skb_set_tunnel_key:
4369                 return &bpf_skb_set_tunnel_key_proto;
4370         case BPF_FUNC_skb_set_tunnel_opt:
4371                 return &bpf_skb_set_tunnel_opt_proto;
4372         default:
4373                 return NULL;
4374         }
4375 }
4376
4377 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4378            u32, idx)
4379 {
4380         struct bpf_array *array = container_of(map, struct bpf_array, map);
4381         struct cgroup *cgrp;
4382         struct sock *sk;
4383
4384         sk = skb_to_full_sk(skb);
4385         if (!sk || !sk_fullsock(sk))
4386                 return -ENOENT;
4387         if (unlikely(idx >= array->map.max_entries))
4388                 return -E2BIG;
4389
4390         cgrp = READ_ONCE(array->ptrs[idx]);
4391         if (unlikely(!cgrp))
4392                 return -EAGAIN;
4393
4394         return sk_under_cgroup_hierarchy(sk, cgrp);
4395 }
4396
4397 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4398         .func           = bpf_skb_under_cgroup,
4399         .gpl_only       = false,
4400         .ret_type       = RET_INTEGER,
4401         .arg1_type      = ARG_PTR_TO_CTX,
4402         .arg2_type      = ARG_CONST_MAP_PTR,
4403         .arg3_type      = ARG_ANYTHING,
4404 };
4405
4406 #ifdef CONFIG_SOCK_CGROUP_DATA
4407 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4408 {
4409         struct cgroup *cgrp;
4410
4411         sk = sk_to_full_sk(sk);
4412         if (!sk || !sk_fullsock(sk))
4413                 return 0;
4414
4415         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4416         return cgroup_id(cgrp);
4417 }
4418
4419 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4420 {
4421         return __bpf_sk_cgroup_id(skb->sk);
4422 }
4423
4424 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4425         .func           = bpf_skb_cgroup_id,
4426         .gpl_only       = false,
4427         .ret_type       = RET_INTEGER,
4428         .arg1_type      = ARG_PTR_TO_CTX,
4429 };
4430
4431 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4432                                               int ancestor_level)
4433 {
4434         struct cgroup *ancestor;
4435         struct cgroup *cgrp;
4436
4437         sk = sk_to_full_sk(sk);
4438         if (!sk || !sk_fullsock(sk))
4439                 return 0;
4440
4441         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4442         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4443         if (!ancestor)
4444                 return 0;
4445
4446         return cgroup_id(ancestor);
4447 }
4448
4449 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4450            ancestor_level)
4451 {
4452         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4453 }
4454
4455 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4456         .func           = bpf_skb_ancestor_cgroup_id,
4457         .gpl_only       = false,
4458         .ret_type       = RET_INTEGER,
4459         .arg1_type      = ARG_PTR_TO_CTX,
4460         .arg2_type      = ARG_ANYTHING,
4461 };
4462
4463 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4464 {
4465         return __bpf_sk_cgroup_id(sk);
4466 }
4467
4468 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4469         .func           = bpf_sk_cgroup_id,
4470         .gpl_only       = false,
4471         .ret_type       = RET_INTEGER,
4472         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4473 };
4474
4475 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4476 {
4477         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4478 }
4479
4480 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4481         .func           = bpf_sk_ancestor_cgroup_id,
4482         .gpl_only       = false,
4483         .ret_type       = RET_INTEGER,
4484         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4485         .arg2_type      = ARG_ANYTHING,
4486 };
4487 #endif
4488
4489 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4490                                   unsigned long off, unsigned long len)
4491 {
4492         memcpy(dst_buff, src_buff + off, len);
4493         return 0;
4494 }
4495
4496 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4497            u64, flags, void *, meta, u64, meta_size)
4498 {
4499         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4500
4501         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4502                 return -EINVAL;
4503         if (unlikely(!xdp ||
4504                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4505                 return -EFAULT;
4506
4507         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4508                                 xdp_size, bpf_xdp_copy);
4509 }
4510
4511 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4512         .func           = bpf_xdp_event_output,
4513         .gpl_only       = true,
4514         .ret_type       = RET_INTEGER,
4515         .arg1_type      = ARG_PTR_TO_CTX,
4516         .arg2_type      = ARG_CONST_MAP_PTR,
4517         .arg3_type      = ARG_ANYTHING,
4518         .arg4_type      = ARG_PTR_TO_MEM,
4519         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4520 };
4521
4522 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4523
4524 const struct bpf_func_proto bpf_xdp_output_proto = {
4525         .func           = bpf_xdp_event_output,
4526         .gpl_only       = true,
4527         .ret_type       = RET_INTEGER,
4528         .arg1_type      = ARG_PTR_TO_BTF_ID,
4529         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4530         .arg2_type      = ARG_CONST_MAP_PTR,
4531         .arg3_type      = ARG_ANYTHING,
4532         .arg4_type      = ARG_PTR_TO_MEM,
4533         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4534 };
4535
4536 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4537 {
4538         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4539 }
4540
4541 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4542         .func           = bpf_get_socket_cookie,
4543         .gpl_only       = false,
4544         .ret_type       = RET_INTEGER,
4545         .arg1_type      = ARG_PTR_TO_CTX,
4546 };
4547
4548 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4549 {
4550         return __sock_gen_cookie(ctx->sk);
4551 }
4552
4553 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4554         .func           = bpf_get_socket_cookie_sock_addr,
4555         .gpl_only       = false,
4556         .ret_type       = RET_INTEGER,
4557         .arg1_type      = ARG_PTR_TO_CTX,
4558 };
4559
4560 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4561 {
4562         return __sock_gen_cookie(ctx);
4563 }
4564
4565 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4566         .func           = bpf_get_socket_cookie_sock,
4567         .gpl_only       = false,
4568         .ret_type       = RET_INTEGER,
4569         .arg1_type      = ARG_PTR_TO_CTX,
4570 };
4571
4572 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4573 {
4574         return sk ? sock_gen_cookie(sk) : 0;
4575 }
4576
4577 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4578         .func           = bpf_get_socket_ptr_cookie,
4579         .gpl_only       = false,
4580         .ret_type       = RET_INTEGER,
4581         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4582 };
4583
4584 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4585 {
4586         return __sock_gen_cookie(ctx->sk);
4587 }
4588
4589 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4590         .func           = bpf_get_socket_cookie_sock_ops,
4591         .gpl_only       = false,
4592         .ret_type       = RET_INTEGER,
4593         .arg1_type      = ARG_PTR_TO_CTX,
4594 };
4595
4596 static u64 __bpf_get_netns_cookie(struct sock *sk)
4597 {
4598         const struct net *net = sk ? sock_net(sk) : &init_net;
4599
4600         return net->net_cookie;
4601 }
4602
4603 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4604 {
4605         return __bpf_get_netns_cookie(ctx);
4606 }
4607
4608 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4609         .func           = bpf_get_netns_cookie_sock,
4610         .gpl_only       = false,
4611         .ret_type       = RET_INTEGER,
4612         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4613 };
4614
4615 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4616 {
4617         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4618 }
4619
4620 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4621         .func           = bpf_get_netns_cookie_sock_addr,
4622         .gpl_only       = false,
4623         .ret_type       = RET_INTEGER,
4624         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4625 };
4626
4627 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4628 {
4629         struct sock *sk = sk_to_full_sk(skb->sk);
4630         kuid_t kuid;
4631
4632         if (!sk || !sk_fullsock(sk))
4633                 return overflowuid;
4634         kuid = sock_net_uid(sock_net(sk), sk);
4635         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4636 }
4637
4638 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4639         .func           = bpf_get_socket_uid,
4640         .gpl_only       = false,
4641         .ret_type       = RET_INTEGER,
4642         .arg1_type      = ARG_PTR_TO_CTX,
4643 };
4644
4645 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4646                            char *optval, int optlen)
4647 {
4648         char devname[IFNAMSIZ];
4649         int val, valbool;
4650         struct net *net;
4651         int ifindex;
4652         int ret = 0;
4653
4654         if (!sk_fullsock(sk))
4655                 return -EINVAL;
4656
4657         sock_owned_by_me(sk);
4658
4659         if (level == SOL_SOCKET) {
4660                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4661                         return -EINVAL;
4662                 val = *((int *)optval);
4663                 valbool = val ? 1 : 0;
4664
4665                 /* Only some socketops are supported */
4666                 switch (optname) {
4667                 case SO_RCVBUF:
4668                         val = min_t(u32, val, sysctl_rmem_max);
4669                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4670                         WRITE_ONCE(sk->sk_rcvbuf,
4671                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4672                         break;
4673                 case SO_SNDBUF:
4674                         val = min_t(u32, val, sysctl_wmem_max);
4675                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4676                         WRITE_ONCE(sk->sk_sndbuf,
4677                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4678                         break;
4679                 case SO_MAX_PACING_RATE: /* 32bit version */
4680                         if (val != ~0U)
4681                                 cmpxchg(&sk->sk_pacing_status,
4682                                         SK_PACING_NONE,
4683                                         SK_PACING_NEEDED);
4684                         sk->sk_max_pacing_rate = (val == ~0U) ?
4685                                                  ~0UL : (unsigned int)val;
4686                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4687                                                  sk->sk_max_pacing_rate);
4688                         break;
4689                 case SO_PRIORITY:
4690                         sk->sk_priority = val;
4691                         break;
4692                 case SO_RCVLOWAT:
4693                         if (val < 0)
4694                                 val = INT_MAX;
4695                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4696                         break;
4697                 case SO_MARK:
4698                         if (sk->sk_mark != val) {
4699                                 sk->sk_mark = val;
4700                                 sk_dst_reset(sk);
4701                         }
4702                         break;
4703                 case SO_BINDTODEVICE:
4704                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4705                         strncpy(devname, optval, optlen);
4706                         devname[optlen] = 0;
4707
4708                         ifindex = 0;
4709                         if (devname[0] != '\0') {
4710                                 struct net_device *dev;
4711
4712                                 ret = -ENODEV;
4713
4714                                 net = sock_net(sk);
4715                                 dev = dev_get_by_name(net, devname);
4716                                 if (!dev)
4717                                         break;
4718                                 ifindex = dev->ifindex;
4719                                 dev_put(dev);
4720                         }
4721                         fallthrough;
4722                 case SO_BINDTOIFINDEX:
4723                         if (optname == SO_BINDTOIFINDEX)
4724                                 ifindex = val;
4725                         ret = sock_bindtoindex(sk, ifindex, false);
4726                         break;
4727                 case SO_KEEPALIVE:
4728                         if (sk->sk_prot->keepalive)
4729                                 sk->sk_prot->keepalive(sk, valbool);
4730                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4731                         break;
4732                 case SO_REUSEPORT:
4733                         sk->sk_reuseport = valbool;
4734                         break;
4735                 default:
4736                         ret = -EINVAL;
4737                 }
4738 #ifdef CONFIG_INET
4739         } else if (level == SOL_IP) {
4740                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4741                         return -EINVAL;
4742
4743                 val = *((int *)optval);
4744                 /* Only some options are supported */
4745                 switch (optname) {
4746                 case IP_TOS:
4747                         if (val < -1 || val > 0xff) {
4748                                 ret = -EINVAL;
4749                         } else {
4750                                 struct inet_sock *inet = inet_sk(sk);
4751
4752                                 if (val == -1)
4753                                         val = 0;
4754                                 inet->tos = val;
4755                         }
4756                         break;
4757                 default:
4758                         ret = -EINVAL;
4759                 }
4760 #if IS_ENABLED(CONFIG_IPV6)
4761         } else if (level == SOL_IPV6) {
4762                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4763                         return -EINVAL;
4764
4765                 val = *((int *)optval);
4766                 /* Only some options are supported */
4767                 switch (optname) {
4768                 case IPV6_TCLASS:
4769                         if (val < -1 || val > 0xff) {
4770                                 ret = -EINVAL;
4771                         } else {
4772                                 struct ipv6_pinfo *np = inet6_sk(sk);
4773
4774                                 if (val == -1)
4775                                         val = 0;
4776                                 np->tclass = val;
4777                         }
4778                         break;
4779                 default:
4780                         ret = -EINVAL;
4781                 }
4782 #endif
4783         } else if (level == SOL_TCP &&
4784                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4785                 if (optname == TCP_CONGESTION) {
4786                         char name[TCP_CA_NAME_MAX];
4787
4788                         strncpy(name, optval, min_t(long, optlen,
4789                                                     TCP_CA_NAME_MAX-1));
4790                         name[TCP_CA_NAME_MAX-1] = 0;
4791                         ret = tcp_set_congestion_control(sk, name, false, true);
4792                 } else {
4793                         struct inet_connection_sock *icsk = inet_csk(sk);
4794                         struct tcp_sock *tp = tcp_sk(sk);
4795                         unsigned long timeout;
4796
4797                         if (optlen != sizeof(int))
4798                                 return -EINVAL;
4799
4800                         val = *((int *)optval);
4801                         /* Only some options are supported */
4802                         switch (optname) {
4803                         case TCP_BPF_IW:
4804                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4805                                         ret = -EINVAL;
4806                                 else
4807                                         tp->snd_cwnd = val;
4808                                 break;
4809                         case TCP_BPF_SNDCWND_CLAMP:
4810                                 if (val <= 0) {
4811                                         ret = -EINVAL;
4812                                 } else {
4813                                         tp->snd_cwnd_clamp = val;
4814                                         tp->snd_ssthresh = val;
4815                                 }
4816                                 break;
4817                         case TCP_BPF_DELACK_MAX:
4818                                 timeout = usecs_to_jiffies(val);
4819                                 if (timeout > TCP_DELACK_MAX ||
4820                                     timeout < TCP_TIMEOUT_MIN)
4821                                         return -EINVAL;
4822                                 inet_csk(sk)->icsk_delack_max = timeout;
4823                                 break;
4824                         case TCP_BPF_RTO_MIN:
4825                                 timeout = usecs_to_jiffies(val);
4826                                 if (timeout > TCP_RTO_MIN ||
4827                                     timeout < TCP_TIMEOUT_MIN)
4828                                         return -EINVAL;
4829                                 inet_csk(sk)->icsk_rto_min = timeout;
4830                                 break;
4831                         case TCP_SAVE_SYN:
4832                                 if (val < 0 || val > 1)
4833                                         ret = -EINVAL;
4834                                 else
4835                                         tp->save_syn = val;
4836                                 break;
4837                         case TCP_KEEPIDLE:
4838                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4839                                 break;
4840                         case TCP_KEEPINTVL:
4841                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4842                                         ret = -EINVAL;
4843                                 else
4844                                         tp->keepalive_intvl = val * HZ;
4845                                 break;
4846                         case TCP_KEEPCNT:
4847                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4848                                         ret = -EINVAL;
4849                                 else
4850                                         tp->keepalive_probes = val;
4851                                 break;
4852                         case TCP_SYNCNT:
4853                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4854                                         ret = -EINVAL;
4855                                 else
4856                                         icsk->icsk_syn_retries = val;
4857                                 break;
4858                         case TCP_USER_TIMEOUT:
4859                                 if (val < 0)
4860                                         ret = -EINVAL;
4861                                 else
4862                                         icsk->icsk_user_timeout = val;
4863                                 break;
4864                         case TCP_NOTSENT_LOWAT:
4865                                 tp->notsent_lowat = val;
4866                                 sk->sk_write_space(sk);
4867                                 break;
4868                         case TCP_WINDOW_CLAMP:
4869                                 ret = tcp_set_window_clamp(sk, val);
4870                                 break;
4871                         default:
4872                                 ret = -EINVAL;
4873                         }
4874                 }
4875 #endif
4876         } else {
4877                 ret = -EINVAL;
4878         }
4879         return ret;
4880 }
4881
4882 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4883                            char *optval, int optlen)
4884 {
4885         if (!sk_fullsock(sk))
4886                 goto err_clear;
4887
4888         sock_owned_by_me(sk);
4889
4890         if (level == SOL_SOCKET) {
4891                 if (optlen != sizeof(int))
4892                         goto err_clear;
4893
4894                 switch (optname) {
4895                 case SO_MARK:
4896                         *((int *)optval) = sk->sk_mark;
4897                         break;
4898                 case SO_PRIORITY:
4899                         *((int *)optval) = sk->sk_priority;
4900                         break;
4901                 case SO_BINDTOIFINDEX:
4902                         *((int *)optval) = sk->sk_bound_dev_if;
4903                         break;
4904                 case SO_REUSEPORT:
4905                         *((int *)optval) = sk->sk_reuseport;
4906                         break;
4907                 default:
4908                         goto err_clear;
4909                 }
4910 #ifdef CONFIG_INET
4911         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4912                 struct inet_connection_sock *icsk;
4913                 struct tcp_sock *tp;
4914
4915                 switch (optname) {
4916                 case TCP_CONGESTION:
4917                         icsk = inet_csk(sk);
4918
4919                         if (!icsk->icsk_ca_ops || optlen <= 1)
4920                                 goto err_clear;
4921                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4922                         optval[optlen - 1] = 0;
4923                         break;
4924                 case TCP_SAVED_SYN:
4925                         tp = tcp_sk(sk);
4926
4927                         if (optlen <= 0 || !tp->saved_syn ||
4928                             optlen > tcp_saved_syn_len(tp->saved_syn))
4929                                 goto err_clear;
4930                         memcpy(optval, tp->saved_syn->data, optlen);
4931                         break;
4932                 default:
4933                         goto err_clear;
4934                 }
4935         } else if (level == SOL_IP) {
4936                 struct inet_sock *inet = inet_sk(sk);
4937
4938                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4939                         goto err_clear;
4940
4941                 /* Only some options are supported */
4942                 switch (optname) {
4943                 case IP_TOS:
4944                         *((int *)optval) = (int)inet->tos;
4945                         break;
4946                 default:
4947                         goto err_clear;
4948                 }
4949 #if IS_ENABLED(CONFIG_IPV6)
4950         } else if (level == SOL_IPV6) {
4951                 struct ipv6_pinfo *np = inet6_sk(sk);
4952
4953                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4954                         goto err_clear;
4955
4956                 /* Only some options are supported */
4957                 switch (optname) {
4958                 case IPV6_TCLASS:
4959                         *((int *)optval) = (int)np->tclass;
4960                         break;
4961                 default:
4962                         goto err_clear;
4963                 }
4964 #endif
4965 #endif
4966         } else {
4967                 goto err_clear;
4968         }
4969         return 0;
4970 err_clear:
4971         memset(optval, 0, optlen);
4972         return -EINVAL;
4973 }
4974
4975 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4976            int, level, int, optname, char *, optval, int, optlen)
4977 {
4978         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
4979 }
4980
4981 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
4982         .func           = bpf_sock_addr_setsockopt,
4983         .gpl_only       = false,
4984         .ret_type       = RET_INTEGER,
4985         .arg1_type      = ARG_PTR_TO_CTX,
4986         .arg2_type      = ARG_ANYTHING,
4987         .arg3_type      = ARG_ANYTHING,
4988         .arg4_type      = ARG_PTR_TO_MEM,
4989         .arg5_type      = ARG_CONST_SIZE,
4990 };
4991
4992 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
4993            int, level, int, optname, char *, optval, int, optlen)
4994 {
4995         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
4996 }
4997
4998 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
4999         .func           = bpf_sock_addr_getsockopt,
5000         .gpl_only       = false,
5001         .ret_type       = RET_INTEGER,
5002         .arg1_type      = ARG_PTR_TO_CTX,
5003         .arg2_type      = ARG_ANYTHING,
5004         .arg3_type      = ARG_ANYTHING,
5005         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5006         .arg5_type      = ARG_CONST_SIZE,
5007 };
5008
5009 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5010            int, level, int, optname, char *, optval, int, optlen)
5011 {
5012         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5013 }
5014
5015 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5016         .func           = bpf_sock_ops_setsockopt,
5017         .gpl_only       = false,
5018         .ret_type       = RET_INTEGER,
5019         .arg1_type      = ARG_PTR_TO_CTX,
5020         .arg2_type      = ARG_ANYTHING,
5021         .arg3_type      = ARG_ANYTHING,
5022         .arg4_type      = ARG_PTR_TO_MEM,
5023         .arg5_type      = ARG_CONST_SIZE,
5024 };
5025
5026 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5027                                 int optname, const u8 **start)
5028 {
5029         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5030         const u8 *hdr_start;
5031         int ret;
5032
5033         if (syn_skb) {
5034                 /* sk is a request_sock here */
5035
5036                 if (optname == TCP_BPF_SYN) {
5037                         hdr_start = syn_skb->data;
5038                         ret = tcp_hdrlen(syn_skb);
5039                 } else if (optname == TCP_BPF_SYN_IP) {
5040                         hdr_start = skb_network_header(syn_skb);
5041                         ret = skb_network_header_len(syn_skb) +
5042                                 tcp_hdrlen(syn_skb);
5043                 } else {
5044                         /* optname == TCP_BPF_SYN_MAC */
5045                         hdr_start = skb_mac_header(syn_skb);
5046                         ret = skb_mac_header_len(syn_skb) +
5047                                 skb_network_header_len(syn_skb) +
5048                                 tcp_hdrlen(syn_skb);
5049                 }
5050         } else {
5051                 struct sock *sk = bpf_sock->sk;
5052                 struct saved_syn *saved_syn;
5053
5054                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5055                         /* synack retransmit. bpf_sock->syn_skb will
5056                          * not be available.  It has to resort to
5057                          * saved_syn (if it is saved).
