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         int optmem_max = READ_ONCE(sysctl_optmem_max);
1216
1217         /* same check as in sock_kmalloc() */
1218         if (filter_size <= optmem_max &&
1219             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1220                 atomic_add(filter_size, &sk->sk_omem_alloc);
1221                 return true;
1222         }
1223         return false;
1224 }
1225
1226 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1227 {
1228         if (!refcount_inc_not_zero(&fp->refcnt))
1229                 return false;
1230
1231         if (!__sk_filter_charge(sk, fp)) {
1232                 sk_filter_release(fp);
1233                 return false;
1234         }
1235         return true;
1236 }
1237
1238 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1239 {
1240         struct sock_filter *old_prog;
1241         struct bpf_prog *old_fp;
1242         int err, new_len, old_len = fp->len;
1243         bool seen_ld_abs = false;
1244
1245         /* We are free to overwrite insns et al right here as it
1246          * won't be used at this point in time anymore internally
1247          * after the migration to the internal BPF instruction
1248          * representation.
1249          */
1250         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1251                      sizeof(struct bpf_insn));
1252
1253         /* Conversion cannot happen on overlapping memory areas,
1254          * so we need to keep the user BPF around until the 2nd
1255          * pass. At this time, the user BPF is stored in fp->insns.
1256          */
1257         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1258                            GFP_KERNEL | __GFP_NOWARN);
1259         if (!old_prog) {
1260                 err = -ENOMEM;
1261                 goto out_err;
1262         }
1263
1264         /* 1st pass: calculate the new program length. */
1265         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1266                                  &seen_ld_abs);
1267         if (err)
1268                 goto out_err_free;
1269
1270         /* Expand fp for appending the new filter representation. */
1271         old_fp = fp;
1272         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1273         if (!fp) {
1274                 /* The old_fp is still around in case we couldn't
1275                  * allocate new memory, so uncharge on that one.
1276                  */
1277                 fp = old_fp;
1278                 err = -ENOMEM;
1279                 goto out_err_free;
1280         }
1281
1282         fp->len = new_len;
1283
1284         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1285         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1286                                  &seen_ld_abs);
1287         if (err)
1288                 /* 2nd bpf_convert_filter() can fail only if it fails
1289                  * to allocate memory, remapping must succeed. Note,
1290                  * that at this time old_fp has already been released
1291                  * by krealloc().
1292                  */
1293                 goto out_err_free;
1294
1295         fp = bpf_prog_select_runtime(fp, &err);
1296         if (err)
1297                 goto out_err_free;
1298
1299         kfree(old_prog);
1300         return fp;
1301
1302 out_err_free:
1303         kfree(old_prog);
1304 out_err:
1305         __bpf_prog_release(fp);
1306         return ERR_PTR(err);
1307 }
1308
1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1310                                            bpf_aux_classic_check_t trans)
1311 {
1312         int err;
1313
1314         fp->bpf_func = NULL;
1315         fp->jited = 0;
1316
1317         err = bpf_check_classic(fp->insns, fp->len);
1318         if (err) {
1319                 __bpf_prog_release(fp);
1320                 return ERR_PTR(err);
1321         }
1322
1323         /* There might be additional checks and transformations
1324          * needed on classic filters, f.e. in case of seccomp.
1325          */
1326         if (trans) {
1327                 err = trans(fp->insns, fp->len);
1328                 if (err) {
1329                         __bpf_prog_release(fp);
1330                         return ERR_PTR(err);
1331                 }
1332         }
1333
1334         /* Probe if we can JIT compile the filter and if so, do
1335          * the compilation of the filter.
1336          */
1337         bpf_jit_compile(fp);
1338
1339         /* JIT compiler couldn't process this filter, so do the
1340          * internal BPF translation for the optimized interpreter.
1341          */
1342         if (!fp->jited)
1343                 fp = bpf_migrate_filter(fp);
1344
1345         return fp;
1346 }
1347
1348 /**
1349  *      bpf_prog_create - create an unattached filter
1350  *      @pfp: the unattached filter that is created
1351  *      @fprog: the filter program
1352  *
1353  * Create a filter independent of any socket. We first run some
1354  * sanity checks on it to make sure it does not explode on us later.
1355  * If an error occurs or there is insufficient memory for the filter
1356  * a negative errno code is returned. On success the return is zero.
1357  */
1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1359 {
1360         unsigned int fsize = bpf_classic_proglen(fprog);
1361         struct bpf_prog *fp;
1362
1363         /* Make sure new filter is there and in the right amounts. */
1364         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1365                 return -EINVAL;
1366
1367         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1368         if (!fp)
1369                 return -ENOMEM;
1370
1371         memcpy(fp->insns, fprog->filter, fsize);
1372
1373         fp->len = fprog->len;
1374         /* Since unattached filters are not copied back to user
1375          * space through sk_get_filter(), we do not need to hold
1376          * a copy here, and can spare us the work.
1377          */
1378         fp->orig_prog = NULL;
1379
1380         /* bpf_prepare_filter() already takes care of freeing
1381          * memory in case something goes wrong.
1382          */
1383         fp = bpf_prepare_filter(fp, NULL);
1384         if (IS_ERR(fp))
1385                 return PTR_ERR(fp);
1386
1387         *pfp = fp;
1388         return 0;
1389 }
1390 EXPORT_SYMBOL_GPL(bpf_prog_create);
1391
1392 /**
1393  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1394  *      @pfp: the unattached filter that is created
1395  *      @fprog: the filter program
1396  *      @trans: post-classic verifier transformation handler
1397  *      @save_orig: save classic BPF program
1398  *
1399  * This function effectively does the same as bpf_prog_create(), only
1400  * that it builds up its insns buffer from user space provided buffer.
1401  * It also allows for passing a bpf_aux_classic_check_t handler.
1402  */
1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1404                               bpf_aux_classic_check_t trans, bool save_orig)
1405 {
1406         unsigned int fsize = bpf_classic_proglen(fprog);
1407         struct bpf_prog *fp;
1408         int err;
1409
1410         /* Make sure new filter is there and in the right amounts. */
1411         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1412                 return -EINVAL;
1413
1414         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1415         if (!fp)
1416                 return -ENOMEM;
1417
1418         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1419                 __bpf_prog_free(fp);
1420                 return -EFAULT;
1421         }
1422
1423         fp->len = fprog->len;
1424         fp->orig_prog = NULL;
1425
1426         if (save_orig) {
1427                 err = bpf_prog_store_orig_filter(fp, fprog);
1428                 if (err) {
1429                         __bpf_prog_free(fp);
1430                         return -ENOMEM;
1431                 }
1432         }
1433
1434         /* bpf_prepare_filter() already takes care of freeing
1435          * memory in case something goes wrong.
1436          */
1437         fp = bpf_prepare_filter(fp, trans);
1438         if (IS_ERR(fp))
1439                 return PTR_ERR(fp);
1440
1441         *pfp = fp;
1442         return 0;
1443 }
1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1445
1446 void bpf_prog_destroy(struct bpf_prog *fp)
1447 {
1448         __bpf_prog_release(fp);
1449 }
1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1451
1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1453 {
1454         struct sk_filter *fp, *old_fp;
1455
1456         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1457         if (!fp)
1458                 return -ENOMEM;
1459
1460         fp->prog = prog;
1461
1462         if (!__sk_filter_charge(sk, fp)) {
1463                 kfree(fp);
1464                 return -ENOMEM;
1465         }
1466         refcount_set(&fp->refcnt, 1);
1467
1468         old_fp = rcu_dereference_protected(sk->sk_filter,
1469                                            lockdep_sock_is_held(sk));
1470         rcu_assign_pointer(sk->sk_filter, fp);
1471
1472         if (old_fp)
1473                 sk_filter_uncharge(sk, old_fp);
1474
1475         return 0;
1476 }
1477
1478 static
1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1480 {
1481         unsigned int fsize = bpf_classic_proglen(fprog);
1482         struct bpf_prog *prog;
1483         int err;
1484
1485         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1486                 return ERR_PTR(-EPERM);
1487
1488         /* Make sure new filter is there and in the right amounts. */
1489         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1490                 return ERR_PTR(-EINVAL);
1491
1492         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1493         if (!prog)
1494                 return ERR_PTR(-ENOMEM);
1495
1496         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1497                 __bpf_prog_free(prog);
1498                 return ERR_PTR(-EFAULT);
1499         }
1500
1501         prog->len = fprog->len;
1502
1503         err = bpf_prog_store_orig_filter(prog, fprog);
1504         if (err) {
1505                 __bpf_prog_free(prog);
1506                 return ERR_PTR(-ENOMEM);
1507         }
1508
1509         /* bpf_prepare_filter() already takes care of freeing
1510          * memory in case something goes wrong.
1511          */
1512         return bpf_prepare_filter(prog, NULL);
1513 }
1514
1515 /**
1516  *      sk_attach_filter - attach a socket filter
1517  *      @fprog: the filter program
1518  *      @sk: the socket to use
1519  *
1520  * Attach the user's filter code. We first run some sanity checks on
1521  * it to make sure it does not explode on us later. If an error
1522  * occurs or there is insufficient memory for the filter a negative
1523  * errno code is returned. On success the return is zero.
1524  */
1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1526 {
1527         struct bpf_prog *prog = __get_filter(fprog, sk);
1528         int err;
1529
1530         if (IS_ERR(prog))
1531                 return PTR_ERR(prog);
1532
1533         err = __sk_attach_prog(prog, sk);
1534         if (err < 0) {
1535                 __bpf_prog_release(prog);
1536                 return err;
1537         }
1538
1539         return 0;
1540 }
1541 EXPORT_SYMBOL_GPL(sk_attach_filter);
1542
1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1544 {
1545         struct bpf_prog *prog = __get_filter(fprog, sk);
1546         int err;
1547
1548         if (IS_ERR(prog))
1549                 return PTR_ERR(prog);
1550
1551         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1552                 err = -ENOMEM;
1553         else
1554                 err = reuseport_attach_prog(sk, prog);
1555
1556         if (err)
1557                 __bpf_prog_release(prog);
1558
1559         return err;
1560 }
1561
1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1563 {
1564         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1565                 return ERR_PTR(-EPERM);
1566
1567         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1568 }
1569
1570 int sk_attach_bpf(u32 ufd, struct sock *sk)
1571 {
1572         struct bpf_prog *prog = __get_bpf(ufd, sk);
1573         int err;
1574
1575         if (IS_ERR(prog))
1576                 return PTR_ERR(prog);
1577
1578         err = __sk_attach_prog(prog, sk);
1579         if (err < 0) {
1580                 bpf_prog_put(prog);
1581                 return err;
1582         }
1583
1584         return 0;
1585 }
1586
1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1588 {
1589         struct bpf_prog *prog;
1590         int err;
1591
1592         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1593                 return -EPERM;
1594
1595         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1596         if (PTR_ERR(prog) == -EINVAL)
1597                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1598         if (IS_ERR(prog))
1599                 return PTR_ERR(prog);
1600
1601         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1602                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1603                  * bpf prog (e.g. sockmap).  It depends on the
1604                  * limitation imposed by bpf_prog_load().
1605                  * Hence, sysctl_optmem_max is not checked.
1606                  */
1607                 if ((sk->sk_type != SOCK_STREAM &&
1608                      sk->sk_type != SOCK_DGRAM) ||
1609                     (sk->sk_protocol != IPPROTO_UDP &&
1610                      sk->sk_protocol != IPPROTO_TCP) ||
1611                     (sk->sk_family != AF_INET &&
1612                      sk->sk_family != AF_INET6)) {
1613                         err = -ENOTSUPP;
1614                         goto err_prog_put;
1615                 }
1616         } else {
1617                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1618                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1619                         err = -ENOMEM;
1620                         goto err_prog_put;
1621                 }
1622         }
1623
1624         err = reuseport_attach_prog(sk, prog);
1625 err_prog_put:
1626         if (err)
1627                 bpf_prog_put(prog);
1628
1629         return err;
1630 }
1631
1632 void sk_reuseport_prog_free(struct bpf_prog *prog)
1633 {
1634         if (!prog)
1635                 return;
1636
1637         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1638                 bpf_prog_put(prog);
1639         else
1640                 bpf_prog_destroy(prog);
1641 }
1642
1643 struct bpf_scratchpad {
1644         union {
1645                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1646                 u8     buff[MAX_BPF_STACK];
1647         };
1648 };
1649
1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1651
1652 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1653                                           unsigned int write_len)
1654 {
1655         return skb_ensure_writable(skb, write_len);
1656 }
1657
1658 static inline int bpf_try_make_writable(struct sk_buff *skb,
1659                                         unsigned int write_len)
1660 {
1661         int err = __bpf_try_make_writable(skb, write_len);
1662
1663         bpf_compute_data_pointers(skb);
1664         return err;
1665 }
1666
1667 static int bpf_try_make_head_writable(struct sk_buff *skb)
1668 {
1669         return bpf_try_make_writable(skb, skb_headlen(skb));
1670 }
1671
1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1673 {
1674         if (skb_at_tc_ingress(skb))
1675                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1676 }
1677
1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1679 {
1680         if (skb_at_tc_ingress(skb))
1681                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1682 }
1683
1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1685            const void *, from, u32, len, u64, flags)
1686 {
1687         void *ptr;
1688
1689         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1690                 return -EINVAL;
1691         if (unlikely(offset > INT_MAX))
1692                 return -EFAULT;
1693         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1694                 return -EFAULT;
1695
1696         ptr = skb->data + offset;
1697         if (flags & BPF_F_RECOMPUTE_CSUM)
1698                 __skb_postpull_rcsum(skb, ptr, len, offset);
1699
1700         memcpy(ptr, from, len);
1701
1702         if (flags & BPF_F_RECOMPUTE_CSUM)
1703                 __skb_postpush_rcsum(skb, ptr, len, offset);
1704         if (flags & BPF_F_INVALIDATE_HASH)
1705                 skb_clear_hash(skb);
1706
1707         return 0;
1708 }
1709
1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1711         .func           = bpf_skb_store_bytes,
1712         .gpl_only       = false,
1713         .ret_type       = RET_INTEGER,
1714         .arg1_type      = ARG_PTR_TO_CTX,
1715         .arg2_type      = ARG_ANYTHING,
1716         .arg3_type      = ARG_PTR_TO_MEM,
1717         .arg4_type      = ARG_CONST_SIZE,
1718         .arg5_type      = ARG_ANYTHING,
1719 };
1720
1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1722            void *, to, u32, len)
1723 {
1724         void *ptr;
1725
1726         if (unlikely(offset > INT_MAX))
1727                 goto err_clear;
1728
1729         ptr = skb_header_pointer(skb, offset, len, to);
1730         if (unlikely(!ptr))
1731                 goto err_clear;
1732         if (ptr != to)
1733                 memcpy(to, ptr, len);
1734
1735         return 0;
1736 err_clear:
1737         memset(to, 0, len);
1738         return -EFAULT;
1739 }
1740
1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1742         .func           = bpf_skb_load_bytes,
1743         .gpl_only       = false,
1744         .ret_type       = RET_INTEGER,
1745         .arg1_type      = ARG_PTR_TO_CTX,
1746         .arg2_type      = ARG_ANYTHING,
1747         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1748         .arg4_type      = ARG_CONST_SIZE,
1749 };
1750
1751 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1752            const struct bpf_flow_dissector *, ctx, u32, offset,
1753            void *, to, u32, len)
1754 {
1755         void *ptr;
1756
1757         if (unlikely(offset > 0xffff))
1758                 goto err_clear;
1759
1760         if (unlikely(!ctx->skb))
1761                 goto err_clear;
1762
1763         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1764         if (unlikely(!ptr))
1765                 goto err_clear;
1766         if (ptr != to)
1767                 memcpy(to, ptr, len);
1768
1769         return 0;
1770 err_clear:
1771         memset(to, 0, len);
1772         return -EFAULT;
1773 }
1774
1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1776         .func           = bpf_flow_dissector_load_bytes,
1777         .gpl_only       = false,
1778         .ret_type       = RET_INTEGER,
1779         .arg1_type      = ARG_PTR_TO_CTX,
1780         .arg2_type      = ARG_ANYTHING,
1781         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1782         .arg4_type      = ARG_CONST_SIZE,
1783 };
1784
1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1786            u32, offset, void *, to, u32, len, u32, start_header)
1787 {
1788         u8 *end = skb_tail_pointer(skb);
1789         u8 *start, *ptr;
1790
1791         if (unlikely(offset > 0xffff))
1792                 goto err_clear;
1793
1794         switch (start_header) {
1795         case BPF_HDR_START_MAC:
1796                 if (unlikely(!skb_mac_header_was_set(skb)))
1797                         goto err_clear;
1798                 start = skb_mac_header(skb);
1799                 break;
1800         case BPF_HDR_START_NET:
1801                 start = skb_network_header(skb);
1802                 break;
1803         default:
1804                 goto err_clear;
1805         }
1806
1807         ptr = start + offset;
1808
1809         if (likely(ptr + len <= end)) {
1810                 memcpy(to, ptr, len);
1811                 return 0;
1812         }
1813
1814 err_clear:
1815         memset(to, 0, len);
1816         return -EFAULT;
1817 }
1818
1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1820         .func           = bpf_skb_load_bytes_relative,
1821         .gpl_only       = false,
1822         .ret_type       = RET_INTEGER,
1823         .arg1_type      = ARG_PTR_TO_CTX,
1824         .arg2_type      = ARG_ANYTHING,
1825         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1826         .arg4_type      = ARG_CONST_SIZE,
1827         .arg5_type      = ARG_ANYTHING,
1828 };
1829
1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1831 {
1832         /* Idea is the following: should the needed direct read/write
1833          * test fail during runtime, we can pull in more data and redo
1834          * again, since implicitly, we invalidate previous checks here.