5058                          */
5059                         saved_syn = inet_reqsk(sk)->saved_syn;
5060                 else
5061                         saved_syn = tcp_sk(sk)->saved_syn;
5062
5063                 if (!saved_syn)
5064                         return -ENOENT;
5065
5066                 if (optname == TCP_BPF_SYN) {
5067                         hdr_start = saved_syn->data +
5068                                 saved_syn->mac_hdrlen +
5069                                 saved_syn->network_hdrlen;
5070                         ret = saved_syn->tcp_hdrlen;
5071                 } else if (optname == TCP_BPF_SYN_IP) {
5072                         hdr_start = saved_syn->data +
5073                                 saved_syn->mac_hdrlen;
5074                         ret = saved_syn->network_hdrlen +
5075                                 saved_syn->tcp_hdrlen;
5076                 } else {
5077                         /* optname == TCP_BPF_SYN_MAC */
5078
5079                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5080                         if (!saved_syn->mac_hdrlen)
5081                                 return -ENOENT;
5082
5083                         hdr_start = saved_syn->data;
5084                         ret = saved_syn->mac_hdrlen +
5085                                 saved_syn->network_hdrlen +
5086                                 saved_syn->tcp_hdrlen;
5087                 }
5088         }
5089
5090         *start = hdr_start;
5091         return ret;
5092 }
5093
5094 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5095            int, level, int, optname, char *, optval, int, optlen)
5096 {
5097         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5098             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5099                 int ret, copy_len = 0;
5100                 const u8 *start;
5101
5102                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5103                 if (ret > 0) {
5104                         copy_len = ret;
5105                         if (optlen < copy_len) {
5106                                 copy_len = optlen;
5107                                 ret = -ENOSPC;
5108                         }
5109
5110                         memcpy(optval, start, copy_len);
5111                 }
5112
5113                 /* Zero out unused buffer at the end */
5114                 memset(optval + copy_len, 0, optlen - copy_len);
5115
5116                 return ret;
5117         }
5118
5119         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5120 }
5121
5122 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5123         .func           = bpf_sock_ops_getsockopt,
5124         .gpl_only       = false,
5125         .ret_type       = RET_INTEGER,
5126         .arg1_type      = ARG_PTR_TO_CTX,
5127         .arg2_type      = ARG_ANYTHING,
5128         .arg3_type      = ARG_ANYTHING,
5129         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5130         .arg5_type      = ARG_CONST_SIZE,
5131 };
5132
5133 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5134            int, argval)
5135 {
5136         struct sock *sk = bpf_sock->sk;
5137         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5138
5139         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5140                 return -EINVAL;
5141
5142         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5143
5144         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5145 }
5146
5147 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5148         .func           = bpf_sock_ops_cb_flags_set,
5149         .gpl_only       = false,
5150         .ret_type       = RET_INTEGER,
5151         .arg1_type      = ARG_PTR_TO_CTX,
5152         .arg2_type      = ARG_ANYTHING,
5153 };
5154
5155 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5156 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5157
5158 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5159            int, addr_len)
5160 {
5161 #ifdef CONFIG_INET
5162         struct sock *sk = ctx->sk;
5163         u32 flags = BIND_FROM_BPF;
5164         int err;
5165
5166         err = -EINVAL;
5167         if (addr_len < offsetofend(struct sockaddr, sa_family))
5168                 return err;
5169         if (addr->sa_family == AF_INET) {
5170                 if (addr_len < sizeof(struct sockaddr_in))
5171                         return err;
5172                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5173                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5174                 return __inet_bind(sk, addr, addr_len, flags);
5175 #if IS_ENABLED(CONFIG_IPV6)
5176         } else if (addr->sa_family == AF_INET6) {
5177                 if (addr_len < SIN6_LEN_RFC2133)
5178                         return err;
5179                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5180                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5181                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5182                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5183                  */
5184                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5185 #endif /* CONFIG_IPV6 */
5186         }
5187 #endif /* CONFIG_INET */
5188
5189         return -EAFNOSUPPORT;
5190 }
5191
5192 static const struct bpf_func_proto bpf_bind_proto = {
5193         .func           = bpf_bind,
5194         .gpl_only       = false,
5195         .ret_type       = RET_INTEGER,
5196         .arg1_type      = ARG_PTR_TO_CTX,
5197         .arg2_type      = ARG_PTR_TO_MEM,
5198         .arg3_type      = ARG_CONST_SIZE,
5199 };
5200
5201 #ifdef CONFIG_XFRM
5202 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5203            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5204 {
5205         const struct sec_path *sp = skb_sec_path(skb);
5206         const struct xfrm_state *x;
5207
5208         if (!sp || unlikely(index >= sp->len || flags))
5209                 goto err_clear;
5210
5211         x = sp->xvec[index];
5212
5213         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5214                 goto err_clear;
5215
5216         to->reqid = x->props.reqid;
5217         to->spi = x->id.spi;
5218         to->family = x->props.family;
5219         to->ext = 0;
5220
5221         if (to->family == AF_INET6) {
5222                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5223                        sizeof(to->remote_ipv6));
5224         } else {
5225                 to->remote_ipv4 = x->props.saddr.a4;
5226                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5227         }
5228
5229         return 0;
5230 err_clear:
5231         memset(to, 0, size);
5232         return -EINVAL;
5233 }
5234
5235 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5236         .func           = bpf_skb_get_xfrm_state,
5237         .gpl_only       = false,
5238         .ret_type       = RET_INTEGER,
5239         .arg1_type      = ARG_PTR_TO_CTX,
5240         .arg2_type      = ARG_ANYTHING,
5241         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5242         .arg4_type      = ARG_CONST_SIZE,
5243         .arg5_type      = ARG_ANYTHING,
5244 };
5245 #endif
5246
5247 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5248 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5249                                   const struct neighbour *neigh,
5250                                   const struct net_device *dev, u32 mtu)
5251 {
5252         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5253         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5254         params->h_vlan_TCI = 0;
5255         params->h_vlan_proto = 0;
5256         if (mtu)
5257                 params->mtu_result = mtu; /* union with tot_len */
5258
5259         return 0;
5260 }
5261 #endif
5262
5263 #if IS_ENABLED(CONFIG_INET)
5264 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5265                                u32 flags, bool check_mtu)
5266 {
5267         struct fib_nh_common *nhc;
5268         struct in_device *in_dev;
5269         struct neighbour *neigh;
5270         struct net_device *dev;
5271         struct fib_result res;
5272         struct flowi4 fl4;
5273         u32 mtu = 0;
5274         int err;
5275
5276         dev = dev_get_by_index_rcu(net, params->ifindex);
5277         if (unlikely(!dev))
5278                 return -ENODEV;
5279
5280         /* verify forwarding is enabled on this interface */
5281         in_dev = __in_dev_get_rcu(dev);
5282         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5283                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5284
5285         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5286                 fl4.flowi4_iif = 1;
5287                 fl4.flowi4_oif = params->ifindex;
5288         } else {
5289                 fl4.flowi4_iif = params->ifindex;
5290                 fl4.flowi4_oif = 0;
5291         }
5292         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5293         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5294         fl4.flowi4_flags = 0;
5295
5296         fl4.flowi4_proto = params->l4_protocol;
5297         fl4.daddr = params->ipv4_dst;
5298         fl4.saddr = params->ipv4_src;
5299         fl4.fl4_sport = params->sport;
5300         fl4.fl4_dport = params->dport;
5301         fl4.flowi4_multipath_hash = 0;
5302
5303         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5304                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5305                 struct fib_table *tb;
5306
5307                 tb = fib_get_table(net, tbid);
5308                 if (unlikely(!tb))
5309                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5310
5311                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5312         } else {
5313                 fl4.flowi4_mark = 0;
5314                 fl4.flowi4_secid = 0;
5315                 fl4.flowi4_tun_key.tun_id = 0;
5316                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5317
5318                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5319         }
5320
5321         if (err) {
5322                 /* map fib lookup errors to RTN_ type */
5323                 if (err == -EINVAL)
5324                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5325                 if (err == -EHOSTUNREACH)
5326                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5327                 if (err == -EACCES)
5328                         return BPF_FIB_LKUP_RET_PROHIBIT;
5329
5330                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5331         }
5332
5333         if (res.type != RTN_UNICAST)
5334                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5335
5336         if (fib_info_num_path(res.fi) > 1)
5337                 fib_select_path(net, &res, &fl4, NULL);
5338
5339         if (check_mtu) {
5340                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5341                 if (params->tot_len > mtu) {
5342                         params->mtu_result = mtu; /* union with tot_len */
5343                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5344                 }
5345         }
5346
5347         nhc = res.nhc;
5348
5349         /* do not handle lwt encaps right now */
5350         if (nhc->nhc_lwtstate)
5351                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5352
5353         dev = nhc->nhc_dev;
5354
5355         params->rt_metric = res.fi->fib_priority;
5356         params->ifindex = dev->ifindex;
5357
5358         /* xdp and cls_bpf programs are run in RCU-bh so
5359          * rcu_read_lock_bh is not needed here
5360          */
5361         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5362                 if (nhc->nhc_gw_family)
5363                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5364
5365                 neigh = __ipv4_neigh_lookup_noref(dev,
5366                                                  (__force u32)params->ipv4_dst);
5367         } else {
5368                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5369
5370                 params->family = AF_INET6;
5371                 *dst = nhc->nhc_gw.ipv6;
5372                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5373         }
5374
5375         if (!neigh)
5376                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5377
5378         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5379 }
5380 #endif
5381
5382 #if IS_ENABLED(CONFIG_IPV6)
5383 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5384                                u32 flags, bool check_mtu)
5385 {
5386         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5387         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5388         struct fib6_result res = {};
5389         struct neighbour *neigh;
5390         struct net_device *dev;
5391         struct inet6_dev *idev;
5392         struct flowi6 fl6;
5393         int strict = 0;
5394         int oif, err;
5395         u32 mtu = 0;
5396
5397         /* link local addresses are never forwarded */
5398         if (rt6_need_strict(dst) || rt6_need_strict(src))
5399                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5400
5401         dev = dev_get_by_index_rcu(net, params->ifindex);
5402         if (unlikely(!dev))
5403                 return -ENODEV;
5404
5405         idev = __in6_dev_get_safely(dev);
5406         if (unlikely(!idev || !idev->cnf.forwarding))
5407                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5408
5409         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5410                 fl6.flowi6_iif = 1;
5411                 oif = fl6.flowi6_oif = params->ifindex;
5412         } else {
5413                 oif = fl6.flowi6_iif = params->ifindex;
5414                 fl6.flowi6_oif = 0;
5415                 strict = RT6_LOOKUP_F_HAS_SADDR;
5416         }
5417         fl6.flowlabel = params->flowinfo;
5418         fl6.flowi6_scope = 0;
5419         fl6.flowi6_flags = 0;
5420         fl6.mp_hash = 0;
5421
5422         fl6.flowi6_proto = params->l4_protocol;
5423         fl6.daddr = *dst;
5424         fl6.saddr = *src;
5425         fl6.fl6_sport = params->sport;
5426         fl6.fl6_dport = params->dport;
5427
5428         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5429                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5430                 struct fib6_table *tb;
5431
5432                 tb = ipv6_stub->fib6_get_table(net, tbid);
5433                 if (unlikely(!tb))
5434                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5435
5436                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5437                                                    strict);
5438         } else {
5439                 fl6.flowi6_mark = 0;
5440                 fl6.flowi6_secid = 0;
5441                 fl6.flowi6_tun_key.tun_id = 0;
5442                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5443
5444                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5445         }
5446
5447         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5448                      res.f6i == net->ipv6.fib6_null_entry))
5449                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5450
5451         switch (res.fib6_type) {
5452         /* only unicast is forwarded */
5453         case RTN_UNICAST:
5454                 break;
5455         case RTN_BLACKHOLE:
5456                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5457         case RTN_UNREACHABLE:
5458                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5459         case RTN_PROHIBIT:
5460                 return BPF_FIB_LKUP_RET_PROHIBIT;
5461         default:
5462                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5463         }
5464
5465         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5466                                     fl6.flowi6_oif != 0, NULL, strict);
5467
5468         if (check_mtu) {
5469                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5470                 if (params->tot_len > mtu) {
5471                         params->mtu_result = mtu; /* union with tot_len */
5472                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5473                 }
5474         }
5475
5476         if (res.nh->fib_nh_lws)
5477                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5478
5479         if (res.nh->fib_nh_gw_family)
5480                 *dst = res.nh->fib_nh_gw6;
5481
5482         dev = res.nh->fib_nh_dev;
5483         params->rt_metric = res.f6i->fib6_metric;
5484         params->ifindex = dev->ifindex;
5485
5486         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5487          * not needed here.
5488          */
5489         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5490         if (!neigh)
5491                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5492
5493         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5494 }
5495 #endif
5496
5497 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5498            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5499 {
5500         if (plen < sizeof(*params))
5501                 return -EINVAL;
5502
5503         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5504                 return -EINVAL;
5505
5506         switch (params->family) {
5507 #if IS_ENABLED(CONFIG_INET)
5508         case AF_INET:
5509                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5510                                            flags, true);
5511 #endif
5512 #if IS_ENABLED(CONFIG_IPV6)
5513         case AF_INET6:
5514                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5515                                            flags, true);
5516 #endif
5517         }
5518         return -EAFNOSUPPORT;
5519 }
5520
5521 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5522         .func           = bpf_xdp_fib_lookup,
5523         .gpl_only       = true,
5524         .ret_type       = RET_INTEGER,
5525         .arg1_type      = ARG_PTR_TO_CTX,
5526         .arg2_type      = ARG_PTR_TO_MEM,
5527         .arg3_type      = ARG_CONST_SIZE,
5528         .arg4_type      = ARG_ANYTHING,
5529 };
5530
5531 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5532            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5533 {
5534         struct net *net = dev_net(skb->dev);
5535         int rc = -EAFNOSUPPORT;
5536         bool check_mtu = false;
5537
5538         if (plen < sizeof(*params))
5539                 return -EINVAL;
5540
5541         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5542                 return -EINVAL;
5543
5544         if (params->tot_len)
5545                 check_mtu = true;
5546
5547         switch (params->family) {
5548 #if IS_ENABLED(CONFIG_INET)
5549         case AF_INET:
5550                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5551                 break;
5552 #endif
5553 #if IS_ENABLED(CONFIG_IPV6)
5554         case AF_INET6:
5555                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5556                 break;
5557 #endif
5558         }
5559
5560         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5561                 struct net_device *dev;
5562
5563                 /* When tot_len isn't provided by user, check skb
5564                  * against MTU of FIB lookup resulting net_device
5565                  */
5566                 dev = dev_get_by_index_rcu(net, params->ifindex);
5567                 if (!is_skb_forwardable(dev, skb))
5568                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5569
5570                 params->mtu_result = dev->mtu; /* union with tot_len */
5571         }
5572
5573         return rc;
5574 }
5575
5576 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5577         .func           = bpf_skb_fib_lookup,
5578         .gpl_only       = true,
5579         .ret_type       = RET_INTEGER,
5580         .arg1_type      = ARG_PTR_TO_CTX,
5581         .arg2_type      = ARG_PTR_TO_MEM,
5582         .arg3_type      = ARG_CONST_SIZE,
5583         .arg4_type      = ARG_ANYTHING,
5584 };
5585
5586 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5587                                             u32 ifindex)
5588 {
5589         struct net *netns = dev_net(dev_curr);
5590
5591         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5592         if (ifindex == 0)
5593                 return dev_curr;
5594
5595         return dev_get_by_index_rcu(netns, ifindex);
5596 }
5597
5598 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5599            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5600 {
5601         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5602         struct net_device *dev = skb->dev;
5603         int skb_len, dev_len;
5604         int mtu;
5605
5606         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5607                 return -EINVAL;
5608
5609         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5610                 return -EINVAL;
5611
5612         dev = __dev_via_ifindex(dev, ifindex);
5613         if (unlikely(!dev))
5614                 return -ENODEV;
5615
5616         mtu = READ_ONCE(dev->mtu);
5617
5618         dev_len = mtu + dev->hard_header_len;
5619
5620         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5621         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5622
5623         skb_len += len_diff; /* minus result pass check */
5624         if (skb_len <= dev_len) {
5625                 ret = BPF_MTU_CHK_RET_SUCCESS;
5626                 goto out;
5627         }
5628         /* At this point, skb->len exceed MTU, but as it include length of all
5629          * segments, it can still be below MTU.  The SKB can possibly get
5630          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5631          * must choose if segs are to be MTU checked.
5632          */
5633         if (skb_is_gso(skb)) {
5634                 ret = BPF_MTU_CHK_RET_SUCCESS;
5635
5636                 if (flags & BPF_MTU_CHK_SEGS &&
5637                     !skb_gso_validate_network_len(skb, mtu))
5638                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5639         }
5640 out:
5641         /* BPF verifier guarantees valid pointer */
5642         *mtu_len = mtu;
5643
5644         return ret;
5645 }
5646
5647 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5648            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5649 {
5650         struct net_device *dev = xdp->rxq->dev;
5651         int xdp_len = xdp->data_end - xdp->data;
5652         int ret = BPF_MTU_CHK_RET_SUCCESS;
5653         int mtu, dev_len;
5654
5655         /* XDP variant doesn't support multi-buffer segment check (yet) */
5656         if (unlikely(flags))
5657                 return -EINVAL;
5658
5659         dev = __dev_via_ifindex(dev, ifindex);
5660         if (unlikely(!dev))
5661                 return -ENODEV;
5662
5663         mtu = READ_ONCE(dev->mtu);
5664
5665         /* Add L2-header as dev MTU is L3 size */
5666         dev_len = mtu + dev->hard_header_len;
5667
5668         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5669         if (*mtu_len)
5670                 xdp_len = *mtu_len + dev->hard_header_len;
5671
5672         xdp_len += len_diff; /* minus result pass check */
5673         if (xdp_len > dev_len)
5674                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5675
5676         /* BPF verifier guarantees valid pointer */
5677         *mtu_len = mtu;
5678
5679         return ret;
5680 }
5681
5682 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5683         .func           = bpf_skb_check_mtu,
5684         .gpl_only       = true,
5685         .ret_type       = RET_INTEGER,
5686         .arg1_type      = ARG_PTR_TO_CTX,
5687         .arg2_type      = ARG_ANYTHING,
5688         .arg3_type      = ARG_PTR_TO_INT,
5689         .arg4_type      = ARG_ANYTHING,
5690         .arg5_type      = ARG_ANYTHING,
5691 };
5692
5693 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5694         .func           = bpf_xdp_check_mtu,
5695         .gpl_only       = true,
5696         .ret_type       = RET_INTEGER,
5697         .arg1_type      = ARG_PTR_TO_CTX,
5698         .arg2_type      = ARG_ANYTHING,
5699         .arg3_type      = ARG_PTR_TO_INT,
5700         .arg4_type      = ARG_ANYTHING,
5701         .arg5_type      = ARG_ANYTHING,
5702 };
5703
5704 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5705 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5706 {
5707         int err;
5708         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5709
5710         if (!seg6_validate_srh(srh, len, false))
5711                 return -EINVAL;
5712
5713         switch (type) {
5714         case BPF_LWT_ENCAP_SEG6_INLINE:
5715                 if (skb->protocol != htons(ETH_P_IPV6))
5716                         return -EBADMSG;
5717
5718                 err = seg6_do_srh_inline(skb, srh);
5719                 break;
5720         case BPF_LWT_ENCAP_SEG6:
5721                 skb_reset_inner_headers(skb);
5722                 skb->encapsulation = 1;
5723                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5724                 break;
5725         default:
5726                 return -EINVAL;
5727         }
5728
5729         bpf_compute_data_pointers(skb);
5730         if (err)
5731                 return err;
5732
5733         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5734         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5735
5736         return seg6_lookup_nexthop(skb, NULL, 0);
5737 }
5738 #endif /* CONFIG_IPV6_SEG6_BPF */
5739
5740 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5741 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5742                              bool ingress)
5743 {
5744         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5745 }
5746 #endif
5747
5748 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5749            u32, len)
5750 {
5751         switch (type) {
5752 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5753         case BPF_LWT_ENCAP_SEG6:
5754         case BPF_LWT_ENCAP_SEG6_INLINE:
5755                 return bpf_push_seg6_encap(skb, type, hdr, len);
5756 #endif
5757 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5758         case BPF_LWT_ENCAP_IP:
5759                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5760 #endif
5761         default:
5762                 return -EINVAL;
5763         }
5764 }
5765
5766 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5767            void *, hdr, u32, len)
5768 {
5769         switch (type) {
5770 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5771         case BPF_LWT_ENCAP_IP:
5772                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5773 #endif
5774         default:
5775                 return -EINVAL;
5776         }
5777 }
5778
5779 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5780         .func           = bpf_lwt_in_push_encap,
5781         .gpl_only       = false,
5782         .ret_type       = RET_INTEGER,
5783         .arg1_type      = ARG_PTR_TO_CTX,
5784         .arg2_type      = ARG_ANYTHING,
5785         .arg3_type      = ARG_PTR_TO_MEM,
5786         .arg4_type      = ARG_CONST_SIZE
5787 };
5788
5789 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5790         .func           = bpf_lwt_xmit_push_encap,
5791         .gpl_only       = false,
5792         .ret_type       = RET_INTEGER,
5793         .arg1_type      = ARG_PTR_TO_CTX,
5794         .arg2_type      = ARG_ANYTHING,
5795         .arg3_type      = ARG_PTR_TO_MEM,
5796         .arg4_type      = ARG_CONST_SIZE
5797 };
5798
5799 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5800 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5801            const void *, from, u32, len)
5802 {
5803         struct seg6_bpf_srh_state *srh_state =
5804                 this_cpu_ptr(&seg6_bpf_srh_states);
5805         struct ipv6_sr_hdr *srh = srh_state->srh;
5806         void *srh_tlvs, *srh_end, *ptr;
5807         int srhoff = 0;
5808
5809         if (srh == NULL)
5810                 return -EINVAL;
5811
5812         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5813         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5814
5815         ptr = skb->data + offset;
5816         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5817                 srh_state->valid = false;
5818         else if (ptr < (void *)&srh->flags ||
5819                  ptr + len > (void *)&srh->segments)
5820                 return -EFAULT;
5821
5822         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5823                 return -EFAULT;
5824         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5825                 return -EINVAL;
5826         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5827
5828         memcpy(skb->data + offset, from, len);
5829         return 0;
5830 }
5831
5832 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5833         .func           = bpf_lwt_seg6_store_bytes,
5834         .gpl_only       = false,
5835         .ret_type       = RET_INTEGER,
5836         .arg1_type      = ARG_PTR_TO_CTX,
5837         .arg2_type      = ARG_ANYTHING,
5838         .arg3_type      = ARG_PTR_TO_MEM,
5839         .arg4_type      = ARG_CONST_SIZE
5840 };
5841
5842 static void bpf_update_srh_state(struct sk_buff *skb)
5843 {
5844         struct seg6_bpf_srh_state *srh_state =
5845                 this_cpu_ptr(&seg6_bpf_srh_states);
5846         int srhoff = 0;
5847
5848         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5849                 srh_state->srh = NULL;
5850         } else {
5851                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5852                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5853                 srh_state->valid = true;
5854         }
5855 }
5856
5857 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5858            u32, action, void *, param, u32, param_len)
5859 {
5860         struct seg6_bpf_srh_state *srh_state =
5861                 this_cpu_ptr(&seg6_bpf_srh_states);
5862         int hdroff = 0;
5863         int err;
5864
5865         switch (action) {
5866         case SEG6_LOCAL_ACTION_END_X:
5867                 if (!seg6_bpf_has_valid_srh(skb))
5868                         return -EBADMSG;
5869                 if (param_len != sizeof(struct in6_addr))
5870                         return -EINVAL;
5871                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5872         case SEG6_LOCAL_ACTION_END_T:
5873                 if (!seg6_bpf_has_valid_srh(skb))
5874                         return -EBADMSG;
5875                 if (param_len != sizeof(int))
5876                         return -EINVAL;
5877                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5878         case SEG6_LOCAL_ACTION_END_DT6:
5879                 if (!seg6_bpf_has_valid_srh(skb))
5880                         return -EBADMSG;
5881                 if (param_len != sizeof(int))
5882                         return -EINVAL;
5883
5884                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5885                         return -EBADMSG;
5886                 if (!pskb_pull(skb, hdroff))
5887                         return -EBADMSG;
5888
5889                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5890                 skb_reset_network_header(skb);
5891                 skb_reset_transport_header(skb);
5892                 skb->encapsulation = 0;
5893
5894                 bpf_compute_data_pointers(skb);
5895                 bpf_update_srh_state(skb);
5896                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5897         case SEG6_LOCAL_ACTION_END_B6:
5898                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5899                         return -EBADMSG;
5900                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5901                                           param, param_len);
5902                 if (!err)
5903                         bpf_update_srh_state(skb);
5904
5905                 return err;
5906         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5907                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5908                         return -EBADMSG;
5909                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5910                                           param, param_len);
5911                 if (!err)
5912                         bpf_update_srh_state(skb);
5913
5914                 return err;
5915         default:
5916                 return -EINVAL;
5917         }
5918 }
5919
5920 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5921         .func           = bpf_lwt_seg6_action,
5922         .gpl_only       = false,
5923         .ret_type       = RET_INTEGER,
5924         .arg1_type      = ARG_PTR_TO_CTX,
5925         .arg2_type      = ARG_ANYTHING,
5926         .arg3_type      = ARG_PTR_TO_MEM,
5927         .arg4_type      = ARG_CONST_SIZE
5928 };
5929
5930 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5931            s32, len)
5932 {
5933         struct seg6_bpf_srh_state *srh_state =
5934                 this_cpu_ptr(&seg6_bpf_srh_states);
5935         struct ipv6_sr_hdr *srh = srh_state->srh;
5936         void *srh_end, *srh_tlvs, *ptr;
5937         struct ipv6hdr *hdr;
5938         int srhoff = 0;
5939         int ret;
5940
5941         if (unlikely(srh == NULL))
5942                 return -EINVAL;
5943
5944         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5945                         ((srh->first_segment + 1) << 4));
5946         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5947                         srh_state->hdrlen);
5948         ptr = skb->data + offset;
5949
5950         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5951                 return -EFAULT;
5952         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5953                 return -EFAULT;
5954
5955         if (len > 0) {
5956                 ret = skb_cow_head(skb, len);
5957                 if (unlikely(ret < 0))
5958                         return ret;
5959
5960                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5961         } else {
5962                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5963         }
5964
5965         bpf_compute_data_pointers(skb);
5966         if (unlikely(ret < 0))
5967                 return ret;
5968
5969         hdr = (struct ipv6hdr *)skb->data;
5970         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5971
5972         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5973                 return -EINVAL;
5974         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5975         srh_state->hdrlen += len;
5976         srh_state->valid = false;
5977         return 0;
5978 }
5979
5980 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5981         .func           = bpf_lwt_seg6_adjust_srh,
5982         .gpl_only       = false,
5983         .ret_type       = RET_INTEGER,
5984         .arg1_type      = ARG_PTR_TO_CTX,
5985         .arg2_type      = ARG_ANYTHING,
5986         .arg3_type      = ARG_ANYTHING,
5987 };
5988 #endif /* CONFIG_IPV6_SEG6_BPF */
5989
5990 #ifdef CONFIG_INET
5991 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5992                               int dif, int sdif, u8 family, u8 proto)
5993 {
5994         bool refcounted = false;
5995         struct sock *sk = NULL;
5996
5997         if (family == AF_INET) {
5998                 __be32 src4 = tuple->ipv4.saddr;
5999                 __be32 dst4 = tuple->ipv4.daddr;
6000
6001                 if (proto == IPPROTO_TCP)
6002                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6003                                            src4, tuple->ipv4.sport,
6004                                            dst4, tuple->ipv4.dport,
6005                                            dif, sdif, &refcounted);
6006                 else
6007                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6008                                                dst4, tuple->ipv4.dport,
6009                                                dif, sdif, &udp_table, NULL);
6010 #if IS_ENABLED(CONFIG_IPV6)
6011         } else {
6012                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6013                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6014
6015                 if (proto == IPPROTO_TCP)
6016                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6017                                             src6, tuple->ipv6.sport,
6018                                             dst6, ntohs(tuple->ipv6.dport),
6019                                             dif, sdif, &refcounted);
6020                 else if (likely(ipv6_bpf_stub))
6021                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6022                                                             src6, tuple->ipv6.sport,
6023                                                             dst6, tuple->ipv6.dport,
6024                                                             dif, sdif,
6025                                                             &udp_table, NULL);
6026 #endif
6027         }
6028
6029         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6030                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6031                 sk = NULL;
6032         }
6033         return sk;
6034 }
6035
6036 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6037  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6038  * Returns the socket as an 'unsigned long' to simplify the casting in the
6039  * callers to satisfy BPF_CALL declarations.
6040  */
6041 static struct sock *
6042 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6043                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6044                  u64 flags)
6045 {
6046         struct sock *sk = NULL;
6047         u8 family = AF_UNSPEC;
6048         struct net *net;
6049         int sdif;
6050
6051         if (len == sizeof(tuple->ipv4))
6052                 family = AF_INET;
6053         else if (len == sizeof(tuple->ipv6))
6054                 family = AF_INET6;
6055         else
6056                 return NULL;
6057
6058         if (unlikely(family == AF_UNSPEC || flags ||
6059                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6060                 goto out;
6061
6062         if (family == AF_INET)
6063                 sdif = inet_sdif(skb);
6064         else
6065                 sdif = inet6_sdif(skb);
6066
6067         if ((s32)netns_id < 0) {
6068                 net = caller_net;
6069                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6070         } else {
6071                 net = get_net_ns_by_id(caller_net, netns_id);
6072                 if (unlikely(!net))
6073                         goto out;
6074                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6075                 put_net(net);
6076         }
6077
6078 out:
6079         return sk;
6080 }
6081
6082 static struct sock *
6083 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6084                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6085                 u64 flags)
6086 {
6087         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6088                                            ifindex, proto, netns_id, flags);
6089
6090         if (sk) {
6091                 sk = sk_to_full_sk(sk);
6092                 if (!sk_fullsock(sk)) {
6093                         sock_gen_put(sk);
6094                         return NULL;
6095                 }
6096         }
6097
6098         return sk;
6099 }
6100
6101 static struct sock *
6102 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6103                u8 proto, u64 netns_id, u64 flags)
6104 {
6105         struct net *caller_net;
6106         int ifindex;
6107
6108         if (skb->dev) {
6109                 caller_net = dev_net(skb->dev);
6110                 ifindex = skb->dev->ifindex;
6111         } else {
6112                 caller_net = sock_net(skb->sk);
6113                 ifindex = 0;
6114         }
6115
6116         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6117                                 netns_id, flags);
6118 }
6119
6120 static struct sock *
6121 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6122               u8 proto, u64 netns_id, u64 flags)
6123 {
6124         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6125                                          flags);
6126
6127         if (sk) {
6128                 sk = sk_to_full_sk(sk);
6129                 if (!sk_fullsock(sk)) {
6130                         sock_gen_put(sk);
6131                         return NULL;
6132                 }
6133         }
6134
6135         return sk;
6136 }
6137
6138 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6139            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6140 {
6141         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6142                                              netns_id, flags);
6143 }
6144
6145 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6146         .func           = bpf_skc_lookup_tcp,
6147         .gpl_only       = false,
6148         .pkt_access     = true,
6149         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6150         .arg1_type      = ARG_PTR_TO_CTX,
6151         .arg2_type      = ARG_PTR_TO_MEM,
6152         .arg3_type      = ARG_CONST_SIZE,
6153         .arg4_type      = ARG_ANYTHING,
6154         .arg5_type      = ARG_ANYTHING,
6155 };
6156
6157 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6158            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6159 {
6160         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6161                                             netns_id, flags);
6162 }
6163
6164 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6165         .func           = bpf_sk_lookup_tcp,
6166         .gpl_only       = false,
6167         .pkt_access     = true,
6168         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6169         .arg1_type      = ARG_PTR_TO_CTX,
6170         .arg2_type      = ARG_PTR_TO_MEM,
6171         .arg3_type      = ARG_CONST_SIZE,
6172         .arg4_type      = ARG_ANYTHING,
6173         .arg5_type      = ARG_ANYTHING,
6174 };
6175
6176 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6177            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6178 {
6179         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6180                                             netns_id, flags);
6181 }
6182
6183 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6184         .func           = bpf_sk_lookup_udp,
6185         .gpl_only       = false,
6186         .pkt_access     = true,
6187         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6188         .arg1_type      = ARG_PTR_TO_CTX,
6189         .arg2_type      = ARG_PTR_TO_MEM,
6190         .arg3_type      = ARG_CONST_SIZE,
6191         .arg4_type      = ARG_ANYTHING,
6192         .arg5_type      = ARG_ANYTHING,
6193 };
6194
6195 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6196 {
6197         if (sk && sk_is_refcounted(sk))
6198                 sock_gen_put(sk);
6199         return 0;
6200 }
6201
6202 static const struct bpf_func_proto bpf_sk_release_proto = {
6203         .func           = bpf_sk_release,
6204         .gpl_only       = false,
6205         .ret_type       = RET_INTEGER,
6206         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6207 };
6208
6209 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6210            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6211 {
6212         struct net *caller_net = dev_net(ctx->rxq->dev);
6213         int ifindex = ctx->rxq->dev->ifindex;
6214
6215         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6216                                               ifindex, IPPROTO_UDP, netns_id,
6217                                               flags);
6218 }
6219
6220 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6221         .func           = bpf_xdp_sk_lookup_udp,
6222         .gpl_only       = false,
6223         .pkt_access     = true,
6224         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6225         .arg1_type      = ARG_PTR_TO_CTX,
6226         .arg2_type      = ARG_PTR_TO_MEM,
6227         .arg3_type      = ARG_CONST_SIZE,
6228         .arg4_type      = ARG_ANYTHING,
6229         .arg5_type      = ARG_ANYTHING,
6230 };
6231
6232 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6233            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6234 {
6235         struct net *caller_net = dev_net(ctx->rxq->dev);
6236         int ifindex = ctx->rxq->dev->ifindex;
6237
6238         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6239                                                ifindex, IPPROTO_TCP, netns_id,
6240                                                flags);
6241 }
6242
6243 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6244         .func           = bpf_xdp_skc_lookup_tcp,
6245         .gpl_only       = false,
6246         .pkt_access     = true,
6247         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6248         .arg1_type      = ARG_PTR_TO_CTX,
6249         .arg2_type      = ARG_PTR_TO_MEM,
6250         .arg3_type      = ARG_CONST_SIZE,
6251         .arg4_type      = ARG_ANYTHING,
6252         .arg5_type      = ARG_ANYTHING,
6253 };
6254
6255 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6256            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6257 {
6258         struct net *caller_net = dev_net(ctx->rxq->dev);
6259         int ifindex = ctx->rxq->dev->ifindex;
6260
6261         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6262                                               ifindex, IPPROTO_TCP, netns_id,
6263                                               flags);
6264 }
6265
6266 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6267         .func           = bpf_xdp_sk_lookup_tcp,
6268         .gpl_only       = false,
6269         .pkt_access     = true,
6270         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6271         .arg1_type      = ARG_PTR_TO_CTX,
6272         .arg2_type      = ARG_PTR_TO_MEM,
6273         .arg3_type      = ARG_CONST_SIZE,
6274         .arg4_type      = ARG_ANYTHING,
6275         .arg5_type      = ARG_ANYTHING,
6276 };
6277
6278 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6279            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6280 {
6281         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6282                                                sock_net(ctx->sk), 0,
6283                                                IPPROTO_TCP, netns_id, flags);
6284 }
6285
6286 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6287         .func           = bpf_sock_addr_skc_lookup_tcp,
6288         .gpl_only       = false,
6289         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6290         .arg1_type      = ARG_PTR_TO_CTX,
6291         .arg2_type      = ARG_PTR_TO_MEM,
6292         .arg3_type      = ARG_CONST_SIZE,
6293         .arg4_type      = ARG_ANYTHING,
6294         .arg5_type      = ARG_ANYTHING,
6295 };
6296
6297 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6298            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6299 {
6300         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6301                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6302                                               netns_id, flags);
6303 }
6304
6305 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6306         .func           = bpf_sock_addr_sk_lookup_tcp,
6307         .gpl_only       = false,
6308         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6309         .arg1_type      = ARG_PTR_TO_CTX,
6310         .arg2_type      = ARG_PTR_TO_MEM,
6311         .arg3_type      = ARG_CONST_SIZE,
6312         .arg4_type      = ARG_ANYTHING,
6313         .arg5_type      = ARG_ANYTHING,
6314 };
6315
6316 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6317            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6318 {
6319         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6320                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6321                                               netns_id, flags);
6322 }
6323
6324 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6325         .func           = bpf_sock_addr_sk_lookup_udp,
6326         .gpl_only       = false,
6327         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6328         .arg1_type      = ARG_PTR_TO_CTX,
6329         .arg2_type      = ARG_PTR_TO_MEM,
6330         .arg3_type      = ARG_CONST_SIZE,
6331         .arg4_type      = ARG_ANYTHING,
6332         .arg5_type      = ARG_ANYTHING,
6333 };
6334
6335 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6336                                   struct bpf_insn_access_aux *info)
6337 {
6338         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6339                                           icsk_retransmits))
6340                 return false;
6341
6342         if (off % size != 0)
6343                 return false;
6344
6345         switch (off) {
6346         case offsetof(struct bpf_tcp_sock, bytes_received):
6347         case offsetof(struct bpf_tcp_sock, bytes_acked):
6348                 return size == sizeof(__u64);
6349         default:
6350                 return size == sizeof(__u32);
6351         }
6352 }
6353
6354 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6355                                     const struct bpf_insn *si,
6356                                     struct bpf_insn *insn_buf,
6357                                     struct bpf_prog *prog, u32 *target_size)
6358 {
6359         struct bpf_insn *insn = insn_buf;
6360
6361 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6362         do {                                                            \
6363                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6364                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6365                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6366                                       si->dst_reg, si->src_reg,         \
6367                                       offsetof(struct tcp_sock, FIELD)); \
6368         } while (0)
6369
6370 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6371         do {                                                            \
6372                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6373                                           FIELD) >                      \
6374                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6375                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6376                                         struct inet_connection_sock,    \
6377                                         FIELD),                         \
6378                                       si->dst_reg, si->src_reg,         \
6379                                       offsetof(                         \
6380                                         struct inet_connection_sock,    \
6381                                         FIELD));                        \
6382         } while (0)
6383
6384         if (insn > insn_buf)
6385                 return insn - insn_buf;
6386
6387         switch (si->off) {
6388         case offsetof(struct bpf_tcp_sock, rtt_min):
6389                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6390                              sizeof(struct minmax));
6391                 BUILD_BUG_ON(sizeof(struct minmax) <
6392                              sizeof(struct minmax_sample));
6393
6394                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6395                                       offsetof(struct tcp_sock, rtt_min) +
6396                                       offsetof(struct minmax_sample, v));
6397                 break;
6398         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6399                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6400                 break;
6401         case offsetof(struct bpf_tcp_sock, srtt_us):
6402                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6403                 break;
6404         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6405                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6406                 break;
6407         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6408                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6409                 break;
6410         case offsetof(struct bpf_tcp_sock, snd_nxt):
6411                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6412                 break;
6413         case offsetof(struct bpf_tcp_sock, snd_una):
6414                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6415                 break;
6416         case offsetof(struct bpf_tcp_sock, mss_cache):
6417                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6418                 break;
6419         case offsetof(struct bpf_tcp_sock, ecn_flags):
6420                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6421                 break;
6422         case offsetof(struct bpf_tcp_sock, rate_delivered):
6423                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6424                 break;
6425         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6426                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6427                 break;
6428         case offsetof(struct bpf_tcp_sock, packets_out):
6429                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6430                 break;
6431         case offsetof(struct bpf_tcp_sock, retrans_out):
6432                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6433                 break;
6434         case offsetof(struct bpf_tcp_sock, total_retrans):
6435                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6436                 break;
6437         case offsetof(struct bpf_tcp_sock, segs_in):
6438                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6439                 break;
6440         case offsetof(struct bpf_tcp_sock, data_segs_in):
6441                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6442                 break;
6443         case offsetof(struct bpf_tcp_sock, segs_out):
6444                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6445                 break;
6446         case offsetof(struct bpf_tcp_sock, data_segs_out):
6447                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6448                 break;
6449         case offsetof(struct bpf_tcp_sock, lost_out):
6450                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6451                 break;
6452         case offsetof(struct bpf_tcp_sock, sacked_out):
6453                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6454                 break;
6455         case offsetof(struct bpf_tcp_sock, bytes_received):
6456                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6457                 break;
6458         case offsetof(struct bpf_tcp_sock, bytes_acked):
6459                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6460                 break;
6461         case offsetof(struct bpf_tcp_sock, dsack_dups):
6462                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6463                 break;
6464         case offsetof(struct bpf_tcp_sock, delivered):
6465                 BPF_TCP_SOCK_GET_COMMON(delivered);
6466                 break;
6467         case offsetof(struct bpf_tcp_sock, delivered_ce):
6468                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6469                 break;
6470         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6471                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6472                 break;
6473         }
6474
6475         return insn - insn_buf;
6476 }
6477
6478 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6479 {
6480         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6481                 return (unsigned long)sk;
6482
6483         return (unsigned long)NULL;
6484 }
6485
6486 const struct bpf_func_proto bpf_tcp_sock_proto = {
6487         .func           = bpf_tcp_sock,
6488         .gpl_only       = false,
6489         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6490         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6491 };
6492
6493 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6494 {
6495         sk = sk_to_full_sk(sk);
6496
6497         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6498                 return (unsigned long)sk;
6499
6500         return (unsigned long)NULL;
6501 }
6502
6503 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6504         .func           = bpf_get_listener_sock,
6505         .gpl_only       = false,
6506         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6507         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6508 };
6509
6510 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6511 {
6512         unsigned int iphdr_len;
6513
6514         switch (skb_protocol(skb, true)) {
6515         case cpu_to_be16(ETH_P_IP):
6516                 iphdr_len = sizeof(struct iphdr);
6517                 break;
6518         case cpu_to_be16(ETH_P_IPV6):
6519                 iphdr_len = sizeof(struct ipv6hdr);
6520                 break;
6521         default:
6522                 return 0;
6523         }
6524
6525         if (skb_headlen(skb) < iphdr_len)
6526                 return 0;
6527
6528         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6529                 return 0;
6530
6531         return INET_ECN_set_ce(skb);
6532 }
6533
6534 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6535                                   struct bpf_insn_access_aux *info)
6536 {
6537         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6538                 return false;
6539
6540         if (off % size != 0)
6541                 return false;
6542
6543         switch (off) {
6544         default:
6545                 return size == sizeof(__u32);
6546         }
6547 }
6548
6549 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6550                                     const struct bpf_insn *si,
6551                                     struct bpf_insn *insn_buf,
6552                                     struct bpf_prog *prog, u32 *target_size)
6553 {
6554         struct bpf_insn *insn = insn_buf;
6555
6556 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6557         do {                                                            \
6558                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6559                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6560                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6561                                       si->dst_reg, si->src_reg,         \
6562                                       offsetof(struct xdp_sock, FIELD)); \
6563         } while (0)
6564
6565         switch (si->off) {
6566         case offsetof(struct bpf_xdp_sock, queue_id):
6567                 BPF_XDP_SOCK_GET(queue_id);
6568                 break;
6569         }
6570
6571         return insn - insn_buf;
6572 }
6573
6574 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6575         .func           = bpf_skb_ecn_set_ce,
6576         .gpl_only       = false,
6577         .ret_type       = RET_INTEGER,
6578         .arg1_type      = ARG_PTR_TO_CTX,
6579 };
6580
6581 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6582            struct tcphdr *, th, u32, th_len)
6583 {
6584 #ifdef CONFIG_SYN_COOKIES
6585         u32 cookie;
6586         int ret;
6587
6588         if (unlikely(!sk || th_len < sizeof(*th)))
6589                 return -EINVAL;
6590
6591         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6592         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6593                 return -EINVAL;
6594
6595         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6596                 return -EINVAL;
6597
6598         if (!th->ack || th->rst || th->syn)
6599                 return -ENOENT;
6600
6601         if (tcp_synq_no_recent_overflow(sk))
6602                 return -ENOENT;
6603
6604         cookie = ntohl(th->ack_seq) - 1;
6605
6606         switch (sk->sk_family) {
6607         case AF_INET:
6608                 if (unlikely(iph_len < sizeof(struct iphdr)))
6609                         return -EINVAL;
6610
6611                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6612                 break;
6613
6614 #if IS_BUILTIN(CONFIG_IPV6)
6615         case AF_INET6:
6616                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6617                         return -EINVAL;
6618
6619                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6620                 break;
6621 #endif /* CONFIG_IPV6 */
6622
6623         default:
6624                 return -EPROTONOSUPPORT;
6625         }
6626
6627         if (ret > 0)
6628                 return 0;
6629
6630         return -ENOENT;
6631 #else
6632         return -ENOTSUPP;
6633 #endif
6634 }
6635
6636 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6637         .func           = bpf_tcp_check_syncookie,
6638         .gpl_only       = true,
6639         .pkt_access     = true,
6640         .ret_type       = RET_INTEGER,
6641         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6642         .arg2_type      = ARG_PTR_TO_MEM,
6643         .arg3_type      = ARG_CONST_SIZE,
6644         .arg4_type      = ARG_PTR_TO_MEM,
6645         .arg5_type      = ARG_CONST_SIZE,
6646 };
6647
6648 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6649            struct tcphdr *, th, u32, th_len)
6650 {
6651 #ifdef CONFIG_SYN_COOKIES
6652         u32 cookie;
6653         u16 mss;
6654
6655         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6656                 return -EINVAL;
6657
6658         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6659                 return -EINVAL;
6660
6661         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6662                 return -ENOENT;
6663
6664         if (!th->syn || th->ack || th->fin || th->rst)
6665                 return -EINVAL;
6666
6667         if (unlikely(iph_len < sizeof(struct iphdr)))
6668                 return -EINVAL;
6669
6670         /* Both struct iphdr and struct ipv6hdr have the version field at the
6671          * same offset so we can cast to the shorter header (struct iphdr).
6672          */
6673         switch (((struct iphdr *)iph)->version) {
6674         case 4:
6675                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6676                         return -EINVAL;
6677
6678                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6679                 break;
6680
6681 #if IS_BUILTIN(CONFIG_IPV6)
6682         case 6:
6683                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6684                         return -EINVAL;
6685
6686                 if (sk->sk_family != AF_INET6)
6687                         return -EINVAL;
6688
6689                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6690                 break;
6691 #endif /* CONFIG_IPV6 */
6692
6693         default:
6694                 return -EPROTONOSUPPORT;
6695         }
6696         if (mss == 0)
6697                 return -ENOENT;
6698
6699         return cookie | ((u64)mss << 32);
6700 #else
6701         return -EOPNOTSUPP;
6702 #endif /* CONFIG_SYN_COOKIES */
6703 }
6704
6705 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6706         .func           = bpf_tcp_gen_syncookie,
6707         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6708         .pkt_access     = true,
6709         .ret_type       = RET_INTEGER,
6710         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6711         .arg2_type      = ARG_PTR_TO_MEM,
6712         .arg3_type      = ARG_CONST_SIZE,
6713         .arg4_type      = ARG_PTR_TO_MEM,
6714         .arg5_type      = ARG_CONST_SIZE,
6715 };
6716
6717 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6718 {
6719         if (!sk || flags != 0)
6720                 return -EINVAL;
6721         if (!skb_at_tc_ingress(skb))
6722                 return -EOPNOTSUPP;
6723         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6724                 return -ENETUNREACH;
6725         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6726                 return -ESOCKTNOSUPPORT;
6727         if (sk_is_refcounted(sk) &&
6728             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6729                 return -ENOENT;
6730
6731         skb_orphan(skb);
6732         skb->sk = sk;
6733         skb->destructor = sock_pfree;
6734
6735         return 0;
6736 }
6737
6738 static const struct bpf_func_proto bpf_sk_assign_proto = {
6739         .func           = bpf_sk_assign,
6740         .gpl_only       = false,
6741         .ret_type       = RET_INTEGER,
6742         .arg1_type      = ARG_PTR_TO_CTX,
6743         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6744         .arg3_type      = ARG_ANYTHING,
6745 };
6746
6747 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6748                                     u8 search_kind, const u8 *magic,
6749                                     u8 magic_len, bool *eol)
6750 {
6751         u8 kind, kind_len;
6752
6753         *eol = false;
6754
6755         while (op < opend) {
6756                 kind = op[0];
6757
6758                 if (kind == TCPOPT_EOL) {
6759                         *eol = true;
6760                         return ERR_PTR(-ENOMSG);
6761                 } else if (kind == TCPOPT_NOP) {
6762                         op++;
6763                         continue;
6764                 }
6765
6766                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6767                         /* Something is wrong in the received header.
6768                          * Follow the TCP stack's tcp_parse_options()
6769                          * and just bail here.
6770                          */
6771                         return ERR_PTR(-EFAULT);
6772
6773                 kind_len = op[1];
6774                 if (search_kind == kind) {
6775                         if (!magic_len)
6776                                 return op;
6777
6778                         if (magic_len > kind_len - 2)
6779                                 return ERR_PTR(-ENOMSG);
6780
6781                         if (!memcmp(&op[2], magic, magic_len))
6782                                 return op;
6783                 }
6784
6785                 op += kind_len;
6786         }
6787
6788         return ERR_PTR(-ENOMSG);
6789 }
6790
6791 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6792            void *, search_res, u32, len, u64, flags)
6793 {
6794         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6795         const u8 *op, *opend, *magic, *search = search_res;
6796         u8 search_kind, search_len, copy_len, magic_len;
6797         int ret;
6798
6799         /* 2 byte is the minimal option len except TCPOPT_NOP and
6800          * TCPOPT_EOL which are useless for the bpf prog to learn
6801          * and this helper disallow loading them also.
6802          */
6803         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6804                 return -EINVAL;
6805
6806         search_kind = search[0];
6807         search_len = search[1];
6808
6809         if (search_len > len || search_kind == TCPOPT_NOP ||
6810             search_kind == TCPOPT_EOL)
6811                 return -EINVAL;
6812
6813         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6814                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6815                 if (search_len != 4 && search_len != 6)
6816                         return -EINVAL;
6817                 magic = &search[2];
6818                 magic_len = search_len - 2;
6819         } else {
6820                 if (search_len)
6821                         return -EINVAL;
6822                 magic = NULL;
6823                 magic_len = 0;
6824         }
6825
6826         if (load_syn) {
6827                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6828                 if (ret < 0)
6829                         return ret;
6830
6831                 opend = op + ret;
6832                 op += sizeof(struct tcphdr);
6833         } else {
6834                 if (!bpf_sock->skb ||
6835                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6836                         /* This bpf_sock->op cannot call this helper */
6837                         return -EPERM;
6838
6839                 opend = bpf_sock->skb_data_end;
6840                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6841         }
6842
6843         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6844                                 &eol);
6845         if (IS_ERR(op))
6846                 return PTR_ERR(op);
6847
6848         copy_len = op[1];
6849         ret = copy_len;
6850         if (copy_len > len) {
6851                 ret = -ENOSPC;
6852                 copy_len = len;
6853         }
6854
6855         memcpy(search_res, op, copy_len);
6856         return ret;
6857 }
6858
6859 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6860         .func           = bpf_sock_ops_load_hdr_opt,
6861         .gpl_only       = false,
6862         .ret_type       = RET_INTEGER,
6863         .arg1_type      = ARG_PTR_TO_CTX,
6864         .arg2_type      = ARG_PTR_TO_MEM,
6865         .arg3_type      = ARG_CONST_SIZE,
6866         .arg4_type      = ARG_ANYTHING,
6867 };
6868
6869 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6870            const void *, from, u32, len, u64, flags)
6871 {
6872         u8 new_kind, new_kind_len, magic_len = 0, *opend;
6873         const u8 *op, *new_op, *magic = NULL;
6874         struct sk_buff *skb;
6875         bool eol;
6876
6877         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6878                 return -EPERM;
6879
6880         if (len < 2 || flags)
6881                 return -EINVAL;
6882
6883         new_op = from;
6884         new_kind = new_op[0];
6885         new_kind_len = new_op[1];
6886
6887         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6888             new_kind == TCPOPT_EOL)
6889                 return -EINVAL;
6890
6891         if (new_kind_len > bpf_sock->remaining_opt_len)
6892                 return -ENOSPC;
6893
6894         /* 253 is another experimental kind */
6895         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
6896                 if (new_kind_len < 4)
6897                         return -EINVAL;
6898                 /* Match for the 2 byte magic also.