1835          *
1836          * Or, since we know how much we need to make read/writeable,
1837          * this can be done once at the program beginning for direct
1838          * access case. By this we overcome limitations of only current
1839          * headroom being accessible.
1840          */
1841         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1842 }
1843
1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1845         .func           = bpf_skb_pull_data,
1846         .gpl_only       = false,
1847         .ret_type       = RET_INTEGER,
1848         .arg1_type      = ARG_PTR_TO_CTX,
1849         .arg2_type      = ARG_ANYTHING,
1850 };
1851
1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1853 {
1854         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1855 }
1856
1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1858         .func           = bpf_sk_fullsock,
1859         .gpl_only       = false,
1860         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1861         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1862 };
1863
1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1865                                            unsigned int write_len)
1866 {
1867         int err = __bpf_try_make_writable(skb, write_len);
1868
1869         bpf_compute_data_end_sk_skb(skb);
1870         return err;
1871 }
1872
1873 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1874 {
1875         /* Idea is the following: should the needed direct read/write
1876          * test fail during runtime, we can pull in more data and redo
1877          * again, since implicitly, we invalidate previous checks here.
1878          *
1879          * Or, since we know how much we need to make read/writeable,
1880          * this can be done once at the program beginning for direct
1881          * access case. By this we overcome limitations of only current
1882          * headroom being accessible.
1883          */
1884         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1885 }
1886
1887 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1888         .func           = sk_skb_pull_data,
1889         .gpl_only       = false,
1890         .ret_type       = RET_INTEGER,
1891         .arg1_type      = ARG_PTR_TO_CTX,
1892         .arg2_type      = ARG_ANYTHING,
1893 };
1894
1895 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1896            u64, from, u64, to, u64, flags)
1897 {
1898         __sum16 *ptr;
1899
1900         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1901                 return -EINVAL;
1902         if (unlikely(offset > 0xffff || offset & 1))
1903                 return -EFAULT;
1904         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1905                 return -EFAULT;
1906
1907         ptr = (__sum16 *)(skb->data + offset);
1908         switch (flags & BPF_F_HDR_FIELD_MASK) {
1909         case 0:
1910                 if (unlikely(from != 0))
1911                         return -EINVAL;
1912
1913                 csum_replace_by_diff(ptr, to);
1914                 break;
1915         case 2:
1916                 csum_replace2(ptr, from, to);
1917                 break;
1918         case 4:
1919                 csum_replace4(ptr, from, to);
1920                 break;
1921         default:
1922                 return -EINVAL;
1923         }
1924
1925         return 0;
1926 }
1927
1928 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1929         .func           = bpf_l3_csum_replace,
1930         .gpl_only       = false,
1931         .ret_type       = RET_INTEGER,
1932         .arg1_type      = ARG_PTR_TO_CTX,
1933         .arg2_type      = ARG_ANYTHING,
1934         .arg3_type      = ARG_ANYTHING,
1935         .arg4_type      = ARG_ANYTHING,
1936         .arg5_type      = ARG_ANYTHING,
1937 };
1938
1939 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1940            u64, from, u64, to, u64, flags)
1941 {
1942         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1943         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1944         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1945         __sum16 *ptr;
1946
1947         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1948                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1949                 return -EINVAL;
1950         if (unlikely(offset > 0xffff || offset & 1))
1951                 return -EFAULT;
1952         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1953                 return -EFAULT;
1954
1955         ptr = (__sum16 *)(skb->data + offset);
1956         if (is_mmzero && !do_mforce && !*ptr)
1957                 return 0;
1958
1959         switch (flags & BPF_F_HDR_FIELD_MASK) {
1960         case 0:
1961                 if (unlikely(from != 0))
1962                         return -EINVAL;
1963
1964                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1965                 break;
1966         case 2:
1967                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1968                 break;
1969         case 4:
1970                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1971                 break;
1972         default:
1973                 return -EINVAL;
1974         }
1975
1976         if (is_mmzero && !*ptr)
1977                 *ptr = CSUM_MANGLED_0;
1978         return 0;
1979 }
1980
1981 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1982         .func           = bpf_l4_csum_replace,
1983         .gpl_only       = false,
1984         .ret_type       = RET_INTEGER,
1985         .arg1_type      = ARG_PTR_TO_CTX,
1986         .arg2_type      = ARG_ANYTHING,
1987         .arg3_type      = ARG_ANYTHING,
1988         .arg4_type      = ARG_ANYTHING,
1989         .arg5_type      = ARG_ANYTHING,
1990 };
1991
1992 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1993            __be32 *, to, u32, to_size, __wsum, seed)
1994 {
1995         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1996         u32 diff_size = from_size + to_size;
1997         int i, j = 0;
1998
1999         /* This is quite flexible, some examples:
2000          *
2001          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2002          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2003          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2004          *
2005          * Even for diffing, from_size and to_size don't need to be equal.
2006          */
2007         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2008                      diff_size > sizeof(sp->diff)))
2009                 return -EINVAL;
2010
2011         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2012                 sp->diff[j] = ~from[i];
2013         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2014                 sp->diff[j] = to[i];
2015
2016         return csum_partial(sp->diff, diff_size, seed);
2017 }
2018
2019 static const struct bpf_func_proto bpf_csum_diff_proto = {
2020         .func           = bpf_csum_diff,
2021         .gpl_only       = false,
2022         .pkt_access     = true,
2023         .ret_type       = RET_INTEGER,
2024         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
2025         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2026         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
2027         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2028         .arg5_type      = ARG_ANYTHING,
2029 };
2030
2031 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2032 {
2033         /* The interface is to be used in combination with bpf_csum_diff()
2034          * for direct packet writes. csum rotation for alignment as well
2035          * as emulating csum_sub() can be done from the eBPF program.
2036          */
2037         if (skb->ip_summed == CHECKSUM_COMPLETE)
2038                 return (skb->csum = csum_add(skb->csum, csum));
2039
2040         return -ENOTSUPP;
2041 }
2042
2043 static const struct bpf_func_proto bpf_csum_update_proto = {
2044         .func           = bpf_csum_update,
2045         .gpl_only       = false,
2046         .ret_type       = RET_INTEGER,
2047         .arg1_type      = ARG_PTR_TO_CTX,
2048         .arg2_type      = ARG_ANYTHING,
2049 };
2050
2051 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2052 {
2053         /* The interface is to be used in combination with bpf_skb_adjust_room()
2054          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2055          * is passed as flags, for example.
2056          */
2057         switch (level) {
2058         case BPF_CSUM_LEVEL_INC:
2059                 __skb_incr_checksum_unnecessary(skb);
2060                 break;
2061         case BPF_CSUM_LEVEL_DEC:
2062                 __skb_decr_checksum_unnecessary(skb);
2063                 break;
2064         case BPF_CSUM_LEVEL_RESET:
2065                 __skb_reset_checksum_unnecessary(skb);
2066                 break;
2067         case BPF_CSUM_LEVEL_QUERY:
2068                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2069                        skb->csum_level : -EACCES;
2070         default:
2071                 return -EINVAL;
2072         }
2073
2074         return 0;
2075 }
2076
2077 static const struct bpf_func_proto bpf_csum_level_proto = {
2078         .func           = bpf_csum_level,
2079         .gpl_only       = false,
2080         .ret_type       = RET_INTEGER,
2081         .arg1_type      = ARG_PTR_TO_CTX,
2082         .arg2_type      = ARG_ANYTHING,
2083 };
2084
2085 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2086 {
2087         return dev_forward_skb(dev, skb);
2088 }
2089
2090 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2091                                       struct sk_buff *skb)
2092 {
2093         int ret = ____dev_forward_skb(dev, skb);
2094
2095         if (likely(!ret)) {
2096                 skb->dev = dev;
2097                 ret = netif_rx(skb);
2098         }
2099
2100         return ret;
2101 }
2102
2103 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2104 {
2105         int ret;
2106
2107         if (dev_xmit_recursion()) {
2108                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2109                 kfree_skb(skb);
2110                 return -ENETDOWN;
2111         }
2112
2113         skb->dev = dev;
2114         skb->tstamp = 0;
2115
2116         dev_xmit_recursion_inc();
2117         ret = dev_queue_xmit(skb);
2118         dev_xmit_recursion_dec();
2119
2120         return ret;
2121 }
2122
2123 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2124                                  u32 flags)
2125 {
2126         unsigned int mlen = skb_network_offset(skb);
2127
2128         if (mlen) {
2129                 __skb_pull(skb, mlen);
2130
2131                 /* At ingress, the mac header has already been pulled once.
2132                  * At egress, skb_pospull_rcsum has to be done in case that
2133                  * the skb is originated from ingress (i.e. a forwarded skb)
2134                  * to ensure that rcsum starts at net header.
2135                  */
2136                 if (!skb_at_tc_ingress(skb))
2137                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2138         }
2139         skb_pop_mac_header(skb);
2140         skb_reset_mac_len(skb);
2141         return flags & BPF_F_INGRESS ?
2142                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2143 }
2144
2145 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2146                                  u32 flags)
2147 {
2148         /* Verify that a link layer header is carried */
2149         if (unlikely(skb->mac_header >= skb->network_header)) {
2150                 kfree_skb(skb);
2151                 return -ERANGE;
2152         }
2153
2154         bpf_push_mac_rcsum(skb);
2155         return flags & BPF_F_INGRESS ?