6899                  * RFC 6994: the magic could be 2 or 4 bytes.
6900                  * Hence, matching by 2 byte only is on the
6901                  * conservative side but it is the right
6902                  * thing to do for the 'search-for-duplication'
6903                  * purpose.
6904                  */
6905                 magic = &new_op[2];
6906                 magic_len = 2;
6907         }
6908
6909         /* Check for duplication */
6910         skb = bpf_sock->skb;
6911         op = skb->data + sizeof(struct tcphdr);
6912         opend = bpf_sock->skb_data_end;
6913
6914         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6915                                 &eol);
6916         if (!IS_ERR(op))
6917                 return -EEXIST;
6918
6919         if (PTR_ERR(op) != -ENOMSG)
6920                 return PTR_ERR(op);
6921
6922         if (eol)
6923                 /* The option has been ended.  Treat it as no more
6924                  * header option can be written.
6925                  */
6926                 return -ENOSPC;
6927
6928         /* No duplication found.  Store the header option. */
6929         memcpy(opend, from, new_kind_len);
6930
6931         bpf_sock->remaining_opt_len -= new_kind_len;
6932         bpf_sock->skb_data_end += new_kind_len;
6933
6934         return 0;
6935 }
6936
6937 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6938         .func           = bpf_sock_ops_store_hdr_opt,
6939         .gpl_only       = false,
6940         .ret_type       = RET_INTEGER,
6941         .arg1_type      = ARG_PTR_TO_CTX,
6942         .arg2_type      = ARG_PTR_TO_MEM,
6943         .arg3_type      = ARG_CONST_SIZE,
6944         .arg4_type      = ARG_ANYTHING,
6945 };
6946
6947 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6948            u32, len, u64, flags)
6949 {
6950         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6951                 return -EPERM;
6952
6953         if (flags || len < 2)
6954                 return -EINVAL;
6955
6956         if (len > bpf_sock->remaining_opt_len)
6957                 return -ENOSPC;
6958
6959         bpf_sock->remaining_opt_len -= len;
6960
6961         return 0;
6962 }
6963
6964 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6965         .func           = bpf_sock_ops_reserve_hdr_opt,
6966         .gpl_only       = false,
6967         .ret_type       = RET_INTEGER,
6968         .arg1_type      = ARG_PTR_TO_CTX,
6969         .arg2_type      = ARG_ANYTHING,
6970         .arg3_type      = ARG_ANYTHING,
6971 };
6972
6973 #endif /* CONFIG_INET */
6974
6975 bool bpf_helper_changes_pkt_data(void *func)
6976 {
6977         if (func == bpf_skb_vlan_push ||
6978             func == bpf_skb_vlan_pop ||
6979             func == bpf_skb_store_bytes ||
6980             func == bpf_skb_change_proto ||
6981             func == bpf_skb_change_head ||
6982             func == sk_skb_change_head ||
6983             func == bpf_skb_change_tail ||
6984             func == sk_skb_change_tail ||
6985             func == bpf_skb_adjust_room ||
6986             func == sk_skb_adjust_room ||
6987             func == bpf_skb_pull_data ||
6988             func == sk_skb_pull_data ||
6989             func == bpf_clone_redirect ||
6990             func == bpf_l3_csum_replace ||
6991             func == bpf_l4_csum_replace ||
6992             func == bpf_xdp_adjust_head ||
6993             func == bpf_xdp_adjust_meta ||
6994             func == bpf_msg_pull_data ||
6995             func == bpf_msg_push_data ||
6996             func == bpf_msg_pop_data ||
6997             func == bpf_xdp_adjust_tail ||
6998 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6999             func == bpf_lwt_seg6_store_bytes ||
7000             func == bpf_lwt_seg6_adjust_srh ||
7001             func == bpf_lwt_seg6_action ||
7002 #endif
7003 #ifdef CONFIG_INET
7004             func == bpf_sock_ops_store_hdr_opt ||
7005 #endif
7006             func == bpf_lwt_in_push_encap ||
7007             func == bpf_lwt_xmit_push_encap)
7008                 return true;
7009
7010         return false;
7011 }
7012
7013 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7014 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7015
7016 static const struct bpf_func_proto *
7017 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7018 {
7019         switch (func_id) {
7020         /* inet and inet6 sockets are created in a process
7021          * context so there is always a valid uid/gid
7022          */
7023         case BPF_FUNC_get_current_uid_gid:
7024                 return &bpf_get_current_uid_gid_proto;
7025         case BPF_FUNC_get_local_storage:
7026                 return &bpf_get_local_storage_proto;
7027         case BPF_FUNC_get_socket_cookie:
7028                 return &bpf_get_socket_cookie_sock_proto;
7029         case BPF_FUNC_get_netns_cookie:
7030                 return &bpf_get_netns_cookie_sock_proto;
7031         case BPF_FUNC_perf_event_output:
7032                 return &bpf_event_output_data_proto;
7033         case BPF_FUNC_get_current_pid_tgid:
7034                 return &bpf_get_current_pid_tgid_proto;
7035         case BPF_FUNC_get_current_comm:
7036                 return &bpf_get_current_comm_proto;
7037 #ifdef CONFIG_CGROUPS
7038         case BPF_FUNC_get_current_cgroup_id:
7039                 return &bpf_get_current_cgroup_id_proto;
7040         case BPF_FUNC_get_current_ancestor_cgroup_id:
7041                 return &bpf_get_current_ancestor_cgroup_id_proto;
7042 #endif
7043 #ifdef CONFIG_CGROUP_NET_CLASSID
7044         case BPF_FUNC_get_cgroup_classid:
7045                 return &bpf_get_cgroup_classid_curr_proto;
7046 #endif
7047         case BPF_FUNC_sk_storage_get:
7048                 return &bpf_sk_storage_get_cg_sock_proto;
7049         default:
7050                 return bpf_base_func_proto(func_id);
7051         }
7052 }
7053
7054 static const struct bpf_func_proto *
7055 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7056 {
7057         switch (func_id) {
7058         /* inet and inet6 sockets are created in a process
7059          * context so there is always a valid uid/gid
7060          */
7061         case BPF_FUNC_get_current_uid_gid:
7062                 return &bpf_get_current_uid_gid_proto;
7063         case BPF_FUNC_bind:
7064                 switch (prog->expected_attach_type) {
7065                 case BPF_CGROUP_INET4_CONNECT:
7066                 case BPF_CGROUP_INET6_CONNECT:
7067                         return &bpf_bind_proto;
7068                 default:
7069                         return NULL;
7070                 }
7071         case BPF_FUNC_get_socket_cookie:
7072                 return &bpf_get_socket_cookie_sock_addr_proto;
7073         case BPF_FUNC_get_netns_cookie:
7074                 return &bpf_get_netns_cookie_sock_addr_proto;
7075         case BPF_FUNC_get_local_storage:
7076                 return &bpf_get_local_storage_proto;
7077         case BPF_FUNC_perf_event_output:
7078                 return &bpf_event_output_data_proto;
7079         case BPF_FUNC_get_current_pid_tgid:
7080                 return &bpf_get_current_pid_tgid_proto;
7081         case BPF_FUNC_get_current_comm:
7082                 return &bpf_get_current_comm_proto;
7083 #ifdef CONFIG_CGROUPS
7084         case BPF_FUNC_get_current_cgroup_id:
7085                 return &bpf_get_current_cgroup_id_proto;
7086         case BPF_FUNC_get_current_ancestor_cgroup_id:
7087                 return &bpf_get_current_ancestor_cgroup_id_proto;
7088 #endif
7089 #ifdef CONFIG_CGROUP_NET_CLASSID
7090         case BPF_FUNC_get_cgroup_classid:
7091                 return &bpf_get_cgroup_classid_curr_proto;
7092 #endif
7093 #ifdef CONFIG_INET
7094         case BPF_FUNC_sk_lookup_tcp:
7095                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7096         case BPF_FUNC_sk_lookup_udp:
7097                 return &bpf_sock_addr_sk_lookup_udp_proto;
7098         case BPF_FUNC_sk_release:
7099                 return &bpf_sk_release_proto;
7100         case BPF_FUNC_skc_lookup_tcp:
7101                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7102 #endif /* CONFIG_INET */
7103         case BPF_FUNC_sk_storage_get:
7104                 return &bpf_sk_storage_get_proto;
7105         case BPF_FUNC_sk_storage_delete:
7106                 return &bpf_sk_storage_delete_proto;
7107         case BPF_FUNC_setsockopt:
7108                 switch (prog->expected_attach_type) {
7109                 case BPF_CGROUP_INET4_BIND:
7110                 case BPF_CGROUP_INET6_BIND:
7111                 case BPF_CGROUP_INET4_CONNECT:
7112                 case BPF_CGROUP_INET6_CONNECT:
7113                 case BPF_CGROUP_UDP4_RECVMSG:
7114                 case BPF_CGROUP_UDP6_RECVMSG:
7115                 case BPF_CGROUP_UDP4_SENDMSG:
7116                 case BPF_CGROUP_UDP6_SENDMSG:
7117                 case BPF_CGROUP_INET4_GETPEERNAME:
7118                 case BPF_CGROUP_INET6_GETPEERNAME:
7119                 case BPF_CGROUP_INET4_GETSOCKNAME:
7120                 case BPF_CGROUP_INET6_GETSOCKNAME:
7121                         return &bpf_sock_addr_setsockopt_proto;
7122                 default:
7123                         return NULL;
7124                 }
7125         case BPF_FUNC_getsockopt:
7126                 switch (prog->expected_attach_type) {
7127                 case BPF_CGROUP_INET4_BIND:
7128                 case BPF_CGROUP_INET6_BIND:
7129                 case BPF_CGROUP_INET4_CONNECT:
7130                 case BPF_CGROUP_INET6_CONNECT:
7131                 case BPF_CGROUP_UDP4_RECVMSG:
7132                 case BPF_CGROUP_UDP6_RECVMSG:
7133                 case BPF_CGROUP_UDP4_SENDMSG:
7134                 case BPF_CGROUP_UDP6_SENDMSG:
7135                 case BPF_CGROUP_INET4_GETPEERNAME:
7136                 case BPF_CGROUP_INET6_GETPEERNAME:
7137                 case BPF_CGROUP_INET4_GETSOCKNAME:
7138                 case BPF_CGROUP_INET6_GETSOCKNAME:
7139                         return &bpf_sock_addr_getsockopt_proto;
7140                 default:
7141                         return NULL;
7142                 }
7143         default:
7144                 return bpf_sk_base_func_proto(func_id);
7145         }
7146 }
7147
7148 static const struct bpf_func_proto *
7149 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7150 {
7151         switch (func_id) {
7152         case BPF_FUNC_skb_load_bytes:
7153                 return &bpf_skb_load_bytes_proto;
7154         case BPF_FUNC_skb_load_bytes_relative:
7155                 return &bpf_skb_load_bytes_relative_proto;
7156         case BPF_FUNC_get_socket_cookie:
7157                 return &bpf_get_socket_cookie_proto;
7158         case BPF_FUNC_get_socket_uid:
7159                 return &bpf_get_socket_uid_proto;
7160         case BPF_FUNC_perf_event_output:
7161                 return &bpf_skb_event_output_proto;
7162         default:
7163                 return bpf_sk_base_func_proto(func_id);
7164         }
7165 }
7166
7167 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7168 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7169
7170 static const struct bpf_func_proto *
7171 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7172 {
7173         switch (func_id) {
7174         case BPF_FUNC_get_local_storage:
7175                 return &bpf_get_local_storage_proto;
7176         case BPF_FUNC_sk_fullsock:
7177                 return &bpf_sk_fullsock_proto;
7178         case BPF_FUNC_sk_storage_get:
7179                 return &bpf_sk_storage_get_proto;
7180         case BPF_FUNC_sk_storage_delete:
7181                 return &bpf_sk_storage_delete_proto;
7182         case BPF_FUNC_perf_event_output:
7183                 return &bpf_skb_event_output_proto;
7184 #ifdef CONFIG_SOCK_CGROUP_DATA
7185         case BPF_FUNC_skb_cgroup_id:
7186                 return &bpf_skb_cgroup_id_proto;
7187         case BPF_FUNC_skb_ancestor_cgroup_id:
7188                 return &bpf_skb_ancestor_cgroup_id_proto;
7189         case BPF_FUNC_sk_cgroup_id:
7190                 return &bpf_sk_cgroup_id_proto;
7191         case BPF_FUNC_sk_ancestor_cgroup_id:
7192                 return &bpf_sk_ancestor_cgroup_id_proto;
7193 #endif
7194 #ifdef CONFIG_INET
7195         case BPF_FUNC_sk_lookup_tcp:
7196                 return &bpf_sk_lookup_tcp_proto;
7197         case BPF_FUNC_sk_lookup_udp:
7198                 return &bpf_sk_lookup_udp_proto;
7199         case BPF_FUNC_sk_release:
7200                 return &bpf_sk_release_proto;
7201         case BPF_FUNC_skc_lookup_tcp:
7202                 return &bpf_skc_lookup_tcp_proto;
7203         case BPF_FUNC_tcp_sock:
7204                 return &bpf_tcp_sock_proto;
7205         case BPF_FUNC_get_listener_sock:
7206                 return &bpf_get_listener_sock_proto;
7207         case BPF_FUNC_skb_ecn_set_ce:
7208                 return &bpf_skb_ecn_set_ce_proto;
7209 #endif
7210         default:
7211                 return sk_filter_func_proto(func_id, prog);
7212         }
7213 }
7214
7215 static const struct bpf_func_proto *
7216 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7217 {
7218         switch (func_id) {
7219         case BPF_FUNC_skb_store_bytes:
7220                 return &bpf_skb_store_bytes_proto;
7221         case BPF_FUNC_skb_load_bytes:
7222                 return &bpf_skb_load_bytes_proto;
7223         case BPF_FUNC_skb_load_bytes_relative:
7224                 return &bpf_skb_load_bytes_relative_proto;
7225         case BPF_FUNC_skb_pull_data:
7226                 return &bpf_skb_pull_data_proto;
7227         case BPF_FUNC_csum_diff:
7228                 return &bpf_csum_diff_proto;
7229         case BPF_FUNC_csum_update:
7230                 return &bpf_csum_update_proto;
7231         case BPF_FUNC_csum_level:
7232                 return &bpf_csum_level_proto;
7233         case BPF_FUNC_l3_csum_replace:
7234                 return &bpf_l3_csum_replace_proto;
7235         case BPF_FUNC_l4_csum_replace:
7236                 return &bpf_l4_csum_replace_proto;
7237         case BPF_FUNC_clone_redirect:
7238                 return &bpf_clone_redirect_proto;
7239         case BPF_FUNC_get_cgroup_classid:
7240                 return &bpf_get_cgroup_classid_proto;
7241         case BPF_FUNC_skb_vlan_push:
7242                 return &bpf_skb_vlan_push_proto;
7243         case BPF_FUNC_skb_vlan_pop:
7244                 return &bpf_skb_vlan_pop_proto;
7245         case BPF_FUNC_skb_change_proto:
7246                 return &bpf_skb_change_proto_proto;
7247         case BPF_FUNC_skb_change_type:
7248                 return &bpf_skb_change_type_proto;
7249         case BPF_FUNC_skb_adjust_room:
7250                 return &bpf_skb_adjust_room_proto;
7251         case BPF_FUNC_skb_change_tail:
7252                 return &bpf_skb_change_tail_proto;
7253         case BPF_FUNC_skb_change_head:
7254                 return &bpf_skb_change_head_proto;
7255         case BPF_FUNC_skb_get_tunnel_key:
7256                 return &bpf_skb_get_tunnel_key_proto;
7257         case BPF_FUNC_skb_set_tunnel_key:
7258                 return bpf_get_skb_set_tunnel_proto(func_id);
7259         case BPF_FUNC_skb_get_tunnel_opt:
7260                 return &bpf_skb_get_tunnel_opt_proto;
7261         case BPF_FUNC_skb_set_tunnel_opt:
7262                 return bpf_get_skb_set_tunnel_proto(func_id);
7263         case BPF_FUNC_redirect:
7264                 return &bpf_redirect_proto;
7265         case BPF_FUNC_redirect_neigh:
7266                 return &bpf_redirect_neigh_proto;
7267         case BPF_FUNC_redirect_peer:
7268                 return &bpf_redirect_peer_proto;
7269         case BPF_FUNC_get_route_realm:
7270                 return &bpf_get_route_realm_proto;
7271         case BPF_FUNC_get_hash_recalc:
7272                 return &bpf_get_hash_recalc_proto;
7273         case BPF_FUNC_set_hash_invalid:
7274                 return &bpf_set_hash_invalid_proto;
7275         case BPF_FUNC_set_hash:
7276                 return &bpf_set_hash_proto;
7277         case BPF_FUNC_perf_event_output:
7278                 return &bpf_skb_event_output_proto;
7279         case BPF_FUNC_get_smp_processor_id:
7280                 return &bpf_get_smp_processor_id_proto;
7281         case BPF_FUNC_skb_under_cgroup:
7282                 return &bpf_skb_under_cgroup_proto;
7283         case BPF_FUNC_get_socket_cookie:
7284                 return &bpf_get_socket_cookie_proto;
7285         case BPF_FUNC_get_socket_uid:
7286                 return &bpf_get_socket_uid_proto;
7287         case BPF_FUNC_fib_lookup:
7288                 return &bpf_skb_fib_lookup_proto;
7289         case BPF_FUNC_check_mtu:
7290                 return &bpf_skb_check_mtu_proto;
7291         case BPF_FUNC_sk_fullsock:
7292                 return &bpf_sk_fullsock_proto;
7293         case BPF_FUNC_sk_storage_get:
7294                 return &bpf_sk_storage_get_proto;
7295         case BPF_FUNC_sk_storage_delete:
7296                 return &bpf_sk_storage_delete_proto;
7297 #ifdef CONFIG_XFRM
7298         case BPF_FUNC_skb_get_xfrm_state:
7299                 return &bpf_skb_get_xfrm_state_proto;
7300 #endif
7301 #ifdef CONFIG_CGROUP_NET_CLASSID
7302         case BPF_FUNC_skb_cgroup_classid:
7303                 return &bpf_skb_cgroup_classid_proto;
7304 #endif
7305 #ifdef CONFIG_SOCK_CGROUP_DATA
7306         case BPF_FUNC_skb_cgroup_id:
7307                 return &bpf_skb_cgroup_id_proto;
7308         case BPF_FUNC_skb_ancestor_cgroup_id:
7309                 return &bpf_skb_ancestor_cgroup_id_proto;
7310 #endif
7311 #ifdef CONFIG_INET
7312         case BPF_FUNC_sk_lookup_tcp:
7313                 return &bpf_sk_lookup_tcp_proto;
7314         case BPF_FUNC_sk_lookup_udp:
7315                 return &bpf_sk_lookup_udp_proto;
7316         case BPF_FUNC_sk_release:
7317                 return &bpf_sk_release_proto;
7318         case BPF_FUNC_tcp_sock:
7319                 return &bpf_tcp_sock_proto;
7320         case BPF_FUNC_get_listener_sock:
7321                 return &bpf_get_listener_sock_proto;
7322         case BPF_FUNC_skc_lookup_tcp:
7323                 return &bpf_skc_lookup_tcp_proto;
7324         case BPF_FUNC_tcp_check_syncookie:
7325                 return &bpf_tcp_check_syncookie_proto;
7326         case BPF_FUNC_skb_ecn_set_ce:
7327                 return &bpf_skb_ecn_set_ce_proto;
7328         case BPF_FUNC_tcp_gen_syncookie:
7329                 return &bpf_tcp_gen_syncookie_proto;
7330         case BPF_FUNC_sk_assign:
7331                 return &bpf_sk_assign_proto;
7332 #endif
7333         default:
7334                 return bpf_sk_base_func_proto(func_id);
7335         }
7336 }
7337
7338 static const struct bpf_func_proto *
7339 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7340 {
7341         switch (func_id) {
7342         case BPF_FUNC_perf_event_output:
7343                 return &bpf_xdp_event_output_proto;
7344         case BPF_FUNC_get_smp_processor_id:
7345                 return &bpf_get_smp_processor_id_proto;
7346         case BPF_FUNC_csum_diff:
7347                 return &bpf_csum_diff_proto;
7348         case BPF_FUNC_xdp_adjust_head:
7349                 return &bpf_xdp_adjust_head_proto;
7350         case BPF_FUNC_xdp_adjust_meta:
7351                 return &bpf_xdp_adjust_meta_proto;
7352         case BPF_FUNC_redirect:
7353                 return &bpf_xdp_redirect_proto;
7354         case BPF_FUNC_redirect_map:
7355                 return &bpf_xdp_redirect_map_proto;
7356         case BPF_FUNC_xdp_adjust_tail:
7357                 return &bpf_xdp_adjust_tail_proto;
7358         case BPF_FUNC_fib_lookup:
7359                 return &bpf_xdp_fib_lookup_proto;
7360         case BPF_FUNC_check_mtu:
7361                 return &bpf_xdp_check_mtu_proto;
7362 #ifdef CONFIG_INET
7363         case BPF_FUNC_sk_lookup_udp:
7364                 return &bpf_xdp_sk_lookup_udp_proto;
7365         case BPF_FUNC_sk_lookup_tcp:
7366                 return &bpf_xdp_sk_lookup_tcp_proto;
7367         case BPF_FUNC_sk_release:
7368                 return &bpf_sk_release_proto;
7369         case BPF_FUNC_skc_lookup_tcp:
7370                 return &bpf_xdp_skc_lookup_tcp_proto;
7371         case BPF_FUNC_tcp_check_syncookie:
7372                 return &bpf_tcp_check_syncookie_proto;
7373         case BPF_FUNC_tcp_gen_syncookie:
7374                 return &bpf_tcp_gen_syncookie_proto;
7375 #endif
7376         default:
7377                 return bpf_sk_base_func_proto(func_id);
7378         }
7379 }
7380
7381 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7382 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7383
7384 static const struct bpf_func_proto *
7385 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7386 {
7387         switch (func_id) {
7388         case BPF_FUNC_setsockopt:
7389                 return &bpf_sock_ops_setsockopt_proto;
7390         case BPF_FUNC_getsockopt:
7391                 return &bpf_sock_ops_getsockopt_proto;
7392         case BPF_FUNC_sock_ops_cb_flags_set:
7393                 return &bpf_sock_ops_cb_flags_set_proto;
7394         case BPF_FUNC_sock_map_update:
7395                 return &bpf_sock_map_update_proto;
7396         case BPF_FUNC_sock_hash_update:
7397                 return &bpf_sock_hash_update_proto;
7398         case BPF_FUNC_get_socket_cookie:
7399                 return &bpf_get_socket_cookie_sock_ops_proto;
7400         case BPF_FUNC_get_local_storage:
7401                 return &bpf_get_local_storage_proto;
7402         case BPF_FUNC_perf_event_output:
7403                 return &bpf_event_output_data_proto;
7404         case BPF_FUNC_sk_storage_get:
7405                 return &bpf_sk_storage_get_proto;
7406         case BPF_FUNC_sk_storage_delete:
7407                 return &bpf_sk_storage_delete_proto;
7408 #ifdef CONFIG_INET
7409         case BPF_FUNC_load_hdr_opt:
7410                 return &bpf_sock_ops_load_hdr_opt_proto;
7411         case BPF_FUNC_store_hdr_opt:
7412                 return &bpf_sock_ops_store_hdr_opt_proto;
7413         case BPF_FUNC_reserve_hdr_opt:
7414                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7415         case BPF_FUNC_tcp_sock:
7416                 return &bpf_tcp_sock_proto;
7417 #endif /* CONFIG_INET */
7418         default:
7419                 return bpf_sk_base_func_proto(func_id);
7420         }
7421 }
7422
7423 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7424 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7425
7426 static const struct bpf_func_proto *
7427 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7428 {
7429         switch (func_id) {
7430         case BPF_FUNC_msg_redirect_map:
7431                 return &bpf_msg_redirect_map_proto;
7432         case BPF_FUNC_msg_redirect_hash:
7433                 return &bpf_msg_redirect_hash_proto;
7434         case BPF_FUNC_msg_apply_bytes:
7435                 return &bpf_msg_apply_bytes_proto;
7436         case BPF_FUNC_msg_cork_bytes:
7437                 return &bpf_msg_cork_bytes_proto;
7438         case BPF_FUNC_msg_pull_data:
7439                 return &bpf_msg_pull_data_proto;
7440         case BPF_FUNC_msg_push_data:
7441                 return &bpf_msg_push_data_proto;
7442         case BPF_FUNC_msg_pop_data:
7443                 return &bpf_msg_pop_data_proto;
7444         case BPF_FUNC_perf_event_output:
7445                 return &bpf_event_output_data_proto;
7446         case BPF_FUNC_get_current_uid_gid:
7447                 return &bpf_get_current_uid_gid_proto;
7448         case BPF_FUNC_get_current_pid_tgid:
7449                 return &bpf_get_current_pid_tgid_proto;
7450         case BPF_FUNC_sk_storage_get:
7451                 return &bpf_sk_storage_get_proto;
7452         case BPF_FUNC_sk_storage_delete:
7453                 return &bpf_sk_storage_delete_proto;
7454 #ifdef CONFIG_CGROUPS
7455         case BPF_FUNC_get_current_cgroup_id:
7456                 return &bpf_get_current_cgroup_id_proto;
7457         case BPF_FUNC_get_current_ancestor_cgroup_id:
7458                 return &bpf_get_current_ancestor_cgroup_id_proto;
7459 #endif
7460 #ifdef CONFIG_CGROUP_NET_CLASSID
7461         case BPF_FUNC_get_cgroup_classid:
7462                 return &bpf_get_cgroup_classid_curr_proto;
7463 #endif
7464         default:
7465                 return bpf_sk_base_func_proto(func_id);
7466         }
7467 }
7468