2156                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2157 }
2158
2159 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2160                           u32 flags)
2161 {
2162         if (dev_is_mac_header_xmit(dev))
2163                 return __bpf_redirect_common(skb, dev, flags);
2164         else
2165                 return __bpf_redirect_no_mac(skb, dev, flags);
2166 }
2167
2168 #if IS_ENABLED(CONFIG_IPV6)
2169 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2170                             struct net_device *dev, struct bpf_nh_params *nh)
2171 {
2172         u32 hh_len = LL_RESERVED_SPACE(dev);
2173         const struct in6_addr *nexthop;
2174         struct dst_entry *dst = NULL;
2175         struct neighbour *neigh;
2176
2177         if (dev_xmit_recursion()) {
2178                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2179                 goto out_drop;
2180         }
2181
2182         skb->dev = dev;
2183         skb->tstamp = 0;
2184
2185         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2186                 struct sk_buff *skb2;
2187
2188                 skb2 = skb_realloc_headroom(skb, hh_len);
2189                 if (unlikely(!skb2)) {
2190                         kfree_skb(skb);
2191                         return -ENOMEM;
2192                 }
2193                 if (skb->sk)
2194                         skb_set_owner_w(skb2, skb->sk);
2195                 consume_skb(skb);
2196                 skb = skb2;
2197         }
2198
2199         rcu_read_lock_bh();
2200         if (!nh) {
2201                 dst = skb_dst(skb);
2202                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2203                                       &ipv6_hdr(skb)->daddr);
2204         } else {
2205                 nexthop = &nh->ipv6_nh;
2206         }
2207         neigh = ip_neigh_gw6(dev, nexthop);
2208         if (likely(!IS_ERR(neigh))) {
2209                 int ret;
2210
2211                 sock_confirm_neigh(skb, neigh);
2212                 dev_xmit_recursion_inc();
2213                 ret = neigh_output(neigh, skb, false);
2214                 dev_xmit_recursion_dec();
2215                 rcu_read_unlock_bh();
2216                 return ret;
2217         }
2218         rcu_read_unlock_bh();
2219         if (dst)
2220                 IP6_INC_STATS(dev_net(dst->dev),
2221                               ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2222 out_drop:
2223         kfree_skb(skb);
2224         return -ENETDOWN;
2225 }
2226
2227 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2228                                    struct bpf_nh_params *nh)
2229 {
2230         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2231         struct net *net = dev_net(dev);
2232         int err, ret = NET_XMIT_DROP;
2233
2234         if (!nh) {
2235                 struct dst_entry *dst;
2236                 struct flowi6 fl6 = {
2237                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2238                         .flowi6_mark  = skb->mark,
2239                         .flowlabel    = ip6_flowinfo(ip6h),
2240                         .flowi6_oif   = dev->ifindex,
2241                         .flowi6_proto = ip6h->nexthdr,
2242                         .daddr        = ip6h->daddr,
2243                         .saddr        = ip6h->saddr,
2244                 };
2245
2246                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2247                 if (IS_ERR(dst))
2248                         goto out_drop;
2249
2250                 skb_dst_set(skb, dst);
2251         } else if (nh->nh_family != AF_INET6) {
2252                 goto out_drop;
2253         }
2254
2255         err = bpf_out_neigh_v6(net, skb, dev, nh);
2256         if (unlikely(net_xmit_eval(err)))
2257                 dev->stats.tx_errors++;
2258         else
2259                 ret = NET_XMIT_SUCCESS;
2260         goto out_xmit;
2261 out_drop:
2262         dev->stats.tx_errors++;
2263         kfree_skb(skb);
2264 out_xmit:
2265         return ret;
2266 }
2267 #else
2268 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2269                                    struct bpf_nh_params *nh)
2270 {
2271         kfree_skb(skb);
2272         return NET_XMIT_DROP;
2273 }
2274 #endif /* CONFIG_IPV6 */
2275
2276 #if IS_ENABLED(CONFIG_INET)
2277 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2278                             struct net_device *dev, struct bpf_nh_params *nh)
2279 {
2280         u32 hh_len = LL_RESERVED_SPACE(dev);
2281         struct neighbour *neigh;
2282         bool is_v6gw = false;
2283
2284         if (dev_xmit_recursion()) {
2285                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2286                 goto out_drop;
2287         }
2288
2289         skb->dev = dev;
2290         skb->tstamp = 0;
2291
2292         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2293                 struct sk_buff *skb2;
2294
2295                 skb2 = skb_realloc_headroom(skb, hh_len);
2296                 if (unlikely(!skb2)) {
2297                         kfree_skb(skb);
2298                         return -ENOMEM;
2299                 }
2300                 if (skb->sk)
2301                         skb_set_owner_w(skb2, skb->sk);
2302                 consume_skb(skb);
2303                 skb = skb2;
2304         }
2305
2306         rcu_read_lock_bh();
2307         if (!nh) {
2308                 struct dst_entry *dst = skb_dst(skb);
2309                 struct rtable *rt = container_of(dst, struct rtable, dst);
2310
2311                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2312         } else if (nh->nh_family == AF_INET6) {
2313                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2314                 is_v6gw = true;
2315         } else if (nh->nh_family == AF_INET) {
2316                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2317         } else {
2318                 rcu_read_unlock_bh();
2319                 goto out_drop;
2320         }
2321
2322         if (likely(!IS_ERR(neigh))) {
2323                 int ret;
2324
2325                 sock_confirm_neigh(skb, neigh);
2326                 dev_xmit_recursion_inc();
2327                 ret = neigh_output(neigh, skb, is_v6gw);
2328                 dev_xmit_recursion_dec();
2329                 rcu_read_unlock_bh();
2330                 return ret;
2331         }
2332         rcu_read_unlock_bh();
2333 out_drop:
2334         kfree_skb(skb);
2335         return -ENETDOWN;
2336 }
2337
2338 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2339                                    struct bpf_nh_params *nh)
2340 {
2341         const struct iphdr *ip4h = ip_hdr(skb);
2342         struct net *net = dev_net(dev);
2343         int err, ret = NET_XMIT_DROP;
2344
2345         if (!nh) {
2346                 struct flowi4 fl4 = {
2347                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2348                         .flowi4_mark  = skb->mark,
2349                         .flowi4_tos   = RT_TOS(ip4h->tos),
2350                         .flowi4_oif   = dev->ifindex,
2351                         .flowi4_proto = ip4h->protocol,
2352                         .daddr        = ip4h->daddr,
2353                         .saddr        = ip4h->saddr,
2354                 };
2355                 struct rtable *rt;
2356
2357                 rt = ip_route_output_flow(net, &fl4, NULL);
2358                 if (IS_ERR(rt))
2359                         goto out_drop;
2360                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2361                         ip_rt_put(rt);
2362                         goto out_drop;
2363                 }
2364
2365                 skb_dst_set(skb, &rt->dst);
2366         }
2367
2368         err = bpf_out_neigh_v4(net, skb, dev, nh);
2369         if (unlikely(net_xmit_eval(err)))
2370                 dev->stats.tx_errors++;
2371         else
2372                 ret = NET_XMIT_SUCCESS;
2373         goto out_xmit;
2374 out_drop:
2375         dev->stats.tx_errors++;
2376         kfree_skb(skb);
2377 out_xmit:
2378         return ret;
2379 }
2380 #else
2381 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2382                                    struct bpf_nh_params *nh)
2383 {
2384         kfree_skb(skb);
2385         return NET_XMIT_DROP;
2386 }
2387 #endif /* CONFIG_INET */
2388
2389 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2390                                 struct bpf_nh_params *nh)
2391 {
2392         struct ethhdr *ethh = eth_hdr(skb);
2393
2394         if (unlikely(skb->mac_header >= skb->network_header))
2395                 goto out;
2396         bpf_push_mac_rcsum(skb);
2397         if (is_multicast_ether_addr(ethh->h_dest))
2398                 goto out;
2399
2400         skb_pull(skb, sizeof(*ethh));
2401         skb_unset_mac_header(skb);
2402         skb_reset_network_header(skb);
2403
2404         if (skb->protocol == htons(ETH_P_IP))
2405                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2406         else if (skb->protocol == htons(ETH_P_IPV6))
2407                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2408 out:
2409         kfree_skb(skb);
2410         return -ENOTSUPP;
2411 }
2412
2413 /* Internal, non-exposed redirect flags. */
2414 enum {
2415         BPF_F_NEIGH     = (1ULL << 1),
2416         BPF_F_PEER      = (1ULL << 2),
2417         BPF_F_NEXTHOP   = (1ULL << 3),
2418 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2419 };
2420
2421 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2422 {
2423         struct net_device *dev;
2424         struct sk_buff *clone;
2425         int ret;
2426
2427         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2428                 return -EINVAL;
2429
2430         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2431         if (unlikely(!dev))
2432                 return -EINVAL;
2433
2434         clone = skb_clone(skb, GFP_ATOMIC);
2435         if (unlikely(!clone))
2436                 return -ENOMEM;
2437
2438         /* For direct write, we need to keep the invariant that the skbs
2439          * we're dealing with need to be uncloned. Should uncloning fail
2440          * here, we need to free the just generated clone to unclone once
2441          * again.
2442          */
2443         ret = bpf_try_make_head_writable(skb);
2444         if (unlikely(ret)) {
2445                 kfree_skb(clone);
2446                 return -ENOMEM;
2447         }
2448
2449         return __bpf_redirect(clone, dev, flags);
2450 }
2451
2452 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2453         .func           = bpf_clone_redirect,
2454         .gpl_only       = false,
2455         .ret_type       = RET_INTEGER,
2456         .arg1_type      = ARG_PTR_TO_CTX,
2457         .arg2_type      = ARG_ANYTHING,
2458         .arg3_type      = ARG_ANYTHING,
2459 };
2460
2461 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2462 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2463
2464 int skb_do_redirect(struct sk_buff *skb)
2465 {
2466         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2467         struct net *net = dev_net(skb->dev);
2468         struct net_device *dev;
2469         u32 flags = ri->flags;
2470
2471         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2472         ri->tgt_index = 0;
2473         ri->flags = 0;
2474         if (unlikely(!dev))
2475                 goto out_drop;
2476         if (flags & BPF_F_PEER) {
2477                 const struct net_device_ops *ops = dev->netdev_ops;
2478
2479                 if (unlikely(!ops->ndo_get_peer_dev ||
2480                              !skb_at_tc_ingress(skb)))
2481                         goto out_drop;
2482                 dev = ops->ndo_get_peer_dev(dev);
2483                 if (unlikely(!dev ||
2484                              !is_skb_forwardable(dev, skb) ||
2485                              net_eq(net, dev_net(dev))))
2486                         goto out_drop;
2487                 skb->dev = dev;
2488                 return -EAGAIN;
2489         }
2490         return flags & BPF_F_NEIGH ?
2491                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2492                                     &ri->nh : NULL) :
2493                __bpf_redirect(skb, dev, flags);
2494 out_drop:
2495         kfree_skb(skb);
2496         return -EINVAL;
2497 }
2498
2499 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2500 {
2501         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2502
2503         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2504                 return TC_ACT_SHOT;
2505
2506         ri->flags = flags;
2507         ri->tgt_index = ifindex;
2508
2509         return TC_ACT_REDIRECT;
2510 }
2511
2512 static const struct bpf_func_proto bpf_redirect_proto = {
2513         .func           = bpf_redirect,
2514         .gpl_only       = false,
2515         .ret_type       = RET_INTEGER,
2516         .arg1_type      = ARG_ANYTHING,
2517         .arg2_type      = ARG_ANYTHING,
2518 };
2519
2520 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2521 {
2522         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2523
2524         if (unlikely(flags))
2525                 return TC_ACT_SHOT;
2526
2527         ri->flags = BPF_F_PEER;
2528         ri->tgt_index = ifindex;
2529
2530         return TC_ACT_REDIRECT;
2531 }
2532
2533 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2534         .func           = bpf_redirect_peer,
2535         .gpl_only       = false,
2536         .ret_type       = RET_INTEGER,
2537         .arg1_type      = ARG_ANYTHING,
2538         .arg2_type      = ARG_ANYTHING,
2539 };
2540
2541 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2542            int, plen, u64, flags)
2543 {
2544         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2545
2546         if (unlikely((plen && plen < sizeof(*params)) || flags))
2547                 return TC_ACT_SHOT;
2548
2549         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2550         ri->tgt_index = ifindex;
2551
2552         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2553         if (plen)
2554                 memcpy(&ri->nh, params, sizeof(ri->nh));
2555
2556         return TC_ACT_REDIRECT;
2557 }
2558
2559 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2560         .func           = bpf_redirect_neigh,
2561         .gpl_only       = false,
2562         .ret_type       = RET_INTEGER,
2563         .arg1_type      = ARG_ANYTHING,
2564         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
2565         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2566         .arg4_type      = ARG_ANYTHING,
2567 };
2568
2569 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2570 {
2571         msg->apply_bytes = bytes;
2572         return 0;
2573 }
2574
2575 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2576         .func           = bpf_msg_apply_bytes,
2577         .gpl_only       = false,
2578         .ret_type       = RET_INTEGER,
2579         .arg1_type      = ARG_PTR_TO_CTX,
2580         .arg2_type      = ARG_ANYTHING,
2581 };
2582
2583 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2584 {
2585         msg->cork_bytes = bytes;
2586         return 0;
2587 }
2588
2589 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2590         .func           = bpf_msg_cork_bytes,
2591         .gpl_only       = false,
2592         .ret_type       = RET_INTEGER,
2593         .arg1_type      = ARG_PTR_TO_CTX,
2594         .arg2_type      = ARG_ANYTHING,
2595 };
2596
2597 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2598            u32, end, u64, flags)
2599 {
2600         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2601         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2602         struct scatterlist *sge;
2603         u8 *raw, *to, *from;
2604         struct page *page;
2605
2606         if (unlikely(flags || end <= start))
2607                 return -EINVAL;
2608
2609         /* First find the starting scatterlist element */
2610         i = msg->sg.start;
2611         do {
2612                 offset += len;
2613                 len = sk_msg_elem(msg, i)->length;
2614                 if (start < offset + len)
2615                         break;
2616                 sk_msg_iter_var_next(i);
2617         } while (i != msg->sg.end);
2618
2619         if (unlikely(start >= offset + len))
2620                 return -EINVAL;
2621
2622         first_sge = i;
2623         /* The start may point into the sg element so we need to also
2624          * account for the headroom.
2625          */
2626         bytes_sg_total = start - offset + bytes;
2627         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2628                 goto out;
2629
2630         /* At this point we need to linearize multiple scatterlist
2631          * elements or a single shared page. Either way we need to
2632          * copy into a linear buffer exclusively owned by BPF. Then
2633          * place the buffer in the scatterlist and fixup the original
2634          * entries by removing the entries now in the linear buffer
2635          * and shifting the remaining entries. For now we do not try
2636          * to copy partial entries to avoid complexity of running out
2637          * of sg_entry slots. The downside is reading a single byte
2638          * will copy the entire sg entry.
2639          */
2640         do {
2641                 copy += sk_msg_elem(msg, i)->length;
2642                 sk_msg_iter_var_next(i);
2643                 if (bytes_sg_total <= copy)
2644                         break;
2645         } while (i != msg->sg.end);
2646         last_sge = i;
2647
2648         if (unlikely(bytes_sg_total > copy))
2649                 return -EINVAL;
2650
2651         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2652                            get_order(copy));
2653         if (unlikely(!page))
2654                 return -ENOMEM;
2655
2656         raw = page_address(page);
2657         i = first_sge;
2658         do {
2659                 sge = sk_msg_elem(msg, i);
2660                 from = sg_virt(sge);
2661                 len = sge->length;
2662                 to = raw + poffset;
2663
2664                 memcpy(to, from, len);
2665                 poffset += len;
2666                 sge->length = 0;
2667                 put_page(sg_page(sge));
2668
2669                 sk_msg_iter_var_next(i);
2670         } while (i != last_sge);
2671
2672         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2673
2674         /* To repair sg ring we need to shift entries. If we only
2675          * had a single entry though we can just replace it and
2676          * be done. Otherwise walk the ring and shift the entries.
2677          */
2678         WARN_ON_ONCE(last_sge == first_sge);
2679         shift = last_sge > first_sge ?
2680                 last_sge - first_sge - 1 :
2681                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2682         if (!shift)
2683                 goto out;
2684
2685         i = first_sge;
2686         sk_msg_iter_var_next(i);
2687         do {
2688                 u32 move_from;
2689
2690                 if (i + shift >= NR_MSG_FRAG_IDS)
2691                         move_from = i + shift - NR_MSG_FRAG_IDS;
2692                 else
2693                         move_from = i + shift;
2694                 if (move_from == msg->sg.end)
2695                         break;
2696
2697                 msg->sg.data[i] = msg->sg.data[move_from];
2698                 msg->sg.data[move_from].length = 0;
2699                 msg->sg.data[move_from].page_link = 0;
2700                 msg->sg.data[move_from].offset = 0;
2701                 sk_msg_iter_var_next(i);
2702         } while (1);
2703
2704         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2705                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2706                       msg->sg.end - shift;
2707 out:
2708         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2709         msg->data_end = msg->data + bytes;
2710         return 0;
2711 }
2712
2713 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2714         .func           = bpf_msg_pull_data,
2715         .gpl_only       = false,
2716         .ret_type       = RET_INTEGER,
2717         .arg1_type      = ARG_PTR_TO_CTX,
2718         .arg2_type      = ARG_ANYTHING,
2719         .arg3_type      = ARG_ANYTHING,
2720         .arg4_type      = ARG_ANYTHING,
2721 };
2722
2723 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2724            u32, len, u64, flags)
2725 {
2726         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2727         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2728         u8 *raw, *to, *from;
2729         struct page *page;
2730
2731         if (unlikely(flags))
2732                 return -EINVAL;
2733
2734         if (unlikely(len == 0))
2735                 return 0;
2736
2737         /* First find the starting scatterlist element */
2738         i = msg->sg.start;
2739         do {
2740                 offset += l;
2741                 l = sk_msg_elem(msg, i)->length;
2742
2743                 if (start < offset + l)
2744                         break;
2745                 sk_msg_iter_var_next(i);
2746         } while (i != msg->sg.end);
2747
2748         if (start >= offset + l)
2749                 return -EINVAL;
2750
2751         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2752
2753         /* If no space available will fallback to copy, we need at
2754          * least one scatterlist elem available to push data into
2755          * when start aligns to the beginning of an element or two
2756          * when it falls inside an element. We handle the start equals
2757          * offset case because its the common case for inserting a
2758          * header.