7469 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7470 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7471
7472 static const struct bpf_func_proto *
7473 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7474 {
7475         switch (func_id) {
7476         case BPF_FUNC_skb_store_bytes:
7477                 return &bpf_skb_store_bytes_proto;
7478         case BPF_FUNC_skb_load_bytes:
7479                 return &bpf_skb_load_bytes_proto;
7480         case BPF_FUNC_skb_pull_data:
7481                 return &sk_skb_pull_data_proto;
7482         case BPF_FUNC_skb_change_tail:
7483                 return &sk_skb_change_tail_proto;
7484         case BPF_FUNC_skb_change_head:
7485                 return &sk_skb_change_head_proto;
7486         case BPF_FUNC_skb_adjust_room:
7487                 return &sk_skb_adjust_room_proto;
7488         case BPF_FUNC_get_socket_cookie:
7489                 return &bpf_get_socket_cookie_proto;
7490         case BPF_FUNC_get_socket_uid:
7491                 return &bpf_get_socket_uid_proto;
7492         case BPF_FUNC_sk_redirect_map:
7493                 return &bpf_sk_redirect_map_proto;
7494         case BPF_FUNC_sk_redirect_hash:
7495                 return &bpf_sk_redirect_hash_proto;
7496         case BPF_FUNC_perf_event_output:
7497                 return &bpf_skb_event_output_proto;
7498 #ifdef CONFIG_INET
7499         case BPF_FUNC_sk_lookup_tcp:
7500                 return &bpf_sk_lookup_tcp_proto;
7501         case BPF_FUNC_sk_lookup_udp:
7502                 return &bpf_sk_lookup_udp_proto;
7503         case BPF_FUNC_sk_release:
7504                 return &bpf_sk_release_proto;
7505         case BPF_FUNC_skc_lookup_tcp:
7506                 return &bpf_skc_lookup_tcp_proto;
7507 #endif
7508         default:
7509                 return bpf_sk_base_func_proto(func_id);
7510         }
7511 }
7512
7513 static const struct bpf_func_proto *
7514 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7515 {
7516         switch (func_id) {
7517         case BPF_FUNC_skb_load_bytes:
7518                 return &bpf_flow_dissector_load_bytes_proto;
7519         default:
7520                 return bpf_sk_base_func_proto(func_id);
7521         }
7522 }
7523
7524 static const struct bpf_func_proto *
7525 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7526 {
7527         switch (func_id) {
7528         case BPF_FUNC_skb_load_bytes:
7529                 return &bpf_skb_load_bytes_proto;
7530         case BPF_FUNC_skb_pull_data:
7531                 return &bpf_skb_pull_data_proto;
7532         case BPF_FUNC_csum_diff:
7533                 return &bpf_csum_diff_proto;
7534         case BPF_FUNC_get_cgroup_classid:
7535                 return &bpf_get_cgroup_classid_proto;
7536         case BPF_FUNC_get_route_realm:
7537                 return &bpf_get_route_realm_proto;
7538         case BPF_FUNC_get_hash_recalc:
7539                 return &bpf_get_hash_recalc_proto;
7540         case BPF_FUNC_perf_event_output:
7541                 return &bpf_skb_event_output_proto;
7542         case BPF_FUNC_get_smp_processor_id:
7543                 return &bpf_get_smp_processor_id_proto;
7544         case BPF_FUNC_skb_under_cgroup:
7545                 return &bpf_skb_under_cgroup_proto;
7546         default:
7547                 return bpf_sk_base_func_proto(func_id);
7548         }
7549 }
7550
7551 static const struct bpf_func_proto *
7552 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7553 {
7554         switch (func_id) {
7555         case BPF_FUNC_lwt_push_encap:
7556                 return &bpf_lwt_in_push_encap_proto;
7557         default:
7558                 return lwt_out_func_proto(func_id, prog);
7559         }
7560 }
7561
7562 static const struct bpf_func_proto *
7563 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7564 {
7565         switch (func_id) {
7566         case BPF_FUNC_skb_get_tunnel_key:
7567                 return &bpf_skb_get_tunnel_key_proto;
7568         case BPF_FUNC_skb_set_tunnel_key:
7569                 return bpf_get_skb_set_tunnel_proto(func_id);
7570         case BPF_FUNC_skb_get_tunnel_opt:
7571                 return &bpf_skb_get_tunnel_opt_proto;
7572         case BPF_FUNC_skb_set_tunnel_opt:
7573                 return bpf_get_skb_set_tunnel_proto(func_id);
7574         case BPF_FUNC_redirect:
7575                 return &bpf_redirect_proto;
7576         case BPF_FUNC_clone_redirect:
7577                 return &bpf_clone_redirect_proto;
7578         case BPF_FUNC_skb_change_tail:
7579                 return &bpf_skb_change_tail_proto;
7580         case BPF_FUNC_skb_change_head:
7581                 return &bpf_skb_change_head_proto;
7582         case BPF_FUNC_skb_store_bytes:
7583                 return &bpf_skb_store_bytes_proto;
7584         case BPF_FUNC_csum_update:
7585                 return &bpf_csum_update_proto;
7586         case BPF_FUNC_csum_level:
7587                 return &bpf_csum_level_proto;
7588         case BPF_FUNC_l3_csum_replace:
7589                 return &bpf_l3_csum_replace_proto;
7590         case BPF_FUNC_l4_csum_replace:
7591                 return &bpf_l4_csum_replace_proto;
7592         case BPF_FUNC_set_hash_invalid:
7593                 return &bpf_set_hash_invalid_proto;
7594         case BPF_FUNC_lwt_push_encap:
7595                 return &bpf_lwt_xmit_push_encap_proto;
7596         default:
7597                 return lwt_out_func_proto(func_id, prog);
7598         }
7599 }
7600
7601 static const struct bpf_func_proto *
7602 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7603 {
7604         switch (func_id) {
7605 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7606         case BPF_FUNC_lwt_seg6_store_bytes:
7607                 return &bpf_lwt_seg6_store_bytes_proto;
7608         case BPF_FUNC_lwt_seg6_action:
7609                 return &bpf_lwt_seg6_action_proto;
7610         case BPF_FUNC_lwt_seg6_adjust_srh:
7611                 return &bpf_lwt_seg6_adjust_srh_proto;
7612 #endif
7613         default:
7614                 return lwt_out_func_proto(func_id, prog);
7615         }
7616 }
7617
7618 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7619                                     const struct bpf_prog *prog,
7620                                     struct bpf_insn_access_aux *info)
7621 {
7622         const int size_default = sizeof(__u32);
7623
7624         if (off < 0 || off >= sizeof(struct __sk_buff))
7625                 return false;
7626
7627         /* The verifier guarantees that size > 0. */
7628         if (off % size != 0)
7629                 return false;
7630
7631         switch (off) {
7632         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7633                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7634                         return false;
7635                 break;
7636         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7637         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7638         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7639         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7640         case bpf_ctx_range(struct __sk_buff, data):
7641         case bpf_ctx_range(struct __sk_buff, data_meta):
7642         case bpf_ctx_range(struct __sk_buff, data_end):
7643                 if (size != size_default)
7644                         return false;
7645                 break;
7646         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7647                 return false;
7648         case bpf_ctx_range(struct __sk_buff, tstamp):
7649                 if (size != sizeof(__u64))
7650                         return false;
7651                 break;
7652         case offsetof(struct __sk_buff, sk):
7653                 if (type == BPF_WRITE || size != sizeof(__u64))
7654                         return false;
7655                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7656                 break;
7657         default:
7658                 /* Only narrow read access allowed for now. */
7659                 if (type == BPF_WRITE) {
7660                         if (size != size_default)
7661                                 return false;
7662                 } else {
7663                         bpf_ctx_record_field_size(info, size_default);
7664                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7665                                 return false;
7666                 }
7667         }
7668
7669         return true;
7670 }
7671
7672 static bool sk_filter_is_valid_access(int off, int size,
7673                                       enum bpf_access_type type,
7674                                       const struct bpf_prog *prog,
7675                                       struct bpf_insn_access_aux *info)
7676 {
7677         switch (off) {
7678         case bpf_ctx_range(struct __sk_buff, tc_classid):
7679         case bpf_ctx_range(struct __sk_buff, data):
7680         case bpf_ctx_range(struct __sk_buff, data_meta):
7681         case bpf_ctx_range(struct __sk_buff, data_end):
7682         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7683         case bpf_ctx_range(struct __sk_buff, tstamp):
7684         case bpf_ctx_range(struct __sk_buff, wire_len):
7685                 return false;
7686         }
7687
7688         if (type == BPF_WRITE) {
7689                 switch (off) {
7690                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7691                         break;
7692                 default:
7693                         return false;
7694                 }
7695         }
7696
7697         return bpf_skb_is_valid_access(off, size, type, prog, info);
7698 }
7699
7700 static bool cg_skb_is_valid_access(int off, int size,
7701                                    enum bpf_access_type type,
7702                                    const struct bpf_prog *prog,
7703                                    struct bpf_insn_access_aux *info)
7704 {
7705         switch (off) {
7706         case bpf_ctx_range(struct __sk_buff, tc_classid):
7707         case bpf_ctx_range(struct __sk_buff, data_meta):
7708         case bpf_ctx_range(struct __sk_buff, wire_len):
7709                 return false;
7710         case bpf_ctx_range(struct __sk_buff, data):
7711         case bpf_ctx_range(struct __sk_buff, data_end):
7712                 if (!bpf_capable())
7713                         return false;
7714                 break;
7715         }
7716
7717         if (type == BPF_WRITE) {
7718                 switch (off) {
7719                 case bpf_ctx_range(struct __sk_buff, mark):
7720                 case bpf_ctx_range(struct __sk_buff, priority):
7721                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7722                         break;
7723                 case bpf_ctx_range(struct __sk_buff, tstamp):
7724                         if (!bpf_capable())
7725                                 return false;
7726                         break;
7727                 default:
7728                         return false;
7729                 }
7730         }
7731
7732         switch (off) {
7733         case bpf_ctx_range(struct __sk_buff, data):
7734                 info->reg_type = PTR_TO_PACKET;
7735                 break;
7736         case bpf_ctx_range(struct __sk_buff, data_end):
7737                 info->reg_type = PTR_TO_PACKET_END;
7738                 break;
7739         }
7740
7741         return bpf_skb_is_valid_access(off, size, type, prog, info);
7742 }
7743
7744 static bool lwt_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_till(struct __sk_buff, family, local_port):
7752         case bpf_ctx_range(struct __sk_buff, data_meta):
7753         case bpf_ctx_range(struct __sk_buff, tstamp):
7754         case bpf_ctx_range(struct __sk_buff, wire_len):
7755                 return false;
7756         }
7757
7758         if (type == BPF_WRITE) {
7759                 switch (off) {
7760                 case bpf_ctx_range(struct __sk_buff, mark):
7761                 case bpf_ctx_range(struct __sk_buff, priority):
7762                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7763                         break;
7764                 default:
7765                         return false;
7766                 }
7767         }
7768
7769         switch (off) {
7770         case bpf_ctx_range(struct __sk_buff, data):
7771                 info->reg_type = PTR_TO_PACKET;
7772                 break;
7773         case bpf_ctx_range(struct __sk_buff, data_end):
7774                 info->reg_type = PTR_TO_PACKET_END;
7775                 break;
7776         }
7777
7778         return bpf_skb_is_valid_access(off, size, type, prog, info);
7779 }
7780
7781 /* Attach type specific accesses */
7782 static bool __sock_filter_check_attach_type(int off,
7783                                             enum bpf_access_type access_type,
7784                                             enum bpf_attach_type attach_type)
7785 {
7786         switch (off) {
7787         case offsetof(struct bpf_sock, bound_dev_if):
7788         case offsetof(struct bpf_sock, mark):
7789         case offsetof(struct bpf_sock, priority):
7790                 switch (attach_type) {
7791                 case BPF_CGROUP_INET_SOCK_CREATE:
7792                 case BPF_CGROUP_INET_SOCK_RELEASE:
7793                         goto full_access;
7794                 default:
7795                         return false;
7796                 }
7797         case bpf_ctx_range(struct bpf_sock, src_ip4):
7798                 switch (attach_type) {
7799                 case BPF_CGROUP_INET4_POST_BIND:
7800                         goto read_only;
7801                 default:
7802                         return false;
7803                 }
7804         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7805                 switch (attach_type) {
7806                 case BPF_CGROUP_INET6_POST_BIND:
7807                         goto read_only;
7808                 default:
7809                         return false;
7810                 }
7811         case bpf_ctx_range(struct bpf_sock, src_port):
7812                 switch (attach_type) {
7813                 case BPF_CGROUP_INET4_POST_BIND:
7814                 case BPF_CGROUP_INET6_POST_BIND:
7815                         goto read_only;
7816                 default:
7817                         return false;
7818                 }
7819         }
7820 read_only:
7821         return access_type == BPF_READ;
7822 full_access:
7823         return true;
7824 }
7825
7826 bool bpf_sock_common_is_valid_access(int off, int size,
7827                                      enum bpf_access_type type,
7828                                      struct bpf_insn_access_aux *info)
7829 {
7830         switch (off) {
7831         case bpf_ctx_range_till(struct bpf_sock, type, priority):
7832                 return false;
7833         default:
7834                 return bpf_sock_is_valid_access(off, size, type, info);
7835         }
7836 }
7837
7838 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7839                               struct bpf_insn_access_aux *info)
7840 {
7841         const int size_default = sizeof(__u32);
7842
7843         if (off < 0 || off >= sizeof(struct bpf_sock))
7844                 return false;
7845         if (off % size != 0)
7846                 return false;
7847
7848         switch (off) {
7849         case offsetof(struct bpf_sock, state):
7850         case offsetof(struct bpf_sock, family):
7851         case offsetof(struct bpf_sock, type):
7852         case offsetof(struct bpf_sock, protocol):
7853         case offsetof(struct bpf_sock, dst_port):
7854         case offsetof(struct bpf_sock, src_port):
7855         case offsetof(struct bpf_sock, rx_queue_mapping):
7856         case bpf_ctx_range(struct bpf_sock, src_ip4):
7857         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7858         case bpf_ctx_range(struct bpf_sock, dst_ip4):
7859         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7860                 bpf_ctx_record_field_size(info, size_default);
7861                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7862         }
7863
7864         return size == size_default;
7865 }
7866
7867 static bool sock_filter_is_valid_access(int off, int size,
7868                                         enum bpf_access_type type,
7869                                         const struct bpf_prog *prog,
7870                                         struct bpf_insn_access_aux *info)
7871 {
7872         if (!bpf_sock_is_valid_access(off, size, type, info))
7873                 return false;
7874         return __sock_filter_check_attach_type(off, type,
7875                                                prog->expected_attach_type);
7876 }
7877
7878 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7879                              const struct bpf_prog *prog)
7880 {
7881         /* Neither direct read nor direct write requires any preliminary
7882          * action.
7883          */
7884         return 0;
7885 }
7886
7887 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7888                                 const struct bpf_prog *prog, int drop_verdict)
7889 {
7890         struct bpf_insn *insn = insn_buf;
7891
7892         if (!direct_write)
7893                 return 0;
7894
7895         /* if (!skb->cloned)
7896          *       goto start;
7897          *
7898          * (Fast-path, otherwise approximation that we might be
7899          *  a clone, do the rest in helper.)
7900          */
7901         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7902         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7903         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7904
7905         /* ret = bpf_skb_pull_data(skb, 0); */
7906         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7907         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7908         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7909                                BPF_FUNC_skb_pull_data);
7910         /* if (!ret)
7911          *      goto restore;
7912          * return TC_ACT_SHOT;
7913          */
7914         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7915         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7916         *insn++ = BPF_EXIT_INSN();
7917
7918         /* restore: */
7919         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7920         /* start: */
7921         *insn++ = prog->insnsi[0];
7922
7923         return insn - insn_buf;
7924 }
7925
7926 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7927                           struct bpf_insn *insn_buf)
7928 {
7929         bool indirect = BPF_MODE(orig->code) == BPF_IND;
7930         struct bpf_insn *insn = insn_buf;
7931
7932         if (!indirect) {
7933                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7934         } else {
7935                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7936                 if (orig->imm)
7937                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7938         }
7939         /* We're guaranteed here that CTX is in R6. */
7940         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7941
7942         switch (BPF_SIZE(orig->code)) {
7943         case BPF_B:
7944                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7945                 break;
7946         case BPF_H:
7947                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7948                 break;
7949         case BPF_W:
7950                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7951                 break;
7952         }
7953
7954         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7955         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7956         *insn++ = BPF_EXIT_INSN();
7957
7958         return insn - insn_buf;
7959 }
7960
7961 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7962                                const struct bpf_prog *prog)
7963 {
7964         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7965 }
7966
7967 static bool tc_cls_act_is_valid_access(int off, int size,
7968                                        enum bpf_access_type type,
7969                                        const struct bpf_prog *prog,
7970                                        struct bpf_insn_access_aux *info)
7971 {
7972         if (type == BPF_WRITE) {
7973                 switch (off) {
7974                 case bpf_ctx_range(struct __sk_buff, mark):
7975                 case bpf_ctx_range(struct __sk_buff, tc_index):
7976                 case bpf_ctx_range(struct __sk_buff, priority):
7977                 case bpf_ctx_range(struct __sk_buff, tc_classid):
7978                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7979                 case bpf_ctx_range(struct __sk_buff, tstamp):
7980                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7981                         break;
7982                 default:
7983                         return false;
7984                 }
7985         }
7986
7987         switch (off) {
7988         case bpf_ctx_range(struct __sk_buff, data):
7989                 info->reg_type = PTR_TO_PACKET;
7990                 break;
7991         case bpf_ctx_range(struct __sk_buff, data_meta):
7992                 info->reg_type = PTR_TO_PACKET_META;
7993                 break;
7994         case bpf_ctx_range(struct __sk_buff, data_end):
7995                 info->reg_type = PTR_TO_PACKET_END;
7996                 break;
7997         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7998                 return false;
7999         }
8000
8001         return bpf_skb_is_valid_access(off, size, type, prog, info);
8002 }
8003
8004 static bool __is_valid_xdp_access(int off, int size)
8005 {
8006         if (off < 0 || off >= sizeof(struct xdp_md))
8007                 return false;
8008         if (off % size != 0)
8009                 return false;
8010         if (size != sizeof(__u32))
8011                 return false;
8012
8013         return true;
8014 }
8015
8016 static bool xdp_is_valid_access(int off, int size,
8017                                 enum bpf_access_type type,
8018                                 const struct bpf_prog *prog,
8019                                 struct bpf_insn_access_aux *info)
8020 {
8021         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8022                 switch (off) {
8023                 case offsetof(struct xdp_md, egress_ifindex):
8024                         return false;
8025                 }
8026         }
8027
8028         if (type == BPF_WRITE) {
8029                 if (bpf_prog_is_dev_bound(prog->aux)) {
8030                         switch (off) {
8031                         case offsetof(struct xdp_md, rx_queue_index):
8032                                 return __is_valid_xdp_access(off, size);
8033                         }
8034                 }
8035                 return false;
8036         }
8037
8038         switch (off) {
8039         case offsetof(struct xdp_md, data):
8040                 info->reg_type = PTR_TO_PACKET;
8041                 break;
8042         case offsetof(struct xdp_md, data_meta):
8043                 info->reg_type = PTR_TO_PACKET_META;
8044                 break;
8045         case offsetof(struct xdp_md, data_end):
8046                 info->reg_type = PTR_TO_PACKET_END;
8047                 break;
8048         }
8049
8050         return __is_valid_xdp_access(off, size);
8051 }
8052
8053 void bpf_warn_invalid_xdp_action(u32 act)
8054 {
8055         const u32 act_max = XDP_REDIRECT;
8056
8057         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8058                   act > act_max ? "Illegal" : "Driver unsupported",
8059                   act);
8060 }
8061 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8062
8063 static bool sock_addr_is_valid_access(int off, int size,
8064                                       enum bpf_access_type type,
8065                                       const struct bpf_prog *prog,
8066                                       struct bpf_insn_access_aux *info)
8067 {
8068         const int size_default = sizeof(__u32);
8069
8070         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8071                 return false;
8072         if (off % size != 0)
8073                 return false;
8074
8075         /* Disallow access to IPv6 fields from IPv4 contex and vise
8076          * versa.
8077          */
8078         switch (off) {
8079         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8080                 switch (prog->expected_attach_type) {
8081                 case BPF_CGROUP_INET4_BIND:
8082                 case BPF_CGROUP_INET4_CONNECT:
8083                 case BPF_CGROUP_INET4_GETPEERNAME:
8084                 case BPF_CGROUP_INET4_GETSOCKNAME:
8085                 case BPF_CGROUP_UDP4_SENDMSG:
8086                 case BPF_CGROUP_UDP4_RECVMSG:
8087                         break;
8088                 default:
8089                         return false;
8090                 }
8091                 break;
8092         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8093                 switch (prog->expected_attach_type) {
8094                 case BPF_CGROUP_INET6_BIND:
8095                 case BPF_CGROUP_INET6_CONNECT:
8096                 case BPF_CGROUP_INET6_GETPEERNAME:
8097                 case BPF_CGROUP_INET6_GETSOCKNAME:
8098                 case BPF_CGROUP_UDP6_SENDMSG:
8099                 case BPF_CGROUP_UDP6_RECVMSG:
8100                         break;
8101                 default:
8102                         return false;
8103                 }
8104                 break;
8105         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8106                 switch (prog->expected_attach_type) {
8107                 case BPF_CGROUP_UDP4_SENDMSG:
8108                         break;
8109                 default:
8110                         return false;
8111                 }
8112                 break;
8113         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8114                                 msg_src_ip6[3]):
8115                 switch (prog->expected_attach_type) {
8116                 case BPF_CGROUP_UDP6_SENDMSG:
8117                         break;
8118                 default:
8119                         return false;
8120                 }
8121                 break;
8122         }
8123
8124         switch (off) {
8125         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8126         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8127         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8128         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8129                                 msg_src_ip6[3]):
8130         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8131                 if (type == BPF_READ) {
8132                         bpf_ctx_record_field_size(info, size_default);
8133
8134                         if (bpf_ctx_wide_access_ok(off, size,
8135                                                    struct bpf_sock_addr,
8136                                                    user_ip6))
8137                                 return true;
8138
8139                         if (bpf_ctx_wide_access_ok(off, size,
8140                                                    struct bpf_sock_addr,
8141                                                    msg_src_ip6))
8142                                 return true;
8143
8144                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8145                                 return false;
8146                 } else {
8147                         if (bpf_ctx_wide_access_ok(off, size,
8148                                                    struct bpf_sock_addr,
8149                                                    user_ip6))
8150                                 return true;
8151
8152                         if (bpf_ctx_wide_access_ok(off, size,
8153                                                    struct bpf_sock_addr,
8154                                                    msg_src_ip6))
8155                                 return true;
8156
8157                         if (size != size_default)
8158                                 return false;
8159                 }
8160                 break;
8161         case offsetof(struct bpf_sock_addr, sk):
8162                 if (type != BPF_READ)
8163                         return false;
8164                 if (size != sizeof(__u64))
8165                         return false;
8166                 info->reg_type = PTR_TO_SOCKET;
8167                 break;
8168         default:
8169                 if (type == BPF_READ) {
8170                         if (size != size_default)
8171                                 return false;
8172                 } else {
8173                         return false;
8174                 }
8175         }
8176
8177         return true;
8178 }
8179
8180 static bool sock_ops_is_valid_access(int off, int size,
8181                                      enum bpf_access_type type,
8182                                      const struct bpf_prog *prog,
8183                                      struct bpf_insn_access_aux *info)
8184 {
8185         const int size_default = sizeof(__u32);
8186
8187         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8188                 return false;
8189
8190         /* The verifier guarantees that size > 0. */
8191         if (off % size != 0)
8192                 return false;
8193
8194         if (type == BPF_WRITE) {
8195                 switch (off) {
8196                 case offsetof(struct bpf_sock_ops, reply):
8197                 case offsetof(struct bpf_sock_ops, sk_txhash):
8198                         if (size != size_default)
8199                                 return false;
8200                         break;
8201                 default:
8202                         return false;
8203                 }
8204         } else {
8205                 switch (off) {
8206                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8207                                         bytes_acked):
8208                         if (size != sizeof(__u64))
8209                                 return false;
8210                         break;
8211                 case offsetof(struct bpf_sock_ops, sk):
8212                         if (size != sizeof(__u64))
8213                                 return false;
8214                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8215                         break;
8216                 case offsetof(struct bpf_sock_ops, skb_data):
8217                         if (size != sizeof(__u64))
8218                                 return false;
8219                         info->reg_type = PTR_TO_PACKET;
8220                         break;
8221                 case offsetof(struct bpf_sock_ops, skb_data_end):
8222                         if (size != sizeof(__u64))
8223                                 return false;
8224                         info->reg_type = PTR_TO_PACKET_END;
8225                         break;
8226                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8227                         bpf_ctx_record_field_size(info, size_default);
8228                         return bpf_ctx_narrow_access_ok(off, size,
8229                                                         size_default);
8230                 default:
8231                         if (size != size_default)
8232                                 return false;
8233                         break;
8234                 }
8235         }
8236
8237         return true;
8238 }
8239
8240 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8241                            const struct bpf_prog *prog)
8242 {
8243         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8244 }
8245
8246 static bool sk_skb_is_valid_access(int off, int size,
8247                                    enum bpf_access_type type,
8248                                    const struct bpf_prog *prog,
8249                                    struct bpf_insn_access_aux *info)
8250 {
8251         switch (off) {
8252         case bpf_ctx_range(struct __sk_buff, tc_classid):
8253         case bpf_ctx_range(struct __sk_buff, data_meta):
8254         case bpf_ctx_range(struct __sk_buff, tstamp):
8255         case bpf_ctx_range(struct __sk_buff, wire_len):
8256                 return false;
8257         }
8258
8259         if (type == BPF_WRITE) {
8260                 switch (off) {
8261                 case bpf_ctx_range(struct __sk_buff, tc_index):
8262                 case bpf_ctx_range(struct __sk_buff, priority):
8263                         break;
8264                 default:
8265                         return false;
8266                 }
8267         }
8268
8269         switch (off) {
8270         case bpf_ctx_range(struct __sk_buff, mark):
8271                 return false;
8272         case bpf_ctx_range(struct __sk_buff, data):
8273                 info->reg_type = PTR_TO_PACKET;
8274                 break;
8275         case bpf_ctx_range(struct __sk_buff, data_end):
8276                 info->reg_type = PTR_TO_PACKET_END;
8277                 break;
8278         }
8279
8280         return bpf_skb_is_valid_access(off, size, type, prog, info);
8281 }
8282
8283 static bool sk_msg_is_valid_access(int off, int size,
8284                                    enum bpf_access_type type,
8285                                    const struct bpf_prog *prog,
8286                                    struct bpf_insn_access_aux *info)
8287 {
8288         if (type == BPF_WRITE)
8289                 return false;
8290
8291         if (off % size != 0)
8292                 return false;
8293
8294         switch (off) {
8295         case offsetof(struct sk_msg_md, data):
8296                 info->reg_type = PTR_TO_PACKET;
8297                 if (size != sizeof(__u64))
8298                         return false;
8299                 break;
8300         case offsetof(struct sk_msg_md, data_end):
8301                 info->reg_type = PTR_TO_PACKET_END;
8302                 if (size != sizeof(__u64))
8303                         return false;
8304                 break;
8305         case offsetof(struct sk_msg_md, sk):
8306                 if (size != sizeof(__u64))
8307                         return false;
8308                 info->reg_type = PTR_TO_SOCKET;
8309                 break;
8310         case bpf_ctx_range(struct sk_msg_md, family):
8311         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8312         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8313         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8314         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8315         case bpf_ctx_range(struct sk_msg_md, remote_port):
8316         case bpf_ctx_range(struct sk_msg_md, local_port):
8317         case bpf_ctx_range(struct sk_msg_md, size):
8318                 if (size != sizeof(__u32))
8319                         return false;
8320                 break;
8321         default:
8322                 return false;
8323         }
8324         return true;
8325 }
8326
8327 static bool flow_dissector_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         const int size_default = sizeof(__u32);
8333
8334         if (off < 0 || off >= sizeof(struct __sk_buff))
8335                 return false;
8336
8337         if (type == BPF_WRITE)
8338                 return false;
8339
8340         switch (off) {
8341         case bpf_ctx_range(struct __sk_buff, data):
8342                 if (size != size_default)
8343                         return false;
8344                 info->reg_type = PTR_TO_PACKET;
8345                 return true;
8346         case bpf_ctx_range(struct __sk_buff, data_end):
8347                 if (size != size_default)
8348                         return false;
8349                 info->reg_type = PTR_TO_PACKET_END;
8350                 return true;
8351         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8352                 if (size != sizeof(__u64))
8353                         return false;
8354                 info->reg_type = PTR_TO_FLOW_KEYS;
8355                 return true;
8356         default:
8357                 return false;
8358         }
8359 }
8360
8361 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8362                                              const struct bpf_insn *si,
8363                                              struct bpf_insn *insn_buf,
8364                                              struct bpf_prog *prog,
8365                                              u32 *target_size)
8366
8367 {
8368         struct bpf_insn *insn = insn_buf;
8369
8370         switch (si->off) {
8371         case offsetof(struct __sk_buff, data):
8372                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8373                                       si->dst_reg, si->src_reg,
8374                                       offsetof(struct bpf_flow_dissector, data));
8375                 break;
8376
8377         case offsetof(struct __sk_buff, data_end):
8378                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8379                                       si->dst_reg, si->src_reg,
8380                                       offsetof(struct bpf_flow_dissector, data_end));
8381                 break;
8382
8383         case offsetof(struct __sk_buff, flow_keys):
8384                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8385                                       si->dst_reg, si->src_reg,
8386                                       offsetof(struct bpf_flow_dissector, flow_keys));
8387                 break;
8388         }
8389
8390         return insn - insn_buf;
8391 }
8392
8393 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8394                                                   struct bpf_insn *insn)
8395 {
8396         /* si->dst_reg = skb_shinfo(SKB); */
8397 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8398         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8399                               BPF_REG_AX, si->src_reg,
8400                               offsetof(struct sk_buff, end));
8401         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8402                               si->dst_reg, si->src_reg,
8403                               offsetof(struct sk_buff, head));
8404         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8405 #else
8406         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8407                               si->dst_reg, si->src_reg,
8408                               offsetof(struct sk_buff, end));
8409 #endif
8410
8411         return insn;
8412 }
8413
8414 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8415                                   const struct bpf_insn *si,
8416                                   struct bpf_insn *insn_buf,
8417                                   struct bpf_prog *prog, u32 *target_size)
8418 {
8419         struct bpf_insn *insn = insn_buf;
8420         int off;
8421
8422         switch (si->off) {
8423         case offsetof(struct __sk_buff, len):
8424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8425                                       bpf_target_off(struct sk_buff, len, 4,
8426                                                      target_size));
8427                 break;
8428
8429         case offsetof(struct __sk_buff, protocol):
8430                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8431                                       bpf_target_off(struct sk_buff, protocol, 2,
8432                                                      target_size));
8433                 break;
8434
8435         case offsetof(struct __sk_buff, vlan_proto):
8436                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8437                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8438                                                      target_size));
8439                 break;
8440
8441         case offsetof(struct __sk_buff, priority):
8442                 if (type == BPF_WRITE)
8443                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8444                                               bpf_target_off(struct sk_buff, priority, 4,
8445                                                              target_size));
8446                 else
8447                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8448                                               bpf_target_off(struct sk_buff, priority, 4,
8449                                                              target_size));
8450                 break;
8451
8452         case offsetof(struct __sk_buff, ingress_ifindex):
8453                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8454                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8455                                                      target_size));
8456                 break;
8457
8458         case offsetof(struct __sk_buff, ifindex):
8459                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8460                                       si->dst_reg, si->src_reg,
8461                                       offsetof(struct sk_buff, dev));
8462                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8463                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8464                                       bpf_target_off(struct net_device, ifindex, 4,
8465                                                      target_size));
8466                 break;
8467
8468         case offsetof(struct __sk_buff, hash):
8469                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8470                                       bpf_target_off(struct sk_buff, hash, 4,
8471                                                      target_size));
8472                 break;
8473
8474         case offsetof(struct __sk_buff, mark):
8475                 if (type == BPF_WRITE)
8476                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8477                                               bpf_target_off(struct sk_buff, mark, 4,
8478                                                              target_size));
8479                 else
8480                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8481                                               bpf_target_off(struct sk_buff, mark, 4,
8482                                                              target_size));
8483                 break;
8484
8485         case offsetof(struct __sk_buff, pkt_type):
8486                 *target_size = 1;
8487                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8488                                       PKT_TYPE_OFFSET());
8489                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8490 #ifdef __BIG_ENDIAN_BITFIELD
8491                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8492 #endif
8493                 break;
8494
8495         case offsetof(struct __sk_buff, queue_mapping):
8496                 if (type == BPF_WRITE) {
8497                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8498                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8499                                               bpf_target_off(struct sk_buff,
8500                                                              queue_mapping,
8501                                                              2, target_size));
8502                 } else {
8503                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8504                                               bpf_target_off(struct sk_buff,
8505                                                              queue_mapping,
8506                                                              2, target_size));
8507                 }
8508                 break;
8509
8510         case offsetof(struct __sk_buff, vlan_present):
8511                 *target_size = 1;
8512                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8513                                       PKT_VLAN_PRESENT_OFFSET());
8514                 if (PKT_VLAN_PRESENT_BIT)
8515                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8516                 if (PKT_VLAN_PRESENT_BIT < 7)
8517                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8518                 break;
8519
8520         case offsetof(struct __sk_buff, vlan_tci):
8521                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8522                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8523                                                      target_size));
8524                 break;
8525
8526         case offsetof(struct __sk_buff, cb[0]) ...
8527              offsetofend(struct __sk_buff, cb[4]) - 1:
8528                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8529                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8530                               offsetof(struct qdisc_skb_cb, data)) %
8531                              sizeof(__u64));
8532
8533                 prog->cb_access = 1;
8534                 off  = si->off;
8535                 off -= offsetof(struct __sk_buff, cb[0]);
8536                 off += offsetof(struct sk_buff, cb);
8537                 off += offsetof(struct qdisc_skb_cb, data);
8538                 if (type == BPF_WRITE)
8539                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8540                                               si->src_reg, off);
8541                 else
8542                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8543                                               si->src_reg, off);
8544                 break;
8545
8546         case offsetof(struct __sk_buff, tc_classid):
8547                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8548
8549                 off  = si->off;
8550                 off -= offsetof(struct __sk_buff, tc_classid);
8551                 off += offsetof(struct sk_buff, cb);
8552                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8553                 *target_size = 2;
8554                 if (type == BPF_WRITE)
8555                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8556                                               si->src_reg, off);
8557                 else
8558                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8559                                               si->src_reg, off);
8560                 break;
8561
8562         case offsetof(struct __sk_buff, data):
8563                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8564                                       si->dst_reg, si->src_reg,
8565                                       offsetof(struct sk_buff, data));
8566                 break;
8567
8568         case offsetof(struct __sk_buff, data_meta):
8569                 off  = si->off;
8570                 off -= offsetof(struct __sk_buff, data_meta);
8571                 off += offsetof(struct sk_buff, cb);
8572                 off += offsetof(struct bpf_skb_data_end, data_meta);
8573                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8574                                       si->src_reg, off);
8575                 break;
8576
8577         case offsetof(struct __sk_buff, data_end):
8578                 off  = si->off;
8579                 off -= offsetof(struct __sk_buff, data_end);
8580                 off += offsetof(struct sk_buff, cb);
8581                 off += offsetof(struct bpf_skb_data_end, data_end);
8582                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8583                                       si->src_reg, off);
8584                 break;
8585
8586         case offsetof(struct __sk_buff, tc_index):
8587 #ifdef CONFIG_NET_SCHED
8588                 if (type == BPF_WRITE)
8589                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8590                                               bpf_target_off(struct sk_buff, tc_index, 2,
8591                                                              target_size));
8592                 else
8593                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8594                                               bpf_target_off(struct sk_buff, tc_index, 2,
8595                                                              target_size));
8596 #else
8597                 *target_size = 2;
8598                 if (type == BPF_WRITE)
8599                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8600                 else
8601                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8602 #endif
8603                 break;
8604
8605         case offsetof(struct __sk_buff, napi_id):
8606 #if defined(CONFIG_NET_RX_BUSY_POLL)
8607                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8608                                       bpf_target_off(struct sk_buff, napi_id, 4,
8609                                                      target_size));
8610                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8611                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8612 #else
8613                 *target_size = 4;
8614                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8615 #endif
8616                 break;
8617         case offsetof(struct __sk_buff, family):
8618                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8619
8620                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8621                                       si->dst_reg, si->src_reg,
8622                                       offsetof(struct sk_buff, sk));
8623                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8624                                       bpf_target_off(struct sock_common,
8625                                                      skc_family,
8626                                                      2, target_size));
8627                 break;
8628         case offsetof(struct __sk_buff, remote_ip4):
8629                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8630
8631                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8632                                       si->dst_reg, si->src_reg,
8633                                       offsetof(struct sk_buff, sk));
8634                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8635                                       bpf_target_off(struct sock_common,
8636                                                      skc_daddr,
8637                                                      4, target_size));
8638                 break;
8639         case offsetof(struct __sk_buff, local_ip4):
8640                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8641                                           skc_rcv_saddr) != 4);
8642
8643                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8644                                       si->dst_reg, si->src_reg,
8645                                       offsetof(struct sk_buff, sk));
8646                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8647                                       bpf_target_off(struct sock_common,
8648                                                      skc_rcv_saddr,
8649                                                      4, target_size));
8650                 break;
8651         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8652              offsetof(struct __sk_buff, remote_ip6[3]):
8653 #if IS_ENABLED(CONFIG_IPV6)
8654                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8655                                           skc_v6_daddr.s6_addr32[0]) != 4);
8656
8657                 off = si->off;
8658                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8659
8660                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8661                                       si->dst_reg, si->src_reg,
8662                                       offsetof(struct sk_buff, sk));
8663                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8664                                       offsetof(struct sock_common,
8665                                                skc_v6_daddr.s6_addr32[0]) +
8666                                       off);
8667 #else
8668                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8669 #endif
8670                 break;
8671         case offsetof(struct __sk_buff, local_ip6[0]) ...
8672              offsetof(struct __sk_buff, local_ip6[3]):
8673 #if IS_ENABLED(CONFIG_IPV6)
8674                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8675                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8676
8677                 off = si->off;
8678                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8679
8680                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8681                                       si->dst_reg, si->src_reg,
8682                                       offsetof(struct sk_buff, sk));
8683                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8684                                       offsetof(struct sock_common,
8685                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8686                                       off);
8687 #else
8688                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8689 #endif
8690                 break;
8691
8692         case offsetof(struct __sk_buff, remote_port):
8693                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8694
8695                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8696                                       si->dst_reg, si->src_reg,
8697                                       offsetof(struct sk_buff, sk));
8698                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8699                                       bpf_target_off(struct sock_common,
8700                                                      skc_dport,
8701                                                      2, target_size));
8702 #ifndef __BIG_ENDIAN_BITFIELD
8703                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8704 #endif
8705                 break;
8706
8707         case offsetof(struct __sk_buff, local_port):
8708                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8709
8710                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8711                                       si->dst_reg, si->src_reg,
8712                                       offsetof(struct sk_buff, sk));
8713                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8714                                       bpf_target_off(struct sock_common,
8715                                                      skc_num, 2, target_size));
8716                 break;
8717
8718         case offsetof(struct __sk_buff, tstamp):
8719                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8720
8721                 if (type == BPF_WRITE)
8722                         *insn++ = BPF_STX_MEM(BPF_DW,
8723                                               si->dst_reg, si->src_reg,
8724                                               bpf_target_off(struct sk_buff,
8725                                                              tstamp, 8,
8726                                                              target_size));
8727                 else
8728                         *insn++ = BPF_LDX_MEM(BPF_DW,
8729                                               si->dst_reg, si->src_reg,
8730                                               bpf_target_off(struct sk_buff,
8731                                                              tstamp, 8,
8732                                                              target_size));
8733                 break;
8734
8735         case offsetof(struct __sk_buff, gso_segs):
8736                 insn = bpf_convert_shinfo_access(si, insn);
8737                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8738                                       si->dst_reg, si->dst_reg,
8739                                       bpf_target_off(struct skb_shared_info,
8740                                                      gso_segs, 2,
8741                                                      target_size));
8742                 break;
8743         case offsetof(struct __sk_buff, gso_size):
8744                 insn = bpf_convert_shinfo_access(si, insn);
8745                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8746                                       si->dst_reg, si->dst_reg,
8747                                       bpf_target_off(struct skb_shared_info,
8748                                                      gso_size, 2,
8749                                                      target_size));
8750                 break;
8751         case offsetof(struct __sk_buff, wire_len):
8752                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8753
8754                 off = si->off;
8755                 off -= offsetof(struct __sk_buff, wire_len);
8756                 off += offsetof(struct sk_buff, cb);
8757                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8758                 *target_size = 4;
8759                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8760                 break;
8761
8762         case offsetof(struct __sk_buff, sk):
8763                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8764                                       si->dst_reg, si->src_reg,
8765                                       offsetof(struct sk_buff, sk));
8766                 break;
8767         }
8768
8769         return insn - insn_buf;
8770 }
8771
8772 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8773                                 const struct bpf_insn *si,
8774                                 struct bpf_insn *insn_buf,
8775                                 struct bpf_prog *prog, u32 *target_size)
8776 {
8777         struct bpf_insn *insn = insn_buf;
8778         int off;
8779
8780         switch (si->off) {
8781         case offsetof(struct bpf_sock, bound_dev_if):
8782                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8783
8784                 if (type == BPF_WRITE)
8785                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8786                                         offsetof(struct sock, sk_bound_dev_if));
8787                 else
8788                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8789                                       offsetof(struct sock, sk_bound_dev_if));
8790                 break;
8791
8792         case offsetof(struct bpf_sock, mark):
8793                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8794
8795                 if (type == BPF_WRITE)
8796                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8797                                         offsetof(struct sock, sk_mark));
8798                 else
8799                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8800                                       offsetof(struct sock, sk_mark));
8801                 break;
8802
8803         case offsetof(struct bpf_sock, priority):
8804                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8805
8806                 if (type == BPF_WRITE)
8807                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8808                                         offsetof(struct sock, sk_priority));
8809                 else
8810                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8811                                       offsetof(struct sock, sk_priority));
8812                 break;
8813
8814         case offsetof(struct bpf_sock, family):
8815                 *insn++ = BPF_LDX_MEM(
8816                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8817                         si->dst_reg, si->src_reg,
8818                         bpf_target_off(struct sock_common,
8819                                        skc_family,
8820                                        sizeof_field(struct sock_common,
8821                                                     skc_family),
8822                                        target_size));
8823                 break;
8824
8825         case offsetof(struct bpf_sock, type):
8826                 *insn++ = BPF_LDX_MEM(
8827                         BPF_FIELD_SIZEOF(struct sock, sk_type),
8828                         si->dst_reg, si->src_reg,
8829                         bpf_target_off(struct sock, sk_type,
8830                                        sizeof_field(struct sock, sk_type),
8831                                        target_size));
8832                 break;
8833
8834         case offsetof(struct bpf_sock, protocol):
8835                 *insn++ = BPF_LDX_MEM(
8836                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8837                         si->dst_reg, si->src_reg,
8838                         bpf_target_off(struct sock, sk_protocol,
8839                                        sizeof_field(struct sock, sk_protocol),
8840                                        target_size));
8841                 break;
8842
8843         case offsetof(struct bpf_sock, src_ip4):
8844                 *insn++ = BPF_LDX_MEM(
8845                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8846                         bpf_target_off(struct sock_common, skc_rcv_saddr,
8847                                        sizeof_field(struct sock_common,
8848                                                     skc_rcv_saddr),
8849                                        target_size));
8850                 break;
8851
8852         case offsetof(struct bpf_sock, dst_ip4):
8853                 *insn++ = BPF_LDX_MEM(
8854                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8855                         bpf_target_off(struct sock_common, skc_daddr,
8856                                        sizeof_field(struct sock_common,
8857                                                     skc_daddr),
8858                                        target_size));
8859                 break;
8860
8861         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8862 #if IS_ENABLED(CONFIG_IPV6)
8863                 off = si->off;
8864                 off -= offsetof(struct bpf_sock, src_ip6[0]);
8865                 *insn++ = BPF_LDX_MEM(
8866                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8867                         bpf_target_off(
8868                                 struct sock_common,
8869                                 skc_v6_rcv_saddr.s6_addr32[0],
8870                                 sizeof_field(struct sock_common,
8871                                              skc_v6_rcv_saddr.s6_addr32[0]),
8872                                 target_size) + off);
8873 #else
8874                 (void)off;
8875                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8876 #endif
8877                 break;
8878
8879         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8880 #if IS_ENABLED(CONFIG_IPV6)
8881                 off = si->off;
8882                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8883                 *insn++ = BPF_LDX_MEM(
8884                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8885                         bpf_target_off(struct sock_common,
8886                                        skc_v6_daddr.s6_addr32[0],
8887                                        sizeof_field(struct sock_common,
8888                                                     skc_v6_daddr.s6_addr32[0]),
8889                                        target_size) + off);
8890 #else
8891                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8892                 *target_size = 4;
8893 #endif
8894                 break;
8895
8896         case offsetof(struct bpf_sock, src_port):
8897                 *insn++ = BPF_LDX_MEM(
8898                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8899                         si->dst_reg, si->src_reg,
8900                         bpf_target_off(struct sock_common, skc_num,
8901                                        sizeof_field(struct sock_common,
8902                                                     skc_num),
8903                                        target_size));
8904                 break;
8905
8906         case offsetof(struct bpf_sock, dst_port):
8907                 *insn++ = BPF_LDX_MEM(
8908                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8909                         si->dst_reg, si->src_reg,
8910                         bpf_target_off(struct sock_common, skc_dport,
8911                                        sizeof_field(struct sock_common,
8912                                                     skc_dport),
8913                                        target_size));
8914                 break;
8915
8916         case offsetof(struct bpf_sock, state):
8917                 *insn++ = BPF_LDX_MEM(
8918                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8919                         si->dst_reg, si->src_reg,
8920                         bpf_target_off(struct sock_common, skc_state,
8921                                        sizeof_field(struct sock_common,
8922                                                     skc_state),
8923                                        target_size));
8924                 break;
8925         case offsetof(struct bpf_sock, rx_queue_mapping):
8926 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8927                 *insn++ = BPF_LDX_MEM(
8928                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8929                         si->dst_reg, si->src_reg,
8930                         bpf_target_off(struct sock, sk_rx_queue_mapping,
8931                                        sizeof_field(struct sock,
8932                                                     sk_rx_queue_mapping),
8933                                        target_size));
8934                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8935                                       1);
8936                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8937 #else
8938                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8939                 *target_size = 2;
8940 #endif
8941                 break;
8942         }
8943
8944         return insn - insn_buf;
8945 }
8946
8947 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8948                                          const struct bpf_insn *si,
8949                                          struct bpf_insn *insn_buf,
8950                                          struct bpf_prog *prog, u32 *target_size)
8951 {
8952         struct bpf_insn *insn = insn_buf;
8953
8954         switch (si->off) {
8955         case offsetof(struct __sk_buff, ifindex):
8956                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8957                                       si->dst_reg, si->src_reg,
8958                                       offsetof(struct sk_buff, dev));
8959                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8960                                       bpf_target_off(struct net_device, ifindex, 4,
8961                                                      target_size));
8962                 break;
8963         default:
8964                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8965                                               target_size);
8966         }
8967
8968         return insn - insn_buf;
8969 }
8970
8971 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8972                                   const struct bpf_insn *si,
8973                                   struct bpf_insn *insn_buf,
8974                                   struct bpf_prog *prog, u32 *target_size)
8975 {
8976         struct bpf_insn *insn = insn_buf;
8977
8978         switch (si->off) {
8979         case offsetof(struct xdp_md, data):
8980                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8981                                       si->dst_reg, si->src_reg,
8982                                       offsetof(struct xdp_buff, data));
8983                 break;
8984         case offsetof(struct xdp_md, data_meta):
8985                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8986                                       si->dst_reg, si->src_reg,
8987                                       offsetof(struct xdp_buff, data_meta));
8988                 break;
8989         case offsetof(struct xdp_md, data_end):
8990                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8991                                       si->dst_reg, si->src_reg,
8992                                       offsetof(struct xdp_buff, data_end));
8993                 break;
8994         case offsetof(struct xdp_md, ingress_ifindex):
8995                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8996                                       si->dst_reg, si->src_reg,
8997                                       offsetof(struct xdp_buff, rxq));
8998                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8999                                       si->dst_reg, si->dst_reg,
9000                                       offsetof(struct xdp_rxq_info, dev));
9001                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9002                                       offsetof(struct net_device, ifindex));
9003                 break;
9004         case offsetof(struct xdp_md, rx_queue_index):
9005                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9006                                       si->dst_reg, si->src_reg,
9007                                       offsetof(struct xdp_buff, rxq));
9008                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9009                                       offsetof(struct xdp_rxq_info,
9010                                                queue_index));
9011                 break;
9012         case offsetof(struct xdp_md, egress_ifindex):
9013                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9014                                       si->dst_reg, si->src_reg,
9015                                       offsetof(struct xdp_buff, txq));
9016                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9017                                       si->dst_reg, si->dst_reg,
9018                                       offsetof(struct xdp_txq_info, dev));
9019                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9020                                       offsetof(struct net_device, ifindex));
9021                 break;
9022         }
9023
9024         return insn - insn_buf;
9025 }
9026
9027 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9028  * context Structure, F is Field in context structure that contains a pointer
9029  * to Nested Structure of type NS that has the field NF.