2759          */
2760         if (!space || (space == 1 && start != offset))
2761                 copy = msg->sg.data[i].length;
2762
2763         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2764                            get_order(copy + len));
2765         if (unlikely(!page))
2766                 return -ENOMEM;
2767
2768         if (copy) {
2769                 int front, back;
2770
2771                 raw = page_address(page);
2772
2773                 psge = sk_msg_elem(msg, i);
2774                 front = start - offset;
2775                 back = psge->length - front;
2776                 from = sg_virt(psge);
2777
2778                 if (front)
2779                         memcpy(raw, from, front);
2780
2781                 if (back) {
2782                         from += front;
2783                         to = raw + front + len;
2784
2785                         memcpy(to, from, back);
2786                 }
2787
2788                 put_page(sg_page(psge));
2789         } else if (start - offset) {
2790                 psge = sk_msg_elem(msg, i);
2791                 rsge = sk_msg_elem_cpy(msg, i);
2792
2793                 psge->length = start - offset;
2794                 rsge.length -= psge->length;
2795                 rsge.offset += start;
2796
2797                 sk_msg_iter_var_next(i);
2798                 sg_unmark_end(psge);
2799                 sg_unmark_end(&rsge);
2800                 sk_msg_iter_next(msg, end);
2801         }
2802
2803         /* Slot(s) to place newly allocated data */
2804         new = i;
2805
2806         /* Shift one or two slots as needed */
2807         if (!copy) {
2808                 sge = sk_msg_elem_cpy(msg, i);
2809
2810                 sk_msg_iter_var_next(i);
2811                 sg_unmark_end(&sge);
2812                 sk_msg_iter_next(msg, end);
2813
2814                 nsge = sk_msg_elem_cpy(msg, i);
2815                 if (rsge.length) {
2816                         sk_msg_iter_var_next(i);
2817                         nnsge = sk_msg_elem_cpy(msg, i);
2818                 }
2819
2820                 while (i != msg->sg.end) {
2821                         msg->sg.data[i] = sge;
2822                         sge = nsge;
2823                         sk_msg_iter_var_next(i);
2824                         if (rsge.length) {
2825                                 nsge = nnsge;
2826                                 nnsge = sk_msg_elem_cpy(msg, i);
2827                         } else {
2828                                 nsge = sk_msg_elem_cpy(msg, i);
2829                         }
2830                 }
2831         }
2832
2833         /* Place newly allocated data buffer */
2834         sk_mem_charge(msg->sk, len);
2835         msg->sg.size += len;
2836         __clear_bit(new, &msg->sg.copy);
2837         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2838         if (rsge.length) {
2839                 get_page(sg_page(&rsge));
2840                 sk_msg_iter_var_next(new);
2841                 msg->sg.data[new] = rsge;
2842         }
2843
2844         sk_msg_compute_data_pointers(msg);
2845         return 0;
2846 }
2847
2848 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2849         .func           = bpf_msg_push_data,
2850         .gpl_only       = false,
2851         .ret_type       = RET_INTEGER,
2852         .arg1_type      = ARG_PTR_TO_CTX,
2853         .arg2_type      = ARG_ANYTHING,
2854         .arg3_type      = ARG_ANYTHING,
2855         .arg4_type      = ARG_ANYTHING,
2856 };
2857
2858 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2859 {
2860         int prev;
2861
2862         do {
2863                 prev = i;
2864                 sk_msg_iter_var_next(i);
2865                 msg->sg.data[prev] = msg->sg.data[i];
2866         } while (i != msg->sg.end);
2867
2868         sk_msg_iter_prev(msg, end);
2869 }
2870
2871 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2872 {
2873         struct scatterlist tmp, sge;
2874
2875         sk_msg_iter_next(msg, end);
2876         sge = sk_msg_elem_cpy(msg, i);
2877         sk_msg_iter_var_next(i);
2878         tmp = sk_msg_elem_cpy(msg, i);
2879
2880         while (i != msg->sg.end) {
2881                 msg->sg.data[i] = sge;
2882                 sk_msg_iter_var_next(i);
2883                 sge = tmp;
2884                 tmp = sk_msg_elem_cpy(msg, i);
2885         }
2886 }
2887
2888 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2889            u32, len, u64, flags)
2890 {
2891         u32 i = 0, l = 0, space, offset = 0;
2892         u64 last = start + len;
2893         int pop;
2894
2895         if (unlikely(flags))
2896                 return -EINVAL;
2897
2898         /* First find the starting scatterlist element */
2899         i = msg->sg.start;
2900         do {
2901                 offset += l;
2902                 l = sk_msg_elem(msg, i)->length;
2903
2904                 if (start < offset + l)
2905                         break;
2906                 sk_msg_iter_var_next(i);
2907         } while (i != msg->sg.end);
2908
2909         /* Bounds checks: start and pop must be inside message */
2910         if (start >= offset + l || last >= msg->sg.size)
2911                 return -EINVAL;
2912
2913         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2914
2915         pop = len;
2916         /* --------------| offset
2917          * -| start      |-------- len -------|
2918          *
2919          *  |----- a ----|-------- pop -------|----- b ----|
2920          *  |______________________________________________| length
2921          *
2922          *
2923          * a:   region at front of scatter element to save
2924          * b:   region at back of scatter element to save when length > A + pop
2925          * pop: region to pop from element, same as input 'pop' here will be
2926          *      decremented below per iteration.
2927          *
2928          * Two top-level cases to handle when start != offset, first B is non
2929          * zero and second B is zero corresponding to when a pop includes more
2930          * than one element.
2931          *
2932          * Then if B is non-zero AND there is no space allocate space and
2933          * compact A, B regions into page. If there is space shift ring to
2934          * the rigth free'ing the next element in ring to place B, leaving
2935          * A untouched except to reduce length.
2936          */
2937         if (start != offset) {
2938                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2939                 int a = start;
2940                 int b = sge->length - pop - a;
2941
2942                 sk_msg_iter_var_next(i);
2943
2944                 if (pop < sge->length - a) {
2945                         if (space) {
2946                                 sge->length = a;
2947                                 sk_msg_shift_right(msg, i);
2948                                 nsge = sk_msg_elem(msg, i);
2949                                 get_page(sg_page(sge));
2950                                 sg_set_page(nsge,
2951                                             sg_page(sge),
2952                                             b, sge->offset + pop + a);
2953                         } else {
2954                                 struct page *page, *orig;
2955                                 u8 *to, *from;
2956
2957                                 page = alloc_pages(__GFP_NOWARN |
2958                                                    __GFP_COMP   | GFP_ATOMIC,
2959                                                    get_order(a + b));
2960                                 if (unlikely(!page))
2961                                         return -ENOMEM;
2962
2963                                 sge->length = a;
2964                                 orig = sg_page(sge);
2965                                 from = sg_virt(sge);
2966                                 to = page_address(page);
2967                                 memcpy(to, from, a);
2968                                 memcpy(to + a, from + a + pop, b);
2969                                 sg_set_page(sge, page, a + b, 0);
2970                                 put_page(orig);
2971                         }
2972                         pop = 0;
2973                 } else if (pop >= sge->length - a) {
2974                         pop -= (sge->length - a);
2975                         sge->length = a;
2976                 }
2977         }
2978
2979         /* From above the current layout _must_ be as follows,
2980          *
2981          * -| offset
2982          * -| start
2983          *
2984          *  |---- pop ---|---------------- b ------------|
2985          *  |____________________________________________| length
2986          *
2987          * Offset and start of the current msg elem are equal because in the
2988          * previous case we handled offset != start and either consumed the
2989          * entire element and advanced to the next element OR pop == 0.
2990          *
2991          * Two cases to handle here are first pop is less than the length
2992          * leaving some remainder b above. Simply adjust the element's layout
2993          * in this case. Or pop >= length of the element so that b = 0. In this
2994          * case advance to next element decrementing pop.
2995          */
2996         while (pop) {
2997                 struct scatterlist *sge = sk_msg_elem(msg, i);
2998
2999                 if (pop < sge->length) {
3000                         sge->length -= pop;
3001                         sge->offset += pop;
3002                         pop = 0;
3003                 } else {
3004                         pop -= sge->length;
3005                         sk_msg_shift_left(msg, i);
3006                 }
3007                 sk_msg_iter_var_next(i);
3008         }
3009
3010         sk_mem_uncharge(msg->sk, len - pop);
3011         msg->sg.size -= (len - pop);
3012         sk_msg_compute_data_pointers(msg);
3013         return 0;
3014 }
3015
3016 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3017         .func           = bpf_msg_pop_data,
3018         .gpl_only       = false,
3019         .ret_type       = RET_INTEGER,
3020         .arg1_type      = ARG_PTR_TO_CTX,
3021         .arg2_type      = ARG_ANYTHING,
3022         .arg3_type      = ARG_ANYTHING,
3023         .arg4_type      = ARG_ANYTHING,
3024 };
3025
3026 #ifdef CONFIG_CGROUP_NET_CLASSID
3027 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3028 {
3029         return __task_get_classid(current);
3030 }
3031
3032 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3033         .func           = bpf_get_cgroup_classid_curr,
3034         .gpl_only       = false,
3035         .ret_type       = RET_INTEGER,
3036 };
3037
3038 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3039 {
3040         struct sock *sk = skb_to_full_sk(skb);
3041
3042         if (!sk || !sk_fullsock(sk))
3043                 return 0;
3044
3045         return sock_cgroup_classid(&sk->sk_cgrp_data);
3046 }
3047
3048 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3049         .func           = bpf_skb_cgroup_classid,
3050         .gpl_only       = false,
3051         .ret_type       = RET_INTEGER,
3052         .arg1_type      = ARG_PTR_TO_CTX,
3053 };
3054 #endif
3055
3056 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3057 {
3058         return task_get_classid(skb);
3059 }
3060
3061 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3062         .func           = bpf_get_cgroup_classid,
3063         .gpl_only       = false,
3064         .ret_type       = RET_INTEGER,
3065         .arg1_type      = ARG_PTR_TO_CTX,
3066 };
3067
3068 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3069 {
3070         return dst_tclassid(skb);
3071 }
3072
3073 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3074         .func           = bpf_get_route_realm,
3075         .gpl_only       = false,
3076         .ret_type       = RET_INTEGER,
3077         .arg1_type      = ARG_PTR_TO_CTX,
3078 };
3079
3080 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3081 {
3082         /* If skb_clear_hash() was called due to mangling, we can
3083          * trigger SW recalculation here. Later access to hash
3084          * can then use the inline skb->hash via context directly
3085          * instead of calling this helper again.
3086          */
3087         return skb_get_hash(skb);
3088 }
3089
3090 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3091         .func           = bpf_get_hash_recalc,
3092         .gpl_only       = false,
3093         .ret_type       = RET_INTEGER,
3094         .arg1_type      = ARG_PTR_TO_CTX,
3095 };
3096
3097 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3098 {
3099         /* After all direct packet write, this can be used once for
3100          * triggering a lazy recalc on next skb_get_hash() invocation.
3101          */
3102         skb_clear_hash(skb);
3103         return 0;
3104 }
3105
3106 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3107         .func           = bpf_set_hash_invalid,
3108         .gpl_only       = false,
3109         .ret_type       = RET_INTEGER,
3110         .arg1_type      = ARG_PTR_TO_CTX,
3111 };
3112
3113 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3114 {
3115         /* Set user specified hash as L4(+), so that it gets returned
3116          * on skb_get_hash() call unless BPF prog later on triggers a
3117          * skb_clear_hash().
3118          */
3119         __skb_set_sw_hash(skb, hash, true);
3120         return 0;
3121 }
3122
3123 static const struct bpf_func_proto bpf_set_hash_proto = {
3124         .func           = bpf_set_hash,
3125         .gpl_only       = false,
3126         .ret_type       = RET_INTEGER,
3127         .arg1_type      = ARG_PTR_TO_CTX,
3128         .arg2_type      = ARG_ANYTHING,
3129 };
3130
3131 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3132            u16, vlan_tci)
3133 {
3134         int ret;
3135
3136         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3137                      vlan_proto != htons(ETH_P_8021AD)))
3138                 vlan_proto = htons(ETH_P_8021Q);
3139
3140         bpf_push_mac_rcsum(skb);
3141         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3142         bpf_pull_mac_rcsum(skb);
3143
3144         bpf_compute_data_pointers(skb);
3145         return ret;
3146 }
3147
3148 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3149         .func           = bpf_skb_vlan_push,
3150         .gpl_only       = false,
3151         .ret_type       = RET_INTEGER,
3152         .arg1_type      = ARG_PTR_TO_CTX,
3153         .arg2_type      = ARG_ANYTHING,
3154         .arg3_type      = ARG_ANYTHING,
3155 };
3156
3157 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3158 {
3159         int ret;
3160
3161         bpf_push_mac_rcsum(skb);
3162         ret = skb_vlan_pop(skb);
3163         bpf_pull_mac_rcsum(skb);
3164
3165         bpf_compute_data_pointers(skb);
3166         return ret;
3167 }
3168
3169 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3170         .func           = bpf_skb_vlan_pop,
3171         .gpl_only       = false,
3172         .ret_type       = RET_INTEGER,
3173         .arg1_type      = ARG_PTR_TO_CTX,
3174 };
3175
3176 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3177 {
3178         /* Caller already did skb_cow() with len as headroom,
3179          * so no need to do it here.
3180          */
3181         skb_push(skb, len);
3182         memmove(skb->data, skb->data + len, off);
3183         memset(skb->data + off, 0, len);
3184
3185         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3186          * needed here as it does not change the skb->csum
3187          * result for checksum complete when summing over
3188          * zeroed blocks.
3189          */
3190         return 0;
3191 }
3192
3193 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3194 {
3195         /* skb_ensure_writable() is not needed here, as we're
3196          * already working on an uncloned skb.
3197          */
3198         if (unlikely(!pskb_may_pull(skb, off + len)))
3199                 return -ENOMEM;
3200
3201         skb_postpull_rcsum(skb, skb->data + off, len);
3202         memmove(skb->data + len, skb->data, off);
3203         __skb_pull(skb, len);
3204
3205         return 0;
3206 }
3207
3208 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3209 {
3210         bool trans_same = skb->transport_header == skb->network_header;
3211         int ret;
3212
3213         /* There's no need for __skb_push()/__skb_pull() pair to
3214          * get to the start of the mac header as we're guaranteed
3215          * to always start from here under eBPF.
3216          */
3217         ret = bpf_skb_generic_push(skb, off, len);
3218         if (likely(!ret)) {
3219                 skb->mac_header -= len;
3220                 skb->network_header -= len;
3221                 if (trans_same)
3222                         skb->transport_header = skb->network_header;
3223         }
3224
3225         return ret;
3226 }
3227
3228 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3229 {
3230         bool trans_same = skb->transport_header == skb->network_header;
3231         int ret;
3232
3233         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3234         ret = bpf_skb_generic_pop(skb, off, len);
3235         if (likely(!ret)) {
3236                 skb->mac_header += len;
3237                 skb->network_header += len;
3238                 if (trans_same)
3239                         skb->transport_header = skb->network_header;
3240         }
3241
3242         return ret;
3243 }
3244
3245 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3246 {
3247         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3248         u32 off = skb_mac_header_len(skb);
3249         int ret;
3250
3251         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3252                 return -ENOTSUPP;
3253
3254         ret = skb_cow(skb, len_diff);
3255         if (unlikely(ret < 0))
3256                 return ret;
3257
3258         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3259         if (unlikely(ret < 0))
3260                 return ret;
3261
3262         if (skb_is_gso(skb)) {
3263                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3264
3265                 /* SKB_GSO_TCPV4 needs to be changed into
3266                  * SKB_GSO_TCPV6.
3267                  */
3268                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3269                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3270                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3271                 }
3272
3273                 /* Header must be checked, and gso_segs recomputed. */
3274                 shinfo->gso_type |= SKB_GSO_DODGY;
3275                 shinfo->gso_segs = 0;
3276         }
3277
3278         skb->protocol = htons(ETH_P_IPV6);
3279         skb_clear_hash(skb);
3280
3281         return 0;
3282 }
3283
3284 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3285 {
3286         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3287         u32 off = skb_mac_header_len(skb);
3288         int ret;
3289
3290         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3291                 return -ENOTSUPP;
3292
3293         ret = skb_unclone(skb, GFP_ATOMIC);
3294         if (unlikely(ret < 0))
3295                 return ret;
3296
3297         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3298         if (unlikely(ret < 0))
3299                 return ret;
3300
3301         if (skb_is_gso(skb)) {
3302                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3303
3304                 /* SKB_GSO_TCPV6 needs to be changed into
3305                  * SKB_GSO_TCPV4.
3306                  */
3307                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3308                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3309                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3310                 }
3311
3312                 /* Header must be checked, and gso_segs recomputed. */
3313                 shinfo->gso_type |= SKB_GSO_DODGY;
3314                 shinfo->gso_segs = 0;
3315         }
3316
3317         skb->protocol = htons(ETH_P_IP);
3318         skb_clear_hash(skb);
3319
3320         return 0;
3321 }
3322
3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3324 {
3325         __be16 from_proto = skb->protocol;
3326
3327         if (from_proto == htons(ETH_P_IP) &&
3328               to_proto == htons(ETH_P_IPV6))
3329                 return bpf_skb_proto_4_to_6(skb);
3330
3331         if (from_proto == htons(ETH_P_IPV6) &&
3332               to_proto == htons(ETH_P_IP))
3333                 return bpf_skb_proto_6_to_4(skb);
3334
3335         return -ENOTSUPP;
3336 }
3337
3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3339            u64, flags)
3340 {
3341         int ret;
3342
3343         if (unlikely(flags))
3344                 return -EINVAL;
3345
3346         /* General idea is that this helper does the basic groundwork
3347          * needed for changing the protocol, and eBPF program fills the
3348          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3349          * and other helpers, rather than passing a raw buffer here.
3350          *
3351          * The rationale is to keep this minimal and without a need to
3352          * deal with raw packet data. F.e. even if we would pass buffers
3353          * here, the program still needs to call the bpf_lX_csum_replace()
3354          * helpers anyway. Plus, this way we keep also separation of
3355          * concerns, since f.e. bpf_skb_store_bytes() should only take
3356          * care of stores.
3357          *
3358          * Currently, additional options and extension header space are
3359          * not supported, but flags register is reserved so we can adapt
3360          * that. For offloads, we mark packet as dodgy, so that headers
3361          * need to be verified first.
3362          */
3363         ret = bpf_skb_proto_xlat(skb, proto);
3364         bpf_compute_data_pointers(skb);
3365         return ret;
3366 }
3367
3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3369         .func           = bpf_skb_change_proto,
3370         .gpl_only       = false,
3371         .ret_type       = RET_INTEGER,
3372         .arg1_type      = ARG_PTR_TO_CTX,
3373         .arg2_type      = ARG_ANYTHING,
3374         .arg3_type      = ARG_ANYTHING,
3375 };
3376
3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3378 {
3379         /* We only allow a restricted subset to be changed for now. */
3380         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3381                      !skb_pkt_type_ok(pkt_type)))
3382                 return -EINVAL;
3383
3384         skb->pkt_type = pkt_type;
3385         return 0;
3386 }
3387
3388 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3389         .func           = bpf_skb_change_type,
3390         .gpl_only       = false,
3391         .ret_type       = RET_INTEGER,
3392         .arg1_type      = ARG_PTR_TO_CTX,
3393         .arg2_type      = ARG_ANYTHING,
3394 };
3395
3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3397 {
3398         switch (skb->protocol) {
3399         case htons(ETH_P_IP):
3400                 return sizeof(struct iphdr);
3401         case htons(ETH_P_IPV6):
3402                 return sizeof(struct ipv6hdr);
3403         default:
3404                 return ~0U;
3405         }
3406 }
3407
3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3409                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3410
3411 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3412                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3413                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3414                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3415                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3416                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3417
3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3419                             u64 flags)
3420 {
3421         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3422         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3423         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3424         unsigned int gso_type = SKB_GSO_DODGY;
3425         int ret;
3426
3427         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3428                 /* udp gso_size delineates datagrams, only allow if fixed */
3429                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3430                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3431                         return -ENOTSUPP;
3432         }
3433
3434         ret = skb_cow_head(skb, len_diff);
3435         if (unlikely(ret < 0))
3436                 return ret;
3437
3438         if (encap) {
3439                 if (skb->protocol != htons(ETH_P_IP) &&
3440                     skb->protocol != htons(ETH_P_IPV6))
3441                         return -ENOTSUPP;
3442
3443                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3444                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3445                         return -EINVAL;
3446
3447                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3448                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3449                         return -EINVAL;
3450
3451                 if (skb->encapsulation)
3452                         return -EALREADY;
3453
3454                 mac_len = skb->network_header - skb->mac_header;
3455                 inner_net = skb->network_header;
3456                 if (inner_mac_len > len_diff)
3457                         return -EINVAL;
3458                 inner_trans = skb->transport_header;
3459         }
3460
3461         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3462         if (unlikely(ret < 0))
3463                 return ret;
3464
3465         if (encap) {
3466                 skb->inner_mac_header = inner_net - inner_mac_len;
3467                 skb->inner_network_header = inner_net;
3468                 skb->inner_transport_header = inner_trans;
3469                 skb_set_inner_protocol(skb, skb->protocol);
3470
3471                 skb->encapsulation = 1;
3472                 skb_set_network_header(skb, mac_len);
3473
3474                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3475                         gso_type |= SKB_GSO_UDP_TUNNEL;
3476                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3477                         gso_type |= SKB_GSO_GRE;
3478                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479                         gso_type |= SKB_GSO_IPXIP6;
3480                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3481                         gso_type |= SKB_GSO_IPXIP4;
3482
3483                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3484                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3485                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3486                                         sizeof(struct ipv6hdr) :
3487                                         sizeof(struct iphdr);
3488
3489                         skb_set_transport_header(skb, mac_len + nh_len);
3490                 }
3491
3492                 /* Match skb->protocol to new outer l3 protocol */
3493                 if (skb->protocol == htons(ETH_P_IP) &&
3494                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3495                         skb->protocol = htons(ETH_P_IPV6);
3496                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3497                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3498                         skb->protocol = htons(ETH_P_IP);
3499         }
3500
3501         if (skb_is_gso(skb)) {
3502                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3503
3504                 /* Due to header grow, MSS needs to be downgraded. */
3505                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3506                         skb_decrease_gso_size(shinfo, len_diff);
3507
3508                 /* Header must be checked, and gso_segs recomputed. */
3509                 shinfo->gso_type |= gso_type;
3510                 shinfo->gso_segs = 0;
3511         }
3512
3513         return 0;
3514 }
3515
3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3517                               u64 flags)
3518 {
3519         int ret;
3520
3521         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3522                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3523                 return -EINVAL;
3524
3525         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3526                 /* udp gso_size delineates datagrams, only allow if fixed */
3527                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3528                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529                         return -ENOTSUPP;
3530         }
3531
3532         ret = skb_unclone(skb, GFP_ATOMIC);
3533         if (unlikely(ret < 0))
3534                 return ret;
3535
3536         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3537         if (unlikely(ret < 0))
3538                 return ret;
3539
3540         if (skb_is_gso(skb)) {
3541                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3542
3543                 /* Due to header shrink, MSS can be upgraded. */
3544                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3545                         skb_increase_gso_size(shinfo, len_diff);
3546
3547                 /* Header must be checked, and gso_segs recomputed. */
3548                 shinfo->gso_type |= SKB_GSO_DODGY;
3549                 shinfo->gso_segs = 0;
3550         }
3551
3552         return 0;
3553 }
3554
3555 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3556
3557 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3558            u32, mode, u64, flags)
3559 {
3560         u32 len_diff_abs = abs(len_diff);
3561         bool shrink = len_diff < 0;
3562         int ret = 0;
3563
3564         if (unlikely(flags || mode))
3565                 return -EINVAL;
3566         if (unlikely(len_diff_abs > 0xfffU))
3567                 return -EFAULT;
3568
3569         if (!shrink) {
3570                 ret = skb_cow(skb, len_diff);
3571                 if (unlikely(ret < 0))
3572                         return ret;
3573                 __skb_push(skb, len_diff_abs);
3574                 memset(skb->data, 0, len_diff_abs);
3575         } else {
3576                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3577                         return -ENOMEM;
3578                 __skb_pull(skb, len_diff_abs);
3579         }
3580         bpf_compute_data_end_sk_skb(skb);
3581         if (tls_sw_has_ctx_rx(skb->sk)) {
3582                 struct strp_msg *rxm = strp_msg(skb);
3583
3584                 rxm->full_len += len_diff;
3585         }
3586         return ret;
3587 }
3588
3589 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3590         .func           = sk_skb_adjust_room,
3591         .gpl_only       = false,
3592         .ret_type       = RET_INTEGER,
3593         .arg1_type      = ARG_PTR_TO_CTX,
3594         .arg2_type      = ARG_ANYTHING,
3595         .arg3_type      = ARG_ANYTHING,
3596         .arg4_type      = ARG_ANYTHING,
3597 };
3598
3599 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3600            u32, mode, u64, flags)
3601 {
3602         u32 len_cur, len_diff_abs = abs(len_diff);
3603         u32 len_min = bpf_skb_net_base_len(skb);
3604         u32 len_max = BPF_SKB_MAX_LEN;
3605         __be16 proto = skb->protocol;
3606         bool shrink = len_diff < 0;
3607         u32 off;
3608         int ret;
3609
3610         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3611                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3612                 return -EINVAL;
3613         if (unlikely(len_diff_abs > 0xfffU))
3614                 return -EFAULT;
3615         if (unlikely(proto != htons(ETH_P_IP) &&
3616                      proto != htons(ETH_P_IPV6)))
3617                 return -ENOTSUPP;
3618
3619         off = skb_mac_header_len(skb);
3620         switch (mode) {
3621         case BPF_ADJ_ROOM_NET:
3622                 off += bpf_skb_net_base_len(skb);
3623                 break;
3624         case BPF_ADJ_ROOM_MAC:
3625                 break;
3626         default:
3627                 return -ENOTSUPP;
3628         }
3629
3630         len_cur = skb->len - skb_network_offset(skb);
3631         if ((shrink && (len_diff_abs >= len_cur ||
3632                         len_cur - len_diff_abs < len_min)) ||
3633             (!shrink && (skb->len + len_diff_abs > len_max &&
3634                          !skb_is_gso(skb))))
3635                 return -ENOTSUPP;
3636
3637         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3638                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3639         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3640                 __skb_reset_checksum_unnecessary(skb);
3641
3642         bpf_compute_data_pointers(skb);
3643         return ret;
3644 }
3645
3646 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3647         .func           = bpf_skb_adjust_room,
3648         .gpl_only       = false,
3649         .ret_type       = RET_INTEGER,
3650         .arg1_type      = ARG_PTR_TO_CTX,
3651         .arg2_type      = ARG_ANYTHING,
3652         .arg3_type      = ARG_ANYTHING,
3653         .arg4_type      = ARG_ANYTHING,
3654 };
3655
3656 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3657 {
3658         u32 min_len = skb_network_offset(skb);
3659
3660         if (skb_transport_header_was_set(skb))
3661                 min_len = skb_transport_offset(skb);
3662         if (skb->ip_summed == CHECKSUM_PARTIAL)
3663                 min_len = skb_checksum_start_offset(skb) +
3664                           skb->csum_offset + sizeof(__sum16);
3665         return min_len;
3666 }
3667
3668 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3669 {
3670         unsigned int old_len = skb->len;
3671         int ret;
3672
3673         ret = __skb_grow_rcsum(skb, new_len);
3674         if (!ret)
3675                 memset(skb->data + old_len, 0, new_len - old_len);
3676         return ret;
3677 }
3678
3679 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3680 {
3681         return __skb_trim_rcsum(skb, new_len);
3682 }
3683
3684 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3685                                         u64 flags)
3686 {
3687         u32 max_len = BPF_SKB_MAX_LEN;
3688         u32 min_len = __bpf_skb_min_len(skb);
3689         int ret;
3690
3691         if (unlikely(flags || new_len > max_len || new_len < min_len))
3692                 return -EINVAL;
3693         if (skb->encapsulation)
3694                 return -ENOTSUPP;
3695
3696         /* The basic idea of this helper is that it's performing the
3697          * needed work to either grow or trim an skb, and eBPF program
3698          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3699          * bpf_lX_csum_replace() and others rather than passing a raw
3700          * buffer here. This one is a slow path helper and intended
3701          * for replies with control messages.
3702          *
3703          * Like in bpf_skb_change_proto(), we want to keep this rather
3704          * minimal and without protocol specifics so that we are able
3705          * to separate concerns as in bpf_skb_store_bytes() should only
3706          * be the one responsible for writing buffers.
3707          *
3708          * It's really expected to be a slow path operation here for
3709          * control message replies, so we're implicitly linearizing,
3710          * uncloning and drop offloads from the skb by this.
3711          */
3712         ret = __bpf_try_make_writable(skb, skb->len);
3713         if (!ret) {
3714                 if (new_len > skb->len)
3715                         ret = bpf_skb_grow_rcsum(skb, new_len);
3716                 else if (new_len < skb->len)
3717                         ret = bpf_skb_trim_rcsum(skb, new_len);
3718                 if (!ret && skb_is_gso(skb))
3719                         skb_gso_reset(skb);
3720         }
3721         return ret;
3722 }
3723
3724 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3725            u64, flags)
3726 {
3727         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3728
3729         bpf_compute_data_pointers(skb);
3730         return ret;
3731 }
3732
3733 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3734         .func           = bpf_skb_change_tail,
3735         .gpl_only       = false,
3736         .ret_type       = RET_INTEGER,
3737         .arg1_type      = ARG_PTR_TO_CTX,
3738         .arg2_type      = ARG_ANYTHING,
3739         .arg3_type      = ARG_ANYTHING,
3740 };
3741
3742 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3743            u64, flags)
3744 {
3745         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3746
3747         bpf_compute_data_end_sk_skb(skb);
3748         return ret;
3749 }
3750
3751 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3752         .func           = sk_skb_change_tail,
3753         .gpl_only       = false,
3754         .ret_type       = RET_INTEGER,
3755         .arg1_type      = ARG_PTR_TO_CTX,
3756         .arg2_type      = ARG_ANYTHING,
3757         .arg3_type      = ARG_ANYTHING,
3758 };
3759
3760 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3761                                         u64 flags)
3762 {
3763         u32 max_len = BPF_SKB_MAX_LEN;
3764         u32 new_len = skb->len + head_room;
3765         int ret;
3766
3767         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3768                      new_len < skb->len))
3769                 return -EINVAL;
3770
3771         ret = skb_cow(skb, head_room);
3772         if (likely(!ret)) {
3773                 /* Idea for this helper is that we currently only
3774                  * allow to expand on mac header. This means that
3775                  * skb->protocol network header, etc, stay as is.