9030  *
9031  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9032  * sure that SIZE is not greater than actual size of S.F.NF.
9033  *
9034  * If offset OFF is provided, the load happens from that offset relative to
9035  * offset of NF.
9036  */
9037 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9038         do {                                                                   \
9039                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9040                                       si->src_reg, offsetof(S, F));            \
9041                 *insn++ = BPF_LDX_MEM(                                         \
9042                         SIZE, si->dst_reg, si->dst_reg,                        \
9043                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9044                                        target_size)                            \
9045                                 + OFF);                                        \
9046         } while (0)
9047
9048 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9049         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9050                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9051
9052 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9053  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9054  *
9055  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9056  * "register" since two registers available in convert_ctx_access are not
9057  * enough: we can't override neither SRC, since it contains value to store, nor
9058  * DST since it contains pointer to context that may be used by later
9059  * instructions. But we need a temporary place to save pointer to nested
9060  * structure whose field we want to store to.
9061  */
9062 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9063         do {                                                                   \
9064                 int tmp_reg = BPF_REG_9;                                       \
9065                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9066                         --tmp_reg;                                             \
9067                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9068                         --tmp_reg;                                             \
9069                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9070                                       offsetof(S, TF));                        \
9071                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9072                                       si->dst_reg, offsetof(S, F));            \
9073                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9074                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9075                                        target_size)                            \
9076                                 + OFF);                                        \
9077                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9078                                       offsetof(S, TF));                        \
9079         } while (0)
9080
9081 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9082                                                       TF)                      \
9083         do {                                                                   \
9084                 if (type == BPF_WRITE) {                                       \
9085                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9086                                                          OFF, TF);             \
9087                 } else {                                                       \
9088                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9089                                 S, NS, F, NF, SIZE, OFF);  \
9090                 }                                                              \
9091         } while (0)
9092
9093 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9094         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9095                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9096
9097 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9098                                         const struct bpf_insn *si,
9099                                         struct bpf_insn *insn_buf,
9100                                         struct bpf_prog *prog, u32 *target_size)
9101 {
9102         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9103         struct bpf_insn *insn = insn_buf;
9104
9105         switch (si->off) {
9106         case offsetof(struct bpf_sock_addr, user_family):
9107                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9108                                             struct sockaddr, uaddr, sa_family);
9109                 break;
9110
9111         case offsetof(struct bpf_sock_addr, user_ip4):
9112                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9113                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9114                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9115                 break;
9116
9117         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9118                 off = si->off;
9119                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9120                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9121                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9122                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9123                         tmp_reg);
9124                 break;
9125
9126         case offsetof(struct bpf_sock_addr, user_port):
9127                 /* To get port we need to know sa_family first and then treat
9128                  * sockaddr as either sockaddr_in or sockaddr_in6.
9129                  * Though we can simplify since port field has same offset and
9130                  * size in both structures.
9131                  * Here we check this invariant and use just one of the
9132                  * structures if it's true.
9133                  */
9134                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9135                              offsetof(struct sockaddr_in6, sin6_port));
9136                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9137                              sizeof_field(struct sockaddr_in6, sin6_port));
9138                 /* Account for sin6_port being smaller than user_port. */
9139                 port_size = min(port_size, BPF_LDST_BYTES(si));
9140                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9141                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9142                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9143                 break;
9144
9145         case offsetof(struct bpf_sock_addr, family):
9146                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9147                                             struct sock, sk, sk_family);
9148                 break;
9149
9150         case offsetof(struct bpf_sock_addr, type):
9151                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9152                                             struct sock, sk, sk_type);
9153                 break;
9154
9155         case offsetof(struct bpf_sock_addr, protocol):
9156                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9157                                             struct sock, sk, sk_protocol);
9158                 break;
9159
9160         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9161                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9162                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9163                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9164                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9165                 break;
9166
9167         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9168                                 msg_src_ip6[3]):
9169                 off = si->off;
9170                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9171                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9172                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9173                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9174                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9175                 break;
9176         case offsetof(struct bpf_sock_addr, sk):
9177                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9178                                       si->dst_reg, si->src_reg,
9179                                       offsetof(struct bpf_sock_addr_kern, sk));
9180                 break;
9181         }
9182
9183         return insn - insn_buf;
9184 }
9185
9186 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9187                                        const struct bpf_insn *si,
9188                                        struct bpf_insn *insn_buf,
9189                                        struct bpf_prog *prog,
9190                                        u32 *target_size)
9191 {
9192         struct bpf_insn *insn = insn_buf;
9193         int off;
9194
9195 /* Helper macro for adding read access to tcp_sock or sock fields. */
9196 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9197         do {                                                                  \
9198                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9199                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9200                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9201                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9202                         reg--;                                                \
9203                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9204                         reg--;                                                \
9205                 if (si->dst_reg == si->src_reg) {                             \
9206                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9207                                           offsetof(struct bpf_sock_ops_kern,  \
9208                                           temp));                             \
9209                         fullsock_reg = reg;                                   \
9210                         jmp += 2;                                             \
9211                 }                                                             \
9212                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9213                                                 struct bpf_sock_ops_kern,     \
9214                                                 is_fullsock),                 \
9215                                       fullsock_reg, si->src_reg,              \
9216                                       offsetof(struct bpf_sock_ops_kern,      \
9217                                                is_fullsock));                 \
9218                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9219                 if (si->dst_reg == si->src_reg)                               \
9220                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9221                                       offsetof(struct bpf_sock_ops_kern,      \
9222                                       temp));                                 \
9223                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9224                                                 struct bpf_sock_ops_kern, sk),\
9225                                       si->dst_reg, si->src_reg,               \
9226                                       offsetof(struct bpf_sock_ops_kern, sk));\
9227                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9228                                                        OBJ_FIELD),            \
9229                                       si->dst_reg, si->dst_reg,               \
9230                                       offsetof(OBJ, OBJ_FIELD));              \
9231                 if (si->dst_reg == si->src_reg) {                             \
9232                         *insn++ = BPF_JMP_A(1);                               \
9233                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9234                                       offsetof(struct bpf_sock_ops_kern,      \
9235                                       temp));                                 \
9236                 }                                                             \
9237         } while (0)
9238
9239 #define SOCK_OPS_GET_SK()                                                             \
9240         do {                                                                  \
9241                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9242                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9243                         reg--;                                                \
9244                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9245                         reg--;                                                \
9246                 if (si->dst_reg == si->src_reg) {                             \
9247                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9248                                           offsetof(struct bpf_sock_ops_kern,  \
9249                                           temp));                             \
9250                         fullsock_reg = reg;                                   \
9251                         jmp += 2;                                             \
9252                 }                                                             \
9253                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9254                                                 struct bpf_sock_ops_kern,     \
9255                                                 is_fullsock),                 \
9256                                       fullsock_reg, si->src_reg,              \
9257                                       offsetof(struct bpf_sock_ops_kern,      \
9258                                                is_fullsock));                 \
9259                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9260                 if (si->dst_reg == si->src_reg)                               \
9261                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9262                                       offsetof(struct bpf_sock_ops_kern,      \
9263                                       temp));                                 \
9264                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9265                                                 struct bpf_sock_ops_kern, sk),\
9266                                       si->dst_reg, si->src_reg,               \
9267                                       offsetof(struct bpf_sock_ops_kern, sk));\
9268                 if (si->dst_reg == si->src_reg) {                             \
9269                         *insn++ = BPF_JMP_A(1);                               \
9270                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9271                                       offsetof(struct bpf_sock_ops_kern,      \
9272                                       temp));                                 \
9273                 }                                                             \
9274         } while (0)
9275
9276 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9277                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9278
9279 /* Helper macro for adding write access to tcp_sock or sock fields.
9280  * The macro is called with two registers, dst_reg which contains a pointer
9281  * to ctx (context) and src_reg which contains the value that should be
9282  * stored. However, we need an additional register since we cannot overwrite
9283  * dst_reg because it may be used later in the program.
9284  * Instead we "borrow" one of the other register. We first save its value
9285  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9286  * it at the end of the macro.
9287  */
9288 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9289         do {                                                                  \
9290                 int reg = BPF_REG_9;                                          \
9291                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9292                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9293                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9294                         reg--;                                                \
9295                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9296                         reg--;                                                \
9297                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9298                                       offsetof(struct bpf_sock_ops_kern,      \
9299                                                temp));                        \
9300                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9301                                                 struct bpf_sock_ops_kern,     \
9302                                                 is_fullsock),                 \
9303                                       reg, si->dst_reg,                       \
9304                                       offsetof(struct bpf_sock_ops_kern,      \
9305                                                is_fullsock));                 \
9306                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9307                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9308                                                 struct bpf_sock_ops_kern, sk),\
9309                                       reg, si->dst_reg,                       \
9310                                       offsetof(struct bpf_sock_ops_kern, sk));\
9311                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9312                                       reg, si->src_reg,                       \
9313                                       offsetof(OBJ, OBJ_FIELD));              \
9314                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9315                                       offsetof(struct bpf_sock_ops_kern,      \
9316                                                temp));                        \
9317         } while (0)
9318
9319 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9320         do {                                                                  \
9321                 if (TYPE == BPF_WRITE)                                        \
9322                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9323                 else                                                          \
9324                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9325         } while (0)
9326
9327         if (insn > insn_buf)
9328                 return insn - insn_buf;
9329
9330         switch (si->off) {
9331         case offsetof(struct bpf_sock_ops, op):
9332                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9333                                                        op),
9334                                       si->dst_reg, si->src_reg,
9335                                       offsetof(struct bpf_sock_ops_kern, op));
9336                 break;
9337
9338         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9339              offsetof(struct bpf_sock_ops, replylong[3]):
9340                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9341                              sizeof_field(struct bpf_sock_ops_kern, reply));
9342                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9343                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9344                 off = si->off;
9345                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9346                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9347                 if (type == BPF_WRITE)
9348                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9349                                               off);
9350                 else
9351                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9352                                               off);
9353                 break;
9354
9355         case offsetof(struct bpf_sock_ops, family):
9356                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9357
9358                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9359                                               struct bpf_sock_ops_kern, sk),
9360                                       si->dst_reg, si->src_reg,
9361                                       offsetof(struct bpf_sock_ops_kern, sk));
9362                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9363                                       offsetof(struct sock_common, skc_family));
9364                 break;
9365
9366         case offsetof(struct bpf_sock_ops, remote_ip4):
9367                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9368
9369                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9370                                                 struct bpf_sock_ops_kern, sk),
9371                                       si->dst_reg, si->src_reg,
9372                                       offsetof(struct bpf_sock_ops_kern, sk));
9373                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9374                                       offsetof(struct sock_common, skc_daddr));
9375                 break;
9376
9377         case offsetof(struct bpf_sock_ops, local_ip4):
9378                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9379                                           skc_rcv_saddr) != 4);
9380
9381                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9382                                               struct bpf_sock_ops_kern, sk),
9383                                       si->dst_reg, si->src_reg,
9384                                       offsetof(struct bpf_sock_ops_kern, sk));
9385                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9386                                       offsetof(struct sock_common,
9387                                                skc_rcv_saddr));
9388                 break;
9389
9390         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9391              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9392 #if IS_ENABLED(CONFIG_IPV6)
9393                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9394                                           skc_v6_daddr.s6_addr32[0]) != 4);
9395
9396                 off = si->off;
9397                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9398                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9399                                                 struct bpf_sock_ops_kern, sk),
9400                                       si->dst_reg, si->src_reg,
9401                                       offsetof(struct bpf_sock_ops_kern, sk));
9402                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9403                                       offsetof(struct sock_common,
9404                                                skc_v6_daddr.s6_addr32[0]) +
9405                                       off);
9406 #else
9407                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9408 #endif
9409                 break;
9410
9411         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9412              offsetof(struct bpf_sock_ops, local_ip6[3]):
9413 #if IS_ENABLED(CONFIG_IPV6)
9414                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9415                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9416
9417                 off = si->off;
9418                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9419                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9420                                                 struct bpf_sock_ops_kern, sk),
9421                                       si->dst_reg, si->src_reg,
9422                                       offsetof(struct bpf_sock_ops_kern, sk));
9423                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9424                                       offsetof(struct sock_common,
9425                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9426                                       off);
9427 #else
9428                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9429 #endif
9430                 break;
9431
9432         case offsetof(struct bpf_sock_ops, remote_port):
9433                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9434
9435                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9436                                                 struct bpf_sock_ops_kern, sk),
9437                                       si->dst_reg, si->src_reg,
9438                                       offsetof(struct bpf_sock_ops_kern, sk));
9439                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9440                                       offsetof(struct sock_common, skc_dport));
9441 #ifndef __BIG_ENDIAN_BITFIELD
9442                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9443 #endif
9444                 break;
9445
9446         case offsetof(struct bpf_sock_ops, local_port):
9447                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9448
9449                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9450                                                 struct bpf_sock_ops_kern, sk),
9451                                       si->dst_reg, si->src_reg,
9452                                       offsetof(struct bpf_sock_ops_kern, sk));
9453                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9454                                       offsetof(struct sock_common, skc_num));
9455                 break;
9456
9457         case offsetof(struct bpf_sock_ops, is_fullsock):
9458                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9459                                                 struct bpf_sock_ops_kern,
9460                                                 is_fullsock),
9461                                       si->dst_reg, si->src_reg,
9462                                       offsetof(struct bpf_sock_ops_kern,
9463                                                is_fullsock));
9464                 break;
9465
9466         case offsetof(struct bpf_sock_ops, state):
9467                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9468
9469                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9470                                                 struct bpf_sock_ops_kern, sk),
9471                                       si->dst_reg, si->src_reg,
9472                                       offsetof(struct bpf_sock_ops_kern, sk));
9473                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9474                                       offsetof(struct sock_common, skc_state));
9475                 break;
9476
9477         case offsetof(struct bpf_sock_ops, rtt_min):
9478                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9479                              sizeof(struct minmax));
9480                 BUILD_BUG_ON(sizeof(struct minmax) <
9481                              sizeof(struct minmax_sample));
9482
9483                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9484                                                 struct bpf_sock_ops_kern, sk),
9485                                       si->dst_reg, si->src_reg,
9486                                       offsetof(struct bpf_sock_ops_kern, sk));
9487                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9488                                       offsetof(struct tcp_sock, rtt_min) +
9489                                       sizeof_field(struct minmax_sample, t));
9490                 break;
9491
9492         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9493                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9494                                    struct tcp_sock);
9495                 break;
9496
9497         case offsetof(struct bpf_sock_ops, sk_txhash):
9498                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9499                                           struct sock, type);
9500                 break;
9501         case offsetof(struct bpf_sock_ops, snd_cwnd):
9502                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9503                 break;
9504         case offsetof(struct bpf_sock_ops, srtt_us):
9505                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9506                 break;
9507         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9508                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9509                 break;
9510         case offsetof(struct bpf_sock_ops, rcv_nxt):
9511                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9512                 break;
9513         case offsetof(struct bpf_sock_ops, snd_nxt):
9514                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9515                 break;
9516         case offsetof(struct bpf_sock_ops, snd_una):
9517                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9518                 break;
9519         case offsetof(struct bpf_sock_ops, mss_cache):
9520                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9521                 break;
9522         case offsetof(struct bpf_sock_ops, ecn_flags):
9523                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9524                 break;
9525         case offsetof(struct bpf_sock_ops, rate_delivered):
9526                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9527                 break;
9528         case offsetof(struct bpf_sock_ops, rate_interval_us):
9529                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9530                 break;
9531         case offsetof(struct bpf_sock_ops, packets_out):
9532                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9533                 break;
9534         case offsetof(struct bpf_sock_ops, retrans_out):
9535                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9536                 break;
9537         case offsetof(struct bpf_sock_ops, total_retrans):
9538                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9539                 break;
9540         case offsetof(struct bpf_sock_ops, segs_in):
9541                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9542                 break;
9543         case offsetof(struct bpf_sock_ops, data_segs_in):
9544                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9545                 break;
9546         case offsetof(struct bpf_sock_ops, segs_out):
9547                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9548                 break;
9549         case offsetof(struct bpf_sock_ops, data_segs_out):
9550                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9551                 break;
9552         case offsetof(struct bpf_sock_ops, lost_out):
9553                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9554                 break;
9555         case offsetof(struct bpf_sock_ops, sacked_out):
9556                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9557                 break;
9558         case offsetof(struct bpf_sock_ops, bytes_received):
9559                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9560                 break;
9561         case offsetof(struct bpf_sock_ops, bytes_acked):
9562                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9563                 break;
9564         case offsetof(struct bpf_sock_ops, sk):
9565                 SOCK_OPS_GET_SK();
9566                 break;
9567         case offsetof(struct bpf_sock_ops, skb_data_end):
9568                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9569                                                        skb_data_end),
9570                                       si->dst_reg, si->src_reg,
9571                                       offsetof(struct bpf_sock_ops_kern,
9572                                                skb_data_end));
9573                 break;
9574         case offsetof(struct bpf_sock_ops, skb_data):
9575                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9576                                                        skb),
9577                                       si->dst_reg, si->src_reg,
9578                                       offsetof(struct bpf_sock_ops_kern,
9579                                                skb));
9580                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9581                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9582                                       si->dst_reg, si->dst_reg,
9583                                       offsetof(struct sk_buff, data));
9584                 break;
9585         case offsetof(struct bpf_sock_ops, skb_len):
9586                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9587                                                        skb),
9588                                       si->dst_reg, si->src_reg,
9589                                       offsetof(struct bpf_sock_ops_kern,
9590                                                skb));
9591                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9592                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9593                                       si->dst_reg, si->dst_reg,
9594                                       offsetof(struct sk_buff, len));
9595                 break;
9596         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9597                 off = offsetof(struct sk_buff, cb);
9598                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9599                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9600                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9601                                                        skb),
9602                                       si->dst_reg, si->src_reg,
9603                                       offsetof(struct bpf_sock_ops_kern,
9604                                                skb));
9605                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9606                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9607                                                        tcp_flags),
9608                                       si->dst_reg, si->dst_reg, off);
9609                 break;
9610         }
9611         return insn - insn_buf;
9612 }
9613
9614 /* data_end = skb->data + skb_headlen() */
9615 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9616                                                     struct bpf_insn *insn)
9617 {
9618         /* si->dst_reg = skb->data */
9619         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9620                               si->dst_reg, si->src_reg,
9621                               offsetof(struct sk_buff, data));
9622         /* AX = skb->len */
9623         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9624                               BPF_REG_AX, si->src_reg,
9625                               offsetof(struct sk_buff, len));
9626         /* si->dst_reg = skb->data + skb->len */
9627         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9628         /* AX = skb->data_len */
9629         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9630                               BPF_REG_AX, si->src_reg,
9631                               offsetof(struct sk_buff, data_len));
9632         /* si->dst_reg = skb->data + skb->len - skb->data_len */
9633         *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9634
9635         return insn;
9636 }
9637
9638 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9639                                      const struct bpf_insn *si,
9640                                      struct bpf_insn *insn_buf,
9641                                      struct bpf_prog *prog, u32 *target_size)
9642 {
9643         struct bpf_insn *insn = insn_buf;
9644
9645         switch (si->off) {
9646         case offsetof(struct __sk_buff, data_end):
9647                 insn = bpf_convert_data_end_access(si, insn);
9648                 break;
9649         default:
9650                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9651                                               target_size);
9652         }
9653
9654         return insn - insn_buf;
9655 }
9656
9657 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9658                                      const struct bpf_insn *si,
9659                                      struct bpf_insn *insn_buf,
9660                                      struct bpf_prog *prog, u32 *target_size)
9661 {
9662         struct bpf_insn *insn = insn_buf;
9663 #if IS_ENABLED(CONFIG_IPV6)
9664         int off;
9665 #endif
9666
9667         /* convert ctx uses the fact sg element is first in struct */
9668         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9669
9670         switch (si->off) {
9671         case offsetof(struct sk_msg_md, data):
9672                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9673                                       si->dst_reg, si->src_reg,
9674                                       offsetof(struct sk_msg, data));
9675                 break;
9676         case offsetof(struct sk_msg_md, data_end):
9677                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9678                                       si->dst_reg, si->src_reg,
9679                                       offsetof(struct sk_msg, data_end));
9680                 break;
9681         case offsetof(struct sk_msg_md, family):
9682                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9683
9684                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9685                                               struct sk_msg, sk),
9686                                       si->dst_reg, si->src_reg,
9687                                       offsetof(struct sk_msg, sk));
9688                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9689                                       offsetof(struct sock_common, skc_family));
9690                 break;
9691
9692         case offsetof(struct sk_msg_md, remote_ip4):
9693                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9694
9695                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9696                                                 struct sk_msg, sk),
9697                                       si->dst_reg, si->src_reg,
9698                                       offsetof(struct sk_msg, sk));
9699                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9700                                       offsetof(struct sock_common, skc_daddr));
9701                 break;
9702
9703         case offsetof(struct sk_msg_md, local_ip4):
9704                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9705                                           skc_rcv_saddr) != 4);
9706
9707                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9708                                               struct sk_msg, sk),
9709                                       si->dst_reg, si->src_reg,
9710                                       offsetof(struct sk_msg, sk));
9711                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9712                                       offsetof(struct sock_common,
9713                                                skc_rcv_saddr));
9714                 break;
9715
9716         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9717              offsetof(struct sk_msg_md, remote_ip6[3]):
9718 #if IS_ENABLED(CONFIG_IPV6)
9719                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9720                                           skc_v6_daddr.s6_addr32[0]) != 4);
9721
9722                 off = si->off;
9723                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9724                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9725                                                 struct sk_msg, sk),
9726                                       si->dst_reg, si->src_reg,
9727                                       offsetof(struct sk_msg, sk));
9728                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9729                                       offsetof(struct sock_common,
9730                                                skc_v6_daddr.s6_addr32[0]) +
9731                                       off);
9732 #else
9733                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9734 #endif
9735                 break;
9736
9737         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9738              offsetof(struct sk_msg_md, local_ip6[3]):
9739 #if IS_ENABLED(CONFIG_IPV6)
9740                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9741                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9742
9743                 off = si->off;
9744                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9745                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9746                                                 struct sk_msg, sk),
9747                                       si->dst_reg, si->src_reg,
9748                                       offsetof(struct sk_msg, sk));
9749                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9750                                       offsetof(struct sock_common,
9751                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9752                                       off);
9753 #else
9754                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9755 #endif
9756                 break;
9757
9758         case offsetof(struct sk_msg_md, remote_port):
9759                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9760
9761                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9762                                                 struct sk_msg, sk),
9763                                       si->dst_reg, si->src_reg,
9764                                       offsetof(struct sk_msg, sk));
9765                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9766                                       offsetof(struct sock_common, skc_dport));
9767 #ifndef __BIG_ENDIAN_BITFIELD
9768                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9769 #endif
9770                 break;
9771
9772         case offsetof(struct sk_msg_md, local_port):
9773                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9774
9775                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9776                                                 struct sk_msg, sk),
9777                                       si->dst_reg, si->src_reg,
9778                                       offsetof(struct sk_msg, sk));
9779                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9780                                       offsetof(struct sock_common, skc_num));
9781                 break;
9782
9783         case offsetof(struct sk_msg_md, size):
9784                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9785                                       si->dst_reg, si->src_reg,
9786                                       offsetof(struct sk_msg_sg, size));
9787                 break;
9788
9789         case offsetof(struct sk_msg_md, sk):
9790                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9791                                       si->dst_reg, si->src_reg,
9792                                       offsetof(struct sk_msg, sk));
9793                 break;
9794         }
9795
9796         return insn - insn_buf;
9797 }
9798
9799 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9800         .get_func_proto         = sk_filter_func_proto,
9801         .is_valid_access        = sk_filter_is_valid_access,
9802         .convert_ctx_access     = bpf_convert_ctx_access,
9803         .gen_ld_abs             = bpf_gen_ld_abs,
9804 };
9805
9806 const struct bpf_prog_ops sk_filter_prog_ops = {
9807         .test_run               = bpf_prog_test_run_skb,
9808 };
9809
9810 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9811         .get_func_proto         = tc_cls_act_func_proto,
9812         .is_valid_access        = tc_cls_act_is_valid_access,
9813         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
9814         .gen_prologue           = tc_cls_act_prologue,
9815         .gen_ld_abs             = bpf_gen_ld_abs,
9816         .check_kfunc_call       = bpf_prog_test_check_kfunc_call,
9817 };
9818
9819 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9820         .test_run               = bpf_prog_test_run_skb,
9821 };
9822
9823 const struct bpf_verifier_ops xdp_verifier_ops = {
9824         .get_func_proto         = xdp_func_proto,
9825         .is_valid_access        = xdp_is_valid_access,
9826         .convert_ctx_access     = xdp_convert_ctx_access,
9827         .gen_prologue           = bpf_noop_prologue,
9828 };
9829
9830 const struct bpf_prog_ops xdp_prog_ops = {
9831         .test_run               = bpf_prog_test_run_xdp,
9832 };
9833
9834 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9835         .get_func_proto         = cg_skb_func_proto,
9836         .is_valid_access        = cg_skb_is_valid_access,
9837         .convert_ctx_access     = bpf_convert_ctx_access,
9838 };
9839
9840 const struct bpf_prog_ops cg_skb_prog_ops = {
9841         .test_run               = bpf_prog_test_run_skb,
9842 };
9843
9844 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9845         .get_func_proto         = lwt_in_func_proto,
9846         .is_valid_access        = lwt_is_valid_access,
9847         .convert_ctx_access     = bpf_convert_ctx_access,
9848 };
9849
9850 const struct bpf_prog_ops lwt_in_prog_ops = {
9851         .test_run               = bpf_prog_test_run_skb,
9852 };
9853
9854 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9855         .get_func_proto         = lwt_out_func_proto,
9856         .is_valid_access        = lwt_is_valid_access,
9857         .convert_ctx_access     = bpf_convert_ctx_access,
9858 };
9859
9860 const struct bpf_prog_ops lwt_out_prog_ops = {
9861         .test_run               = bpf_prog_test_run_skb,
9862 };
9863
9864 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9865         .get_func_proto         = lwt_xmit_func_proto,
9866         .is_valid_access        = lwt_is_valid_access,
9867         .convert_ctx_access     = bpf_convert_ctx_access,
9868         .gen_prologue           = tc_cls_act_prologue,
9869 };
9870
9871 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9872         .test_run               = bpf_prog_test_run_skb,
9873 };
9874
9875 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9876         .get_func_proto         = lwt_seg6local_func_proto,
9877         .is_valid_access        = lwt_is_valid_access,
9878         .convert_ctx_access     = bpf_convert_ctx_access,
9879 };
9880
9881 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9882         .test_run               = bpf_prog_test_run_skb,
9883 };
9884
9885 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9886         .get_func_proto         = sock_filter_func_proto,
9887         .is_valid_access        = sock_filter_is_valid_access,
9888         .convert_ctx_access     = bpf_sock_convert_ctx_access,
9889 };
9890
9891 const struct bpf_prog_ops cg_sock_prog_ops = {
9892 };
9893
9894 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9895         .get_func_proto         = sock_addr_func_proto,
9896         .is_valid_access        = sock_addr_is_valid_access,
9897         .convert_ctx_access     = sock_addr_convert_ctx_access,
9898 };
9899
9900 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9901 };
9902
9903 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9904         .get_func_proto         = sock_ops_func_proto,
9905         .is_valid_access        = sock_ops_is_valid_access,
9906         .convert_ctx_access     = sock_ops_convert_ctx_access,
9907 };
9908
9909 const struct bpf_prog_ops sock_ops_prog_ops = {
9910 };
9911
9912 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9913         .get_func_proto         = sk_skb_func_proto,
9914         .is_valid_access        = sk_skb_is_valid_access,
9915         .convert_ctx_access     = sk_skb_convert_ctx_access,
9916         .gen_prologue           = sk_skb_prologue,
9917 };
9918
9919 const struct bpf_prog_ops sk_skb_prog_ops = {
9920 };
9921
9922 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9923         .get_func_proto         = sk_msg_func_proto,
9924         .is_valid_access        = sk_msg_is_valid_access,
9925         .convert_ctx_access     = sk_msg_convert_ctx_access,
9926         .gen_prologue           = bpf_noop_prologue,
9927 };
9928
9929 const struct bpf_prog_ops sk_msg_prog_ops = {
9930 };
9931
9932 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9933         .get_func_proto         = flow_dissector_func_proto,
9934         .is_valid_access        = flow_dissector_is_valid_access,
9935         .convert_ctx_access     = flow_dissector_convert_ctx_access,
9936 };
9937
9938 const struct bpf_prog_ops flow_dissector_prog_ops = {
9939         .test_run               = bpf_prog_test_run_flow_dissector,
9940 };
9941
9942 int sk_detach_filter(struct sock *sk)
9943 {
9944         int ret = -ENOENT;
9945         struct sk_filter *filter;
9946
9947         if (sock_flag(sk, SOCK_FILTER_LOCKED))
9948                 return -EPERM;
9949
9950         filter = rcu_dereference_protected(sk->sk_filter,
9951                                            lockdep_sock_is_held(sk));
9952         if (filter) {
9953                 RCU_INIT_POINTER(sk->sk_filter, NULL);
9954                 sk_filter_uncharge(sk, filter);
9955                 ret = 0;
9956         }
9957
9958         return ret;
9959 }
9960 EXPORT_SYMBOL_GPL(sk_detach_filter);
9961
9962 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9963                   unsigned int len)
9964 {
9965         struct sock_fprog_kern *fprog;
9966         struct sk_filter *filter;
9967         int ret = 0;
9968
9969         lock_sock(sk);
9970         filter = rcu_dereference_protected(sk->sk_filter,
9971                                            lockdep_sock_is_held(sk));
9972         if (!filter)
9973                 goto out;
9974
9975         /* We're copying the filter that has been originally attached,
9976          * so no conversion/decode needed anymore. eBPF programs that
9977          * have no original program cannot be dumped through this.