3776                  * Compared to bpf_skb_change_tail(), we're more
3777                  * flexible due to not needing to linearize or
3778                  * reset GSO. Intention for this helper is to be
3779                  * used by an L3 skb that needs to push mac header
3780                  * for redirection into L2 device.
3781                  */
3782                 __skb_push(skb, head_room);
3783                 memset(skb->data, 0, head_room);
3784                 skb_reset_mac_header(skb);
3785                 skb_reset_mac_len(skb);
3786         }
3787
3788         return ret;
3789 }
3790
3791 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3792            u64, flags)
3793 {
3794         int ret = __bpf_skb_change_head(skb, head_room, flags);
3795
3796         bpf_compute_data_pointers(skb);
3797         return ret;
3798 }
3799
3800 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3801         .func           = bpf_skb_change_head,
3802         .gpl_only       = false,
3803         .ret_type       = RET_INTEGER,
3804         .arg1_type      = ARG_PTR_TO_CTX,
3805         .arg2_type      = ARG_ANYTHING,
3806         .arg3_type      = ARG_ANYTHING,
3807 };
3808
3809 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3810            u64, flags)
3811 {
3812         int ret = __bpf_skb_change_head(skb, head_room, flags);
3813
3814         bpf_compute_data_end_sk_skb(skb);
3815         return ret;
3816 }
3817
3818 static const struct bpf_func_proto sk_skb_change_head_proto = {
3819         .func           = sk_skb_change_head,
3820         .gpl_only       = false,
3821         .ret_type       = RET_INTEGER,
3822         .arg1_type      = ARG_PTR_TO_CTX,
3823         .arg2_type      = ARG_ANYTHING,
3824         .arg3_type      = ARG_ANYTHING,
3825 };
3826 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3827 {
3828         return xdp_data_meta_unsupported(xdp) ? 0 :
3829                xdp->data - xdp->data_meta;
3830 }
3831
3832 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3833 {
3834         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3835         unsigned long metalen = xdp_get_metalen(xdp);
3836         void *data_start = xdp_frame_end + metalen;
3837         void *data = xdp->data + offset;
3838
3839         if (unlikely(data < data_start ||
3840                      data > xdp->data_end - ETH_HLEN))
3841                 return -EINVAL;
3842
3843         if (metalen)
3844                 memmove(xdp->data_meta + offset,
3845                         xdp->data_meta, metalen);
3846         xdp->data_meta += offset;
3847         xdp->data = data;
3848
3849         return 0;
3850 }
3851
3852 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3853         .func           = bpf_xdp_adjust_head,
3854         .gpl_only       = false,
3855         .ret_type       = RET_INTEGER,
3856         .arg1_type      = ARG_PTR_TO_CTX,
3857         .arg2_type      = ARG_ANYTHING,
3858 };
3859
3860 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3861 {
3862         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3863         void *data_end = xdp->data_end + offset;
3864
3865         /* Notice that xdp_data_hard_end have reserved some tailroom */
3866         if (unlikely(data_end > data_hard_end))
3867                 return -EINVAL;
3868
3869         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3870         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3871                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3872                 return -EINVAL;
3873         }
3874
3875         if (unlikely(data_end < xdp->data + ETH_HLEN))
3876                 return -EINVAL;
3877
3878         /* Clear memory area on grow, can contain uninit kernel memory */
3879         if (offset > 0)
3880                 memset(xdp->data_end, 0, offset);
3881
3882         xdp->data_end = data_end;
3883
3884         return 0;
3885 }
3886
3887 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3888         .func           = bpf_xdp_adjust_tail,
3889         .gpl_only       = false,
3890         .ret_type       = RET_INTEGER,
3891         .arg1_type      = ARG_PTR_TO_CTX,
3892         .arg2_type      = ARG_ANYTHING,
3893 };
3894
3895 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3896 {
3897         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3898         void *meta = xdp->data_meta + offset;
3899         unsigned long metalen = xdp->data - meta;
3900
3901         if (xdp_data_meta_unsupported(xdp))
3902                 return -ENOTSUPP;
3903         if (unlikely(meta < xdp_frame_end ||
3904                      meta > xdp->data))
3905                 return -EINVAL;
3906         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3907                      (metalen > 32)))
3908                 return -EACCES;
3909
3910         xdp->data_meta = meta;
3911
3912         return 0;
3913 }
3914
3915 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3916         .func           = bpf_xdp_adjust_meta,
3917         .gpl_only       = false,
3918         .ret_type       = RET_INTEGER,
3919         .arg1_type      = ARG_PTR_TO_CTX,
3920         .arg2_type      = ARG_ANYTHING,
3921 };
3922
3923 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3924                             struct bpf_map *map, struct xdp_buff *xdp)
3925 {
3926         switch (map->map_type) {
3927         case BPF_MAP_TYPE_DEVMAP:
3928         case BPF_MAP_TYPE_DEVMAP_HASH:
3929                 return dev_map_enqueue(fwd, xdp, dev_rx);
3930         case BPF_MAP_TYPE_CPUMAP:
3931                 return cpu_map_enqueue(fwd, xdp, dev_rx);
3932         case BPF_MAP_TYPE_XSKMAP:
3933                 return __xsk_map_redirect(fwd, xdp);
3934         default:
3935                 return -EBADRQC;
3936         }
3937         return 0;
3938 }
3939
3940 void xdp_do_flush(void)
3941 {
3942         __dev_flush();
3943         __cpu_map_flush();
3944         __xsk_map_flush();
3945 }
3946 EXPORT_SYMBOL_GPL(xdp_do_flush);
3947
3948 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3949 {
3950         switch (map->map_type) {
3951         case BPF_MAP_TYPE_DEVMAP:
3952                 return __dev_map_lookup_elem(map, index);
3953         case BPF_MAP_TYPE_DEVMAP_HASH:
3954                 return __dev_map_hash_lookup_elem(map, index);
3955         case BPF_MAP_TYPE_CPUMAP:
3956                 return __cpu_map_lookup_elem(map, index);
3957         case BPF_MAP_TYPE_XSKMAP:
3958                 return __xsk_map_lookup_elem(map, index);
3959         default:
3960                 return NULL;
3961         }
3962 }
3963
3964 void bpf_clear_redirect_map(struct bpf_map *map)
3965 {
3966         struct bpf_redirect_info *ri;
3967         int cpu;
3968
3969         for_each_possible_cpu(cpu) {
3970                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3971                 /* Avoid polluting remote cacheline due to writes if
3972                  * not needed. Once we pass this test, we need the
3973                  * cmpxchg() to make sure it hasn't been changed in
3974                  * the meantime by remote CPU.
3975                  */
3976                 if (unlikely(READ_ONCE(ri->map) == map))
3977                         cmpxchg(&ri->map, map, NULL);
3978         }
3979 }
3980
3981 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3982                     struct bpf_prog *xdp_prog)
3983 {
3984         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3985         struct bpf_map *map = READ_ONCE(ri->map);
3986         u32 index = ri->tgt_index;
3987         void *fwd = ri->tgt_value;
3988         int err;
3989
3990         ri->tgt_index = 0;
3991         ri->tgt_value = NULL;
3992         WRITE_ONCE(ri->map, NULL);
3993
3994         if (unlikely(!map)) {
3995                 fwd = dev_get_by_index_rcu(dev_net(dev), index);
3996                 if (unlikely(!fwd)) {
3997                         err = -EINVAL;
3998                         goto err;
3999                 }
4000
4001                 err = dev_xdp_enqueue(fwd, xdp, dev);
4002         } else {
4003                 err = __bpf_tx_xdp_map(dev, fwd, map, xdp);
4004         }
4005
4006         if (unlikely(err))
4007                 goto err;
4008
4009         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4010         return 0;
4011 err:
4012         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4013         return err;
4014 }
4015 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4016
4017 static int xdp_do_generic_redirect_map(struct net_device *dev,
4018                                        struct sk_buff *skb,
4019                                        struct xdp_buff *xdp,
4020                                        struct bpf_prog *xdp_prog,
4021                                        struct bpf_map *map)
4022 {
4023         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4024         u32 index = ri->tgt_index;
4025         void *fwd = ri->tgt_value;
4026         int err = 0;
4027
4028         ri->tgt_index = 0;
4029         ri->tgt_value = NULL;
4030         WRITE_ONCE(ri->map, NULL);
4031
4032         if (map->map_type == BPF_MAP_TYPE_DEVMAP ||
4033             map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) {
4034                 struct bpf_dtab_netdev *dst = fwd;
4035
4036                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
4037                 if (unlikely(err))
4038                         goto err;
4039         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
4040                 struct xdp_sock *xs = fwd;
4041
4042                 err = xsk_generic_rcv(xs, xdp);
4043                 if (err)
4044                         goto err;
4045                 consume_skb(skb);
4046         } else {
4047                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4048                 err = -EBADRQC;
4049                 goto err;
4050         }
4051
4052         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
4053         return 0;
4054 err:
4055         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
4056         return err;
4057 }
4058
4059 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4060                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4061 {
4062         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4063         struct bpf_map *map = READ_ONCE(ri->map);
4064         u32 index = ri->tgt_index;
4065         struct net_device *fwd;
4066         int err = 0;
4067
4068         if (map)
4069                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
4070                                                    map);
4071         ri->tgt_index = 0;
4072         fwd = dev_get_by_index_rcu(dev_net(dev), index);
4073         if (unlikely(!fwd)) {
4074                 err = -EINVAL;
4075                 goto err;
4076         }
4077
4078         err = xdp_ok_fwd_dev(fwd, skb->len);
4079         if (unlikely(err))
4080                 goto err;
4081
4082         skb->dev = fwd;
4083         _trace_xdp_redirect(dev, xdp_prog, index);
4084         generic_xdp_tx(skb, xdp_prog);
4085         return 0;
4086 err:
4087         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
4088         return err;
4089 }
4090
4091 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4092 {
4093         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4094
4095         if (unlikely(flags))
4096                 return XDP_ABORTED;
4097
4098         ri->flags = flags;
4099         ri->tgt_index = ifindex;
4100         ri->tgt_value = NULL;
4101         WRITE_ONCE(ri->map, NULL);
4102
4103         return XDP_REDIRECT;
4104 }
4105
4106 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4107         .func           = bpf_xdp_redirect,
4108         .gpl_only       = false,
4109         .ret_type       = RET_INTEGER,
4110         .arg1_type      = ARG_ANYTHING,
4111         .arg2_type      = ARG_ANYTHING,
4112 };
4113
4114 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4115            u64, flags)
4116 {
4117         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4118
4119         /* Lower bits of the flags are used as return code on lookup failure */
4120         if (unlikely(flags > XDP_TX))
4121                 return XDP_ABORTED;
4122
4123         ri->tgt_value = __xdp_map_lookup_elem(map, ifindex);
4124         if (unlikely(!ri->tgt_value)) {
4125                 /* If the lookup fails we want to clear out the state in the
4126                  * redirect_info struct completely, so that if an eBPF program
4127                  * performs multiple lookups, the last one always takes
4128                  * precedence.
4129                  */
4130                 WRITE_ONCE(ri->map, NULL);
4131                 return flags;
4132         }
4133
4134         ri->flags = flags;
4135         ri->tgt_index = ifindex;
4136         WRITE_ONCE(ri->map, map);
4137
4138         return XDP_REDIRECT;
4139 }
4140
4141 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4142         .func           = bpf_xdp_redirect_map,
4143         .gpl_only       = false,
4144         .ret_type       = RET_INTEGER,
4145         .arg1_type      = ARG_CONST_MAP_PTR,
4146         .arg2_type      = ARG_ANYTHING,
4147         .arg3_type      = ARG_ANYTHING,
4148 };
4149
4150 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4151                                   unsigned long off, unsigned long len)
4152 {
4153         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4154
4155         if (unlikely(!ptr))
4156                 return len;
4157         if (ptr != dst_buff)
4158                 memcpy(dst_buff, ptr, len);
4159
4160         return 0;
4161 }
4162
4163 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4164            u64, flags, void *, meta, u64, meta_size)
4165 {
4166         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4167
4168         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4169                 return -EINVAL;
4170         if (unlikely(!skb || skb_size > skb->len))
4171                 return -EFAULT;
4172
4173         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4174                                 bpf_skb_copy);
4175 }
4176
4177 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4178         .func           = bpf_skb_event_output,
4179         .gpl_only       = true,
4180         .ret_type       = RET_INTEGER,
4181         .arg1_type      = ARG_PTR_TO_CTX,
4182         .arg2_type      = ARG_CONST_MAP_PTR,
4183         .arg3_type      = ARG_ANYTHING,
4184         .arg4_type      = ARG_PTR_TO_MEM,
4185         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4186 };
4187
4188 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4189
4190 const struct bpf_func_proto bpf_skb_output_proto = {
4191         .func           = bpf_skb_event_output,
4192         .gpl_only       = true,
4193         .ret_type       = RET_INTEGER,
4194         .arg1_type      = ARG_PTR_TO_BTF_ID,
4195         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4196         .arg2_type      = ARG_CONST_MAP_PTR,
4197         .arg3_type      = ARG_ANYTHING,
4198         .arg4_type      = ARG_PTR_TO_MEM,
4199         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4200 };
4201
4202 static unsigned short bpf_tunnel_key_af(u64 flags)
4203 {
4204         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4205 }
4206
4207 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4208            u32, size, u64, flags)
4209 {
4210         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4211         u8 compat[sizeof(struct bpf_tunnel_key)];
4212         void *to_orig = to;
4213         int err;
4214
4215         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4216                 err = -EINVAL;
4217                 goto err_clear;
4218         }
4219         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4220                 err = -EPROTO;
4221                 goto err_clear;
4222         }
4223         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4224                 err = -EINVAL;
4225                 switch (size) {
4226                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4227                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4228                         goto set_compat;
4229                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4230                         /* Fixup deprecated structure layouts here, so we have
4231                          * a common path later on.