9978          */
9979         ret = -EACCES;
9980         fprog = filter->prog->orig_prog;
9981         if (!fprog)
9982                 goto out;
9983
9984         ret = fprog->len;
9985         if (!len)
9986                 /* User space only enquires number of filter blocks. */
9987                 goto out;
9988
9989         ret = -EINVAL;
9990         if (len < fprog->len)
9991                 goto out;
9992
9993         ret = -EFAULT;
9994         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9995                 goto out;
9996
9997         /* Instead of bytes, the API requests to return the number
9998          * of filter blocks.
9999          */
10000         ret = fprog->len;
10001 out:
10002         release_sock(sk);
10003         return ret;
10004 }
10005
10006 #ifdef CONFIG_INET
10007 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10008                                     struct sock_reuseport *reuse,
10009                                     struct sock *sk, struct sk_buff *skb,
10010                                     u32 hash)
10011 {
10012         reuse_kern->skb = skb;
10013         reuse_kern->sk = sk;
10014         reuse_kern->selected_sk = NULL;
10015         reuse_kern->data_end = skb->data + skb_headlen(skb);
10016         reuse_kern->hash = hash;
10017         reuse_kern->reuseport_id = reuse->reuseport_id;
10018         reuse_kern->bind_inany = reuse->bind_inany;
10019 }
10020
10021 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10022                                   struct bpf_prog *prog, struct sk_buff *skb,
10023                                   u32 hash)
10024 {
10025         struct sk_reuseport_kern reuse_kern;
10026         enum sk_action action;
10027
10028         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
10029         action = BPF_PROG_RUN(prog, &reuse_kern);
10030
10031         if (action == SK_PASS)
10032                 return reuse_kern.selected_sk;
10033         else
10034                 return ERR_PTR(-ECONNREFUSED);
10035 }
10036
10037 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10038            struct bpf_map *, map, void *, key, u32, flags)
10039 {
10040         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10041         struct sock_reuseport *reuse;
10042         struct sock *selected_sk;
10043
10044         selected_sk = map->ops->map_lookup_elem(map, key);
10045         if (!selected_sk)
10046                 return -ENOENT;
10047
10048         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10049         if (!reuse) {
10050                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10051                 if (sk_is_refcounted(selected_sk))
10052                         sock_put(selected_sk);
10053
10054                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10055                  * The only (!reuse) case here is - the sk has already been
10056                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10057                  *
10058                  * Other maps (e.g. sock_map) do not provide this guarantee and
10059                  * the sk may never be in the reuseport group to begin with.
10060                  */
10061                 return is_sockarray ? -ENOENT : -EINVAL;
10062         }
10063
10064         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10065                 struct sock *sk = reuse_kern->sk;
10066
10067                 if (sk->sk_protocol != selected_sk->sk_protocol)
10068                         return -EPROTOTYPE;
10069                 else if (sk->sk_family != selected_sk->sk_family)
10070                         return -EAFNOSUPPORT;
10071
10072                 /* Catch all. Likely bound to a different sockaddr. */
10073                 return -EBADFD;
10074         }
10075
10076         reuse_kern->selected_sk = selected_sk;
10077
10078         return 0;
10079 }
10080
10081 static const struct bpf_func_proto sk_select_reuseport_proto = {
10082         .func           = sk_select_reuseport,
10083         .gpl_only       = false,
10084         .ret_type       = RET_INTEGER,
10085         .arg1_type      = ARG_PTR_TO_CTX,
10086         .arg2_type      = ARG_CONST_MAP_PTR,
10087         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10088         .arg4_type      = ARG_ANYTHING,
10089 };
10090
10091 BPF_CALL_4(sk_reuseport_load_bytes,
10092            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10093            void *, to, u32, len)
10094 {
10095         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10096 }
10097
10098 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10099         .func           = sk_reuseport_load_bytes,
10100         .gpl_only       = false,
10101         .ret_type       = RET_INTEGER,
10102         .arg1_type      = ARG_PTR_TO_CTX,
10103         .arg2_type      = ARG_ANYTHING,
10104         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10105         .arg4_type      = ARG_CONST_SIZE,
10106 };
10107
10108 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10109            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10110            void *, to, u32, len, u32, start_header)
10111 {
10112         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10113                                                len, start_header);
10114 }
10115
10116 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10117         .func           = sk_reuseport_load_bytes_relative,
10118         .gpl_only       = false,
10119         .ret_type       = RET_INTEGER,
10120         .arg1_type      = ARG_PTR_TO_CTX,
10121         .arg2_type      = ARG_ANYTHING,
10122         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10123         .arg4_type      = ARG_CONST_SIZE,
10124         .arg5_type      = ARG_ANYTHING,
10125 };
10126
10127 static const struct bpf_func_proto *
10128 sk_reuseport_func_proto(enum bpf_func_id func_id,
10129                         const struct bpf_prog *prog)
10130 {
10131         switch (func_id) {
10132         case BPF_FUNC_sk_select_reuseport:
10133                 return &sk_select_reuseport_proto;
10134         case BPF_FUNC_skb_load_bytes:
10135                 return &sk_reuseport_load_bytes_proto;
10136         case BPF_FUNC_skb_load_bytes_relative:
10137                 return &sk_reuseport_load_bytes_relative_proto;
10138         default:
10139                 return bpf_base_func_proto(func_id);
10140         }
10141 }
10142
10143 static bool
10144 sk_reuseport_is_valid_access(int off, int size,
10145                              enum bpf_access_type type,
10146                              const struct bpf_prog *prog,
10147                              struct bpf_insn_access_aux *info)
10148 {
10149         const u32 size_default = sizeof(__u32);
10150
10151         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10152             off % size || type != BPF_READ)
10153                 return false;
10154
10155         switch (off) {
10156         case offsetof(struct sk_reuseport_md, data):
10157                 info->reg_type = PTR_TO_PACKET;
10158                 return size == sizeof(__u64);
10159
10160         case offsetof(struct sk_reuseport_md, data_end):
10161                 info->reg_type = PTR_TO_PACKET_END;
10162                 return size == sizeof(__u64);
10163
10164         case offsetof(struct sk_reuseport_md, hash):
10165                 return size == size_default;
10166
10167         /* Fields that allow narrowing */
10168         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10169                 if (size < sizeof_field(struct sk_buff, protocol))
10170                         return false;
10171                 fallthrough;
10172         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10173         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10174         case bpf_ctx_range(struct sk_reuseport_md, len):
10175                 bpf_ctx_record_field_size(info, size_default);
10176                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10177
10178         default:
10179                 return false;
10180         }
10181 }
10182
10183 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10184         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10185                               si->dst_reg, si->src_reg,                 \
10186                               bpf_target_off(struct sk_reuseport_kern, F, \
10187                                              sizeof_field(struct sk_reuseport_kern, F), \
10188                                              target_size));             \
10189         })
10190
10191 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10192         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10193                                     struct sk_buff,                     \
10194                                     skb,                                \
10195                                     SKB_FIELD)
10196
10197 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10198         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10199                                     struct sock,                        \
10200                                     sk,                                 \
10201                                     SK_FIELD)
10202
10203 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10204                                            const struct bpf_insn *si,
10205                                            struct bpf_insn *insn_buf,
10206                                            struct bpf_prog *prog,
10207                                            u32 *target_size)
10208 {
10209         struct bpf_insn *insn = insn_buf;
10210
10211         switch (si->off) {
10212         case offsetof(struct sk_reuseport_md, data):
10213                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10214                 break;
10215
10216         case offsetof(struct sk_reuseport_md, len):
10217                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10218                 break;
10219
10220         case offsetof(struct sk_reuseport_md, eth_protocol):
10221                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10222                 break;
10223
10224         case offsetof(struct sk_reuseport_md, ip_protocol):
10225                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10226                 break;
10227
10228         case offsetof(struct sk_reuseport_md, data_end):
10229                 SK_REUSEPORT_LOAD_FIELD(data_end);
10230                 break;
10231
10232         case offsetof(struct sk_reuseport_md, hash):
10233                 SK_REUSEPORT_LOAD_FIELD(hash);
10234                 break;
10235
10236         case offsetof(struct sk_reuseport_md, bind_inany):
10237                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10238                 break;
10239         }
10240
10241         return insn - insn_buf;
10242 }
10243
10244 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10245         .get_func_proto         = sk_reuseport_func_proto,
10246         .is_valid_access        = sk_reuseport_is_valid_access,
10247         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10248 };
10249
10250 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10251 };
10252
10253 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10254 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10255
10256 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10257            struct sock *, sk, u64, flags)
10258 {
10259         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10260                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10261                 return -EINVAL;
10262         if (unlikely(sk && sk_is_refcounted(sk)))
10263                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10264         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10265                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10266
10267         /* Check if socket is suitable for packet L3/L4 protocol */
10268         if (sk && sk->sk_protocol != ctx->protocol)
10269                 return -EPROTOTYPE;
10270         if (sk && sk->sk_family != ctx->family &&
10271             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10272                 return -EAFNOSUPPORT;
10273
10274         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10275                 return -EEXIST;
10276
10277         /* Select socket as lookup result */
10278         ctx->selected_sk = sk;
10279         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10280         return 0;
10281 }
10282
10283 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10284         .func           = bpf_sk_lookup_assign,
10285         .gpl_only       = false,
10286         .ret_type       = RET_INTEGER,
10287         .arg1_type      = ARG_PTR_TO_CTX,
10288         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10289         .arg3_type      = ARG_ANYTHING,
10290 };
10291
10292 static const struct bpf_func_proto *
10293 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10294 {
10295         switch (func_id) {
10296         case BPF_FUNC_perf_event_output:
10297                 return &bpf_event_output_data_proto;
10298         case BPF_FUNC_sk_assign:
10299                 return &bpf_sk_lookup_assign_proto;
10300         case BPF_FUNC_sk_release:
10301                 return &bpf_sk_release_proto;
10302         default:
10303                 return bpf_sk_base_func_proto(func_id);
10304         }
10305 }
10306
10307 static bool sk_lookup_is_valid_access(int off, int size,
10308                                       enum bpf_access_type type,
10309                                       const struct bpf_prog *prog,
10310                                       struct bpf_insn_access_aux *info)
10311 {
10312         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10313                 return false;
10314         if (off % size != 0)
10315                 return false;
10316         if (type != BPF_READ)
10317                 return false;
10318
10319         switch (off) {
10320         case offsetof(struct bpf_sk_lookup, sk):
10321                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10322                 return size == sizeof(__u64);
10323
10324         case bpf_ctx_range(struct bpf_sk_lookup, family):
10325         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10326         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10327         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10328         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10329         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10330         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10331         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10332                 bpf_ctx_record_field_size(info, sizeof(__u32));
10333                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10334
10335         default:
10336                 return false;
10337         }
10338 }
10339
10340 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10341                                         const struct bpf_insn *si,
10342                                         struct bpf_insn *insn_buf,
10343                                         struct bpf_prog *prog,
10344                                         u32 *target_size)
10345 {
10346         struct bpf_insn *insn = insn_buf;
10347
10348         switch (si->off) {
10349         case offsetof(struct bpf_sk_lookup, sk):
10350                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10351                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10352                 break;
10353
10354         case offsetof(struct bpf_sk_lookup, family):
10355                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10356                                       bpf_target_off(struct bpf_sk_lookup_kern,
10357                                                      family, 2, target_size));
10358                 break;
10359
10360         case offsetof(struct bpf_sk_lookup, protocol):
10361                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10362                                       bpf_target_off(struct bpf_sk_lookup_kern,
10363                                                      protocol, 2, target_size));
10364                 break;
10365
10366         case offsetof(struct bpf_sk_lookup, remote_ip4):
10367                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10368                                       bpf_target_off(struct bpf_sk_lookup_kern,
10369                                                      v4.saddr, 4, target_size));
10370                 break;
10371
10372         case offsetof(struct bpf_sk_lookup, local_ip4):
10373                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10374                                       bpf_target_off(struct bpf_sk_lookup_kern,
10375                                                      v4.daddr, 4, target_size));
10376                 break;
10377
10378         case bpf_ctx_range_till(struct bpf_sk_lookup,
10379                                 remote_ip6[0], remote_ip6[3]): {
10380 #if IS_ENABLED(CONFIG_IPV6)
10381                 int off = si->off;
10382
10383                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10384                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10385                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10386                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10387                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10388                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10389 #else
10390                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10391 #endif
10392                 break;
10393         }
10394         case bpf_ctx_range_till(struct bpf_sk_lookup,
10395                                 local_ip6[0], local_ip6[3]): {
10396 #if IS_ENABLED(CONFIG_IPV6)
10397                 int off = si->off;
10398
10399                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10400                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10401                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10402                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10403                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10404                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10405 #else
10406                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10407 #endif
10408                 break;
10409         }
10410         case offsetof(struct bpf_sk_lookup, remote_port):
10411                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10412                                       bpf_target_off(struct bpf_sk_lookup_kern,
10413                                                      sport, 2, target_size));
10414                 break;
10415
10416         case offsetof(struct bpf_sk_lookup, local_port):
10417                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10418                                       bpf_target_off(struct bpf_sk_lookup_kern,
10419                                                      dport, 2, target_size));
10420                 break;
10421         }
10422
10423         return insn - insn_buf;
10424 }
10425
10426 const struct bpf_prog_ops sk_lookup_prog_ops = {
10427         .test_run = bpf_prog_test_run_sk_lookup,
10428 };
10429
10430 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10431         .get_func_proto         = sk_lookup_func_proto,
10432         .is_valid_access        = sk_lookup_is_valid_access,
10433         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10434 };
10435
10436 #endif /* CONFIG_INET */
10437
10438 DEFINE_BPF_DISPATCHER(xdp)
10439
10440 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10441 {
10442         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10443 }
10444
10445 #ifdef CONFIG_DEBUG_INFO_BTF
10446 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10447 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10448 BTF_SOCK_TYPE_xxx
10449 #undef BTF_SOCK_TYPE
10450 #else
10451 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10452 #endif
10453
10454 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10455 {
10456         /* tcp6_sock type is not generated in dwarf and hence btf,
10457          * trigger an explicit type generation here.
10458          */
10459         BTF_TYPE_EMIT(struct tcp6_sock);
10460         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10461             sk->sk_family == AF_INET6)
10462                 return (unsigned long)sk;
10463
10464         return (unsigned long)NULL;
10465 }
10466
10467 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10468         .func                   = bpf_skc_to_tcp6_sock,
10469         .gpl_only               = false,
10470         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10471         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10472         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10473 };
10474
10475 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10476 {
10477         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10478                 return (unsigned long)sk;
10479
10480         return (unsigned long)NULL;
10481 }
10482
10483 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10484         .func                   = bpf_skc_to_tcp_sock,
10485         .gpl_only               = false,
10486         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10487         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10488         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10489 };
10490
10491 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10492 {
10493         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10494          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10495          */
10496         BTF_TYPE_EMIT(struct inet_timewait_sock);
10497         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10498
10499 #ifdef CONFIG_INET
10500         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10501                 return (unsigned long)sk;
10502 #endif
10503
10504 #if IS_BUILTIN(CONFIG_IPV6)
10505         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10506                 return (unsigned long)sk;
10507 #endif
10508
10509         return (unsigned long)NULL;
10510 }
10511
10512 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10513         .func                   = bpf_skc_to_tcp_timewait_sock,
10514         .gpl_only               = false,
10515         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10516         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10517         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10518 };
10519
10520 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10521 {
10522 #ifdef CONFIG_INET
10523         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10524                 return (unsigned long)sk;
10525 #endif
10526
10527 #if IS_BUILTIN(CONFIG_IPV6)
10528         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10529                 return (unsigned long)sk;
10530 #endif
10531
10532         return (unsigned long)NULL;
10533 }
10534
10535 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10536         .func                   = bpf_skc_to_tcp_request_sock,
10537         .gpl_only               = false,
10538         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10539         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10540         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10541 };
10542
10543 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10544 {
10545         /* udp6_sock type is not generated in dwarf and hence btf,
10546          * trigger an explicit type generation here.
10547          */
10548         BTF_TYPE_EMIT(struct udp6_sock);
10549         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10550             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10551                 return (unsigned long)sk;
10552
10553         return (unsigned long)NULL;
10554 }
10555
10556 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10557         .func                   = bpf_skc_to_udp6_sock,
10558         .gpl_only               = false,
10559         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10560         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10561         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10562 };
10563
10564 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10565 {
10566         return (unsigned long)sock_from_file(file);
10567 }
10568
10569 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10570 BTF_ID(struct, socket)
10571 BTF_ID(struct, file)
10572
10573 const struct bpf_func_proto bpf_sock_from_file_proto = {
10574         .func           = bpf_sock_from_file,
10575         .gpl_only       = false,
10576         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10577         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10578         .arg1_type      = ARG_PTR_TO_BTF_ID,
10579         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10580 };
10581
10582 static const struct bpf_func_proto *
10583 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10584 {
10585         const struct bpf_func_proto *func;
10586
10587         switch (func_id) {
10588         case BPF_FUNC_skc_to_tcp6_sock:
10589                 func = &bpf_skc_to_tcp6_sock_proto;
10590                 break;
10591         case BPF_FUNC_skc_to_tcp_sock:
10592                 func = &bpf_skc_to_tcp_sock_proto;
10593                 break;
10594         case BPF_FUNC_skc_to_tcp_timewait_sock:
10595                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10596                 break;
10597         case BPF_FUNC_skc_to_tcp_request_sock:
10598                 func = &bpf_skc_to_tcp_request_sock_proto;
10599                 break;
10600         case BPF_FUNC_skc_to_udp6_sock:
10601                 func = &bpf_skc_to_udp6_sock_proto;
10602                 break;
10603         default:
10604                 return bpf_base_func_proto(func_id);
10605         }
10606
10607         if (!perfmon_capable())
10608                 return NULL;
10609
10610         return func;
10611 }