4232                          */
4233                         if (ip_tunnel_info_af(info) != AF_INET)
4234                                 goto err_clear;
4235 set_compat:
4236                         to = (struct bpf_tunnel_key *)compat;
4237                         break;
4238                 default:
4239                         goto err_clear;
4240                 }
4241         }
4242
4243         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4244         to->tunnel_tos = info->key.tos;
4245         to->tunnel_ttl = info->key.ttl;
4246         to->tunnel_ext = 0;
4247
4248         if (flags & BPF_F_TUNINFO_IPV6) {
4249                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4250                        sizeof(to->remote_ipv6));
4251                 to->tunnel_label = be32_to_cpu(info->key.label);
4252         } else {
4253                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4254                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4255                 to->tunnel_label = 0;
4256         }
4257
4258         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4259                 memcpy(to_orig, to, size);
4260
4261         return 0;
4262 err_clear:
4263         memset(to_orig, 0, size);
4264         return err;
4265 }
4266
4267 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4268         .func           = bpf_skb_get_tunnel_key,
4269         .gpl_only       = false,
4270         .ret_type       = RET_INTEGER,
4271         .arg1_type      = ARG_PTR_TO_CTX,
4272         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4273         .arg3_type      = ARG_CONST_SIZE,
4274         .arg4_type      = ARG_ANYTHING,
4275 };
4276
4277 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4278 {
4279         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4280         int err;
4281
4282         if (unlikely(!info ||
4283                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4284                 err = -ENOENT;
4285                 goto err_clear;
4286         }
4287         if (unlikely(size < info->options_len)) {
4288                 err = -ENOMEM;
4289                 goto err_clear;
4290         }
4291
4292         ip_tunnel_info_opts_get(to, info);
4293         if (size > info->options_len)
4294                 memset(to + info->options_len, 0, size - info->options_len);
4295
4296         return info->options_len;
4297 err_clear:
4298         memset(to, 0, size);
4299         return err;
4300 }
4301
4302 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4303         .func           = bpf_skb_get_tunnel_opt,
4304         .gpl_only       = false,
4305         .ret_type       = RET_INTEGER,
4306         .arg1_type      = ARG_PTR_TO_CTX,
4307         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4308         .arg3_type      = ARG_CONST_SIZE,
4309 };
4310
4311 static struct metadata_dst __percpu *md_dst;
4312
4313 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4314            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4315 {
4316         struct metadata_dst *md = this_cpu_ptr(md_dst);
4317         u8 compat[sizeof(struct bpf_tunnel_key)];
4318         struct ip_tunnel_info *info;
4319
4320         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4321                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4322                 return -EINVAL;
4323         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4324                 switch (size) {
4325                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4326                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4327                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4328                         /* Fixup deprecated structure layouts here, so we have
4329                          * a common path later on.
4330                          */
4331                         memcpy(compat, from, size);
4332                         memset(compat + size, 0, sizeof(compat) - size);
4333                         from = (const struct bpf_tunnel_key *) compat;
4334                         break;
4335                 default:
4336                         return -EINVAL;
4337                 }
4338         }
4339         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4340                      from->tunnel_ext))
4341                 return -EINVAL;
4342
4343         skb_dst_drop(skb);
4344         dst_hold((struct dst_entry *) md);
4345         skb_dst_set(skb, (struct dst_entry *) md);
4346
4347         info = &md->u.tun_info;
4348         memset(info, 0, sizeof(*info));
4349         info->mode = IP_TUNNEL_INFO_TX;
4350
4351         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4352         if (flags & BPF_F_DONT_FRAGMENT)
4353                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4354         if (flags & BPF_F_ZERO_CSUM_TX)
4355                 info->key.tun_flags &= ~TUNNEL_CSUM;
4356         if (flags & BPF_F_SEQ_NUMBER)
4357                 info->key.tun_flags |= TUNNEL_SEQ;
4358
4359         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4360         info->key.tos = from->tunnel_tos;
4361         info->key.ttl = from->tunnel_ttl;
4362
4363         if (flags & BPF_F_TUNINFO_IPV6) {
4364                 info->mode |= IP_TUNNEL_INFO_IPV6;
4365                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4366                        sizeof(from->remote_ipv6));
4367                 info->key.label = cpu_to_be32(from->tunnel_label) &
4368                                   IPV6_FLOWLABEL_MASK;
4369         } else {
4370                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4371         }
4372
4373         return 0;
4374 }
4375
4376 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4377         .func           = bpf_skb_set_tunnel_key,
4378         .gpl_only       = false,
4379         .ret_type       = RET_INTEGER,
4380         .arg1_type      = ARG_PTR_TO_CTX,
4381         .arg2_type      = ARG_PTR_TO_MEM,
4382         .arg3_type      = ARG_CONST_SIZE,
4383         .arg4_type      = ARG_ANYTHING,
4384 };
4385
4386 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4387            const u8 *, from, u32, size)
4388 {
4389         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4390         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4391
4392         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4393                 return -EINVAL;
4394         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4395                 return -ENOMEM;
4396
4397         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4398
4399         return 0;
4400 }
4401
4402 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4403         .func           = bpf_skb_set_tunnel_opt,
4404         .gpl_only       = false,
4405         .ret_type       = RET_INTEGER,
4406         .arg1_type      = ARG_PTR_TO_CTX,
4407         .arg2_type      = ARG_PTR_TO_MEM,
4408         .arg3_type      = ARG_CONST_SIZE,
4409 };
4410
4411 static const struct bpf_func_proto *
4412 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4413 {
4414         if (!md_dst) {
4415                 struct metadata_dst __percpu *tmp;
4416
4417                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4418                                                 METADATA_IP_TUNNEL,
4419                                                 GFP_KERNEL);
4420                 if (!tmp)
4421                         return NULL;
4422                 if (cmpxchg(&md_dst, NULL, tmp))
4423                         metadata_dst_free_percpu(tmp);
4424         }
4425
4426         switch (which) {
4427         case BPF_FUNC_skb_set_tunnel_key:
4428                 return &bpf_skb_set_tunnel_key_proto;
4429         case BPF_FUNC_skb_set_tunnel_opt:
4430                 return &bpf_skb_set_tunnel_opt_proto;
4431         default:
4432                 return NULL;
4433         }
4434 }
4435
4436 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4437            u32, idx)
4438 {
4439         struct bpf_array *array = container_of(map, struct bpf_array, map);
4440         struct cgroup *cgrp;
4441         struct sock *sk;
4442
4443         sk = skb_to_full_sk(skb);
4444         if (!sk || !sk_fullsock(sk))
4445                 return -ENOENT;
4446         if (unlikely(idx >= array->map.max_entries))
4447                 return -E2BIG;
4448
4449         cgrp = READ_ONCE(array->ptrs[idx]);
4450         if (unlikely(!cgrp))
4451                 return -EAGAIN;
4452
4453         return sk_under_cgroup_hierarchy(sk, cgrp);
4454 }
4455
4456 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4457         .func           = bpf_skb_under_cgroup,
4458         .gpl_only       = false,
4459         .ret_type       = RET_INTEGER,
4460         .arg1_type      = ARG_PTR_TO_CTX,
4461         .arg2_type      = ARG_CONST_MAP_PTR,
4462         .arg3_type      = ARG_ANYTHING,
4463 };
4464
4465 #ifdef CONFIG_SOCK_CGROUP_DATA
4466 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4467 {
4468         struct cgroup *cgrp;
4469
4470         sk = sk_to_full_sk(sk);
4471         if (!sk || !sk_fullsock(sk))
4472                 return 0;
4473
4474         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4475         return cgroup_id(cgrp);
4476 }
4477
4478 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4479 {
4480         return __bpf_sk_cgroup_id(skb->sk);
4481 }
4482
4483 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4484         .func           = bpf_skb_cgroup_id,
4485         .gpl_only       = false,
4486         .ret_type       = RET_INTEGER,
4487         .arg1_type      = ARG_PTR_TO_CTX,
4488 };
4489
4490 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4491                                               int ancestor_level)
4492 {
4493         struct cgroup *ancestor;
4494         struct cgroup *cgrp;
4495
4496         sk = sk_to_full_sk(sk);
4497         if (!sk || !sk_fullsock(sk))
4498                 return 0;
4499
4500         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4501         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4502         if (!ancestor)
4503                 return 0;
4504
4505         return cgroup_id(ancestor);
4506 }
4507
4508 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4509            ancestor_level)
4510 {
4511         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4512 }
4513
4514 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4515         .func           = bpf_skb_ancestor_cgroup_id,
4516         .gpl_only       = false,
4517         .ret_type       = RET_INTEGER,
4518         .arg1_type      = ARG_PTR_TO_CTX,
4519         .arg2_type      = ARG_ANYTHING,
4520 };
4521
4522 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4523 {
4524         return __bpf_sk_cgroup_id(sk);
4525 }
4526
4527 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4528         .func           = bpf_sk_cgroup_id,
4529         .gpl_only       = false,
4530         .ret_type       = RET_INTEGER,
4531         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4532 };
4533
4534 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4535 {
4536         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4537 }
4538
4539 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4540         .func           = bpf_sk_ancestor_cgroup_id,
4541         .gpl_only       = false,
4542         .ret_type       = RET_INTEGER,
4543         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4544         .arg2_type      = ARG_ANYTHING,
4545 };
4546 #endif
4547
4548 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4549                                   unsigned long off, unsigned long len)
4550 {
4551         memcpy(dst_buff, src_buff + off, len);
4552         return 0;
4553 }
4554
4555 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4556            u64, flags, void *, meta, u64, meta_size)
4557 {
4558         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4559
4560         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4561                 return -EINVAL;
4562         if (unlikely(!xdp ||
4563                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4564                 return -EFAULT;
4565
4566         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4567                                 xdp_size, bpf_xdp_copy);
4568 }
4569
4570 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4571         .func           = bpf_xdp_event_output,
4572         .gpl_only       = true,
4573         .ret_type       = RET_INTEGER,
4574         .arg1_type      = ARG_PTR_TO_CTX,
4575         .arg2_type      = ARG_CONST_MAP_PTR,
4576         .arg3_type      = ARG_ANYTHING,
4577         .arg4_type      = ARG_PTR_TO_MEM,
4578         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4579 };
4580
4581 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4582
4583 const struct bpf_func_proto bpf_xdp_output_proto = {
4584         .func           = bpf_xdp_event_output,
4585         .gpl_only       = true,
4586         .ret_type       = RET_INTEGER,
4587         .arg1_type      = ARG_PTR_TO_BTF_ID,
4588         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4589         .arg2_type      = ARG_CONST_MAP_PTR,
4590         .arg3_type      = ARG_ANYTHING,
4591         .arg4_type      = ARG_PTR_TO_MEM,
4592         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4593 };
4594
4595 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4596 {
4597         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4598 }
4599
4600 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4601         .func           = bpf_get_socket_cookie,
4602         .gpl_only       = false,
4603         .ret_type       = RET_INTEGER,
4604         .arg1_type      = ARG_PTR_TO_CTX,
4605 };
4606
4607 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4608 {
4609         return __sock_gen_cookie(ctx->sk);
4610 }
4611
4612 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4613         .func           = bpf_get_socket_cookie_sock_addr,
4614         .gpl_only       = false,
4615         .ret_type       = RET_INTEGER,
4616         .arg1_type      = ARG_PTR_TO_CTX,
4617 };
4618
4619 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4620 {
4621         return __sock_gen_cookie(ctx);
4622 }
4623
4624 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4625         .func           = bpf_get_socket_cookie_sock,
4626         .gpl_only       = false,
4627         .ret_type       = RET_INTEGER,
4628         .arg1_type      = ARG_PTR_TO_CTX,
4629 };
4630
4631 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4632 {
4633         return __sock_gen_cookie(ctx->sk);
4634 }
4635
4636 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4637         .func           = bpf_get_socket_cookie_sock_ops,
4638         .gpl_only       = false,
4639         .ret_type       = RET_INTEGER,
4640         .arg1_type      = ARG_PTR_TO_CTX,
4641 };
4642
4643 static u64 __bpf_get_netns_cookie(struct sock *sk)
4644 {
4645 #ifdef CONFIG_NET_NS
4646         return __net_gen_cookie(sk ? sk->sk_net.net : &init_net);
4647 #else
4648         return 0;
4649 #endif
4650 }
4651
4652 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4653 {
4654         return __bpf_get_netns_cookie(ctx);
4655 }
4656
4657 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4658         .func           = bpf_get_netns_cookie_sock,
4659         .gpl_only       = false,
4660         .ret_type       = RET_INTEGER,
4661         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4662 };
4663
4664 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4665 {
4666         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4667 }
4668
4669 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4670         .func           = bpf_get_netns_cookie_sock_addr,
4671         .gpl_only       = false,
4672         .ret_type       = RET_INTEGER,
4673         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4674 };
4675
4676 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4677 {
4678         struct sock *sk = sk_to_full_sk(skb->sk);
4679         kuid_t kuid;
4680
4681         if (!sk || !sk_fullsock(sk))
4682                 return overflowuid;
4683         kuid = sock_net_uid(sock_net(sk), sk);
4684         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4685 }
4686
4687 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4688         .func           = bpf_get_socket_uid,
4689         .gpl_only       = false,
4690         .ret_type       = RET_INTEGER,
4691         .arg1_type      = ARG_PTR_TO_CTX,
4692 };
4693
4694 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4695                            char *optval, int optlen)
4696 {
4697         char devname[IFNAMSIZ];
4698         int val, valbool;
4699         struct net *net;
4700         int ifindex;
4701         int ret = 0;
4702
4703         if (!sk_fullsock(sk))
4704                 return -EINVAL;
4705
4706         sock_owned_by_me(sk);
4707
4708         if (level == SOL_SOCKET) {
4709                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4710                         return -EINVAL;
4711                 val = *((int *)optval);
4712                 valbool = val ? 1 : 0;
4713
4714                 /* Only some socketops are supported */
4715                 switch (optname) {
4716                 case SO_RCVBUF:
4717                         val = min_t(u32, val, READ_ONCE(sysctl_rmem_max));
4718                         val = min_t(int, val, INT_MAX / 2);
4719                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4720                         WRITE_ONCE(sk->sk_rcvbuf,
4721                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4722                         break;
4723                 case SO_SNDBUF:
4724                         val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
4725                         val = min_t(int, val, INT_MAX / 2);
4726                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4727                         WRITE_ONCE(sk->sk_sndbuf,
4728                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4729                         break;
4730                 case SO_MAX_PACING_RATE: /* 32bit version */
4731                         if (val != ~0U)
4732                                 cmpxchg(&sk->sk_pacing_status,
4733                                         SK_PACING_NONE,
4734                                         SK_PACING_NEEDED);
4735                         sk->sk_max_pacing_rate = (val == ~0U) ?
4736                                                  ~0UL : (unsigned int)val;
4737                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4738                                                  sk->sk_max_pacing_rate);
4739                         break;
4740                 case SO_PRIORITY:
4741                         sk->sk_priority = val;
4742                         break;
4743                 case SO_RCVLOWAT:
4744                         if (val < 0)
4745                                 val = INT_MAX;
4746                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4747                         break;
4748                 case SO_MARK:
4749                         if (sk->sk_mark != val) {
4750                                 sk->sk_mark = val;
4751                                 sk_dst_reset(sk);
4752                         }
4753                         break;
4754                 case SO_BINDTODEVICE:
4755                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4756                         strncpy(devname, optval, optlen);
4757                         devname[optlen] = 0;
4758
4759                         ifindex = 0;
4760                         if (devname[0] != '\0') {
4761                                 struct net_device *dev;
4762
4763                                 ret = -ENODEV;
4764
4765                                 net = sock_net(sk);
4766                                 dev = dev_get_by_name(net, devname);
4767                                 if (!dev)
4768                                         break;
4769                                 ifindex = dev->ifindex;
4770                                 dev_put(dev);
4771                         }
4772                         ret = sock_bindtoindex(sk, ifindex, false);
4773                         break;
4774                 case SO_KEEPALIVE:
4775                         if (sk->sk_prot->keepalive)
4776                                 sk->sk_prot->keepalive(sk, valbool);
4777                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4778                         break;
4779                 default:
4780                         ret = -EINVAL;
4781                 }
4782 #ifdef CONFIG_INET
4783         } else if (level == SOL_IP) {
4784                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4785                         return -EINVAL;
4786
4787                 val = *((int *)optval);
4788                 /* Only some options are supported */
4789                 switch (optname) {
4790                 case IP_TOS:
4791                         if (val < -1 || val > 0xff) {
4792                                 ret = -EINVAL;
4793                         } else {
4794                                 struct inet_sock *inet = inet_sk(sk);
4795
4796                                 if (val == -1)
4797                                         val = 0;
4798                                 inet->tos = val;
4799                         }
4800                         break;
4801                 default:
4802                         ret = -EINVAL;
4803                 }
4804 #if IS_ENABLED(CONFIG_IPV6)
4805         } else if (level == SOL_IPV6) {
4806                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4807                         return -EINVAL;
4808
4809                 val = *((int *)optval);
4810                 /* Only some options are supported */
4811                 switch (optname) {
4812                 case IPV6_TCLASS:
4813                         if (val < -1 || val > 0xff) {
4814                                 ret = -EINVAL;
4815                         } else {
4816                                 struct ipv6_pinfo *np = inet6_sk(sk);
4817
4818                                 if (val == -1)
4819                                         val = 0;
4820                                 np->tclass = val;
4821                         }
4822                         break;
4823                 default:
4824                         ret = -EINVAL;
4825                 }
4826 #endif
4827         } else if (level == SOL_TCP &&
4828                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4829                 if (optname == TCP_CONGESTION) {
4830                         char name[TCP_CA_NAME_MAX];
4831
4832                         strncpy(name, optval, min_t(long, optlen,
4833                                                     TCP_CA_NAME_MAX-1));
4834                         name[TCP_CA_NAME_MAX-1] = 0;
4835                         ret = tcp_set_congestion_control(sk, name, false, true);
4836                 } else {
4837                         struct inet_connection_sock *icsk = inet_csk(sk);
4838                         struct tcp_sock *tp = tcp_sk(sk);
4839                         unsigned long timeout;
4840
4841                         if (optlen != sizeof(int))
4842                                 return -EINVAL;
4843
4844                         val = *((int *)optval);
4845                         /* Only some options are supported */
4846                         switch (optname) {
4847                         case TCP_BPF_IW:
4848                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4849                                         ret = -EINVAL;
4850                                 else
4851                                         tp->snd_cwnd = val;
4852                                 break;
4853                         case TCP_BPF_SNDCWND_CLAMP:
4854                                 if (val <= 0) {
4855                                         ret = -EINVAL;
4856                                 } else {
4857                                         tp->snd_cwnd_clamp = val;
4858                                         tp->snd_ssthresh = val;
4859                                 }
4860                                 break;
4861                         case TCP_BPF_DELACK_MAX:
4862                                 timeout = usecs_to_jiffies(val);
4863                                 if (timeout > TCP_DELACK_MAX ||
4864                                     timeout < TCP_TIMEOUT_MIN)
4865                                         return -EINVAL;
4866                                 inet_csk(sk)->icsk_delack_max = timeout;
4867                                 break;
4868                         case TCP_BPF_RTO_MIN:
4869                                 timeout = usecs_to_jiffies(val);
4870                                 if (timeout > TCP_RTO_MIN ||
4871                                     timeout < TCP_TIMEOUT_MIN)
4872                                         return -EINVAL;
4873                                 inet_csk(sk)->icsk_rto_min = timeout;
4874                                 break;
4875                         case TCP_SAVE_SYN:
4876                                 if (val < 0 || val > 1)
4877                                         ret = -EINVAL;
4878                                 else
4879                                         tp->save_syn = val;
4880                                 break;
4881                         case TCP_KEEPIDLE:
4882                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4883                                 break;
4884                         case TCP_KEEPINTVL:
4885                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4886                                         ret = -EINVAL;
4887                                 else
4888                                         tp->keepalive_intvl = val * HZ;
4889                                 break;
4890                         case TCP_KEEPCNT:
4891                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4892                                         ret = -EINVAL;
4893                                 else
4894                                         tp->keepalive_probes = val;
4895                                 break;
4896                         case TCP_SYNCNT:
4897                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4898                                         ret = -EINVAL;
4899                                 else
4900                                         icsk->icsk_syn_retries = val;
4901                                 break;
4902                         case TCP_USER_TIMEOUT:
4903                                 if (val < 0)
4904                                         ret = -EINVAL;
4905                                 else
4906                                         icsk->icsk_user_timeout = val;
4907                                 break;
4908                         case TCP_NOTSENT_LOWAT:
4909                                 tp->notsent_lowat = val;
4910                                 sk->sk_write_space(sk);
4911                                 break;
4912                         default:
4913                                 ret = -EINVAL;
4914                         }
4915                 }
4916 #endif
4917         } else {
4918                 ret = -EINVAL;
4919         }
4920         return ret;
4921 }
4922
4923 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4924                            char *optval, int optlen)
4925 {
4926         if (!sk_fullsock(sk))
4927                 goto err_clear;
4928
4929         sock_owned_by_me(sk);
4930
4931 #ifdef CONFIG_INET
4932         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4933                 struct inet_connection_sock *icsk;
4934                 struct tcp_sock *tp;
4935
4936                 switch (optname) {
4937                 case TCP_CONGESTION:
4938                         icsk = inet_csk(sk);
4939
4940                         if (!icsk->icsk_ca_ops || optlen <= 1)
4941                                 goto err_clear;
4942                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4943                         optval[optlen - 1] = 0;
4944                         break;
4945                 case TCP_SAVED_SYN:
4946                         tp = tcp_sk(sk);
4947
4948                         if (optlen <= 0 || !tp->saved_syn ||
4949                             optlen > tcp_saved_syn_len(tp->saved_syn))
4950                                 goto err_clear;
4951                         memcpy(optval, tp->saved_syn->data, optlen);
4952                         break;
4953                 default:
4954                         goto err_clear;
4955                 }
4956         } else if (level == SOL_IP) {
4957                 struct inet_sock *inet = inet_sk(sk);
4958
4959                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4960                         goto err_clear;
4961
4962                 /* Only some options are supported */
4963                 switch (optname) {
4964                 case IP_TOS:
4965                         *((int *)optval) = (int)inet->tos;
4966                         break;
4967                 default:
4968                         goto err_clear;
4969                 }
4970 #if IS_ENABLED(CONFIG_IPV6)
4971         } else if (level == SOL_IPV6) {
4972                 struct ipv6_pinfo *np = inet6_sk(sk);
4973
4974                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4975                         goto err_clear;
4976
4977                 /* Only some options are supported */
4978                 switch (optname) {
4979                 case IPV6_TCLASS:
4980                         *((int *)optval) = (int)np->tclass;
4981                         break;
4982                 default:
4983                         goto err_clear;
4984                 }
4985 #endif
4986         } else {
4987                 goto err_clear;
4988         }
4989         return 0;
4990 #endif
4991 err_clear:
4992         memset(optval, 0, optlen);
4993         return -EINVAL;
4994 }
4995
4996 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4997            int, level, int, optname, char *, optval, int, optlen)
4998 {
4999         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5000 }
5001
5002 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5003         .func           = bpf_sock_addr_setsockopt,
5004         .gpl_only       = false,
5005         .ret_type       = RET_INTEGER,
5006         .arg1_type      = ARG_PTR_TO_CTX,
5007         .arg2_type      = ARG_ANYTHING,
5008         .arg3_type      = ARG_ANYTHING,
5009         .arg4_type      = ARG_PTR_TO_MEM,
5010         .arg5_type      = ARG_CONST_SIZE,
5011 };
5012
5013 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5014            int, level, int, optname, char *, optval, int, optlen)
5015 {
5016         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5017 }
5018
5019 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5020         .func           = bpf_sock_addr_getsockopt,
5021         .gpl_only       = false,
5022         .ret_type       = RET_INTEGER,
5023         .arg1_type      = ARG_PTR_TO_CTX,
5024         .arg2_type      = ARG_ANYTHING,
5025         .arg3_type      = ARG_ANYTHING,
5026         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5027         .arg5_type      = ARG_CONST_SIZE,
5028 };
5029
5030 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5031            int, level, int, optname, char *, optval, int, optlen)
5032 {
5033         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5034 }
5035
5036 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5037         .func           = bpf_sock_ops_setsockopt,
5038         .gpl_only       = false,
5039         .ret_type       = RET_INTEGER,
5040         .arg1_type      = ARG_PTR_TO_CTX,
5041         .arg2_type      = ARG_ANYTHING,
5042         .arg3_type      = ARG_ANYTHING,
5043         .arg4_type      = ARG_PTR_TO_MEM,
5044         .arg5_type      = ARG_CONST_SIZE,
5045 };
5046
5047 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5048                                 int optname, const u8 **start)
5049 {
5050         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5051         const u8 *hdr_start;
5052         int ret;
5053
5054         if (syn_skb) {
5055                 /* sk is a request_sock here */
5056
5057                 if (optname == TCP_BPF_SYN) {
5058                         hdr_start = syn_skb->data;
5059                         ret = tcp_hdrlen(syn_skb);
5060                 } else if (optname == TCP_BPF_SYN_IP) {
5061                         hdr_start = skb_network_header(syn_skb);
5062                         ret = skb_network_header_len(syn_skb) +
5063                                 tcp_hdrlen(syn_skb);
5064                 } else {
5065                         /* optname == TCP_BPF_SYN_MAC */
5066                         hdr_start = skb_mac_header(syn_skb);
5067                         ret = skb_mac_header_len(syn_skb) +
5068                                 skb_network_header_len(syn_skb) +
5069                                 tcp_hdrlen(syn_skb);
5070                 }
5071         } else {
5072                 struct sock *sk = bpf_sock->sk;
5073                 struct saved_syn *saved_syn;
5074
5075                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5076                         /* synack retransmit. bpf_sock->syn_skb will
5077                          * not be available.  It has to resort to
5078                          * saved_syn (if it is saved).
5079                          */
5080                         saved_syn = inet_reqsk(sk)->saved_syn;
5081                 else
5082                         saved_syn = tcp_sk(sk)->saved_syn;
5083
5084                 if (!saved_syn)
5085                         return -ENOENT;
5086
5087                 if (optname == TCP_BPF_SYN) {
5088                         hdr_start = saved_syn->data +
5089                                 saved_syn->mac_hdrlen +
5090                                 saved_syn->network_hdrlen;
5091                         ret = saved_syn->tcp_hdrlen;
5092                 } else if (optname == TCP_BPF_SYN_IP) {
5093                         hdr_start = saved_syn->data +
5094                                 saved_syn->mac_hdrlen;
5095                         ret = saved_syn->network_hdrlen +
5096                                 saved_syn->tcp_hdrlen;
5097                 } else {
5098                         /* optname == TCP_BPF_SYN_MAC */
5099
5100                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5101                         if (!saved_syn->mac_hdrlen)
5102                                 return -ENOENT;
5103
5104                         hdr_start = saved_syn->data;
5105                         ret = saved_syn->mac_hdrlen +
5106                                 saved_syn->network_hdrlen +
5107                                 saved_syn->tcp_hdrlen;
5108                 }
5109         }
5110
5111         *start = hdr_start;
5112         return ret;
5113 }
5114
5115 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5116            int, level, int, optname, char *, optval, int, optlen)
5117 {
5118         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5119             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5120                 int ret, copy_len = 0;
5121                 const u8 *start;
5122
5123                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5124                 if (ret > 0) {
5125                         copy_len = ret;
5126                         if (optlen < copy_len) {
5127                                 copy_len = optlen;
5128                                 ret = -ENOSPC;
5129                         }
5130
5131                         memcpy(optval, start, copy_len);
5132                 }
5133
5134                 /* Zero out unused buffer at the end */
5135                 memset(optval + copy_len, 0, optlen - copy_len);
5136
5137                 return ret;
5138         }
5139
5140         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5141 }
5142
5143 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5144         .func           = bpf_sock_ops_getsockopt,
5145         .gpl_only       = false,
5146         .ret_type       = RET_INTEGER,
5147         .arg1_type      = ARG_PTR_TO_CTX,
5148         .arg2_type      = ARG_ANYTHING,
5149         .arg3_type      = ARG_ANYTHING,
5150         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5151         .arg5_type      = ARG_CONST_SIZE,
5152 };
5153
5154 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5155            int, argval)
5156 {
5157         struct sock *sk = bpf_sock->sk;
5158         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5159
5160         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5161                 return -EINVAL;
5162
5163         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5164
5165         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5166 }
5167
5168 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5169         .func           = bpf_sock_ops_cb_flags_set,
5170         .gpl_only       = false,
5171         .ret_type       = RET_INTEGER,
5172         .arg1_type      = ARG_PTR_TO_CTX,
5173         .arg2_type      = ARG_ANYTHING,
5174 };
5175
5176 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5177 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5178
5179 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5180            int, addr_len)
5181 {
5182 #ifdef CONFIG_INET
5183         struct sock *sk = ctx->sk;
5184         u32 flags = BIND_FROM_BPF;
5185         int err;
5186
5187         err = -EINVAL;
5188         if (addr_len < offsetofend(struct sockaddr, sa_family))
5189                 return err;
5190         if (addr->sa_family == AF_INET) {
5191                 if (addr_len < sizeof(struct sockaddr_in))
5192                         return err;
5193                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5194                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5195                 return __inet_bind(sk, addr, addr_len, flags);
5196 #if IS_ENABLED(CONFIG_IPV6)
5197         } else if (addr->sa_family == AF_INET6) {
5198                 if (addr_len < SIN6_LEN_RFC2133)
5199                         return err;
5200                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5201                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5202                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5203                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5204                  */
5205                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5206 #endif /* CONFIG_IPV6 */
5207         }
5208 #endif /* CONFIG_INET */
5209
5210         return -EAFNOSUPPORT;
5211 }
5212
5213 static const struct bpf_func_proto bpf_bind_proto = {
5214         .func           = bpf_bind,
5215         .gpl_only       = false,
5216         .ret_type       = RET_INTEGER,
5217         .arg1_type      = ARG_PTR_TO_CTX,
5218         .arg2_type      = ARG_PTR_TO_MEM,
5219         .arg3_type      = ARG_CONST_SIZE,
5220 };
5221
5222 #ifdef CONFIG_XFRM
5223 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5224            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5225 {
5226         const struct sec_path *sp = skb_sec_path(skb);
5227         const struct xfrm_state *x;
5228
5229         if (!sp || unlikely(index >= sp->len || flags))
5230                 goto err_clear;
5231
5232         x = sp->xvec[index];
5233
5234         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5235                 goto err_clear;
5236
5237         to->reqid = x->props.reqid;
5238         to->spi = x->id.spi;
5239         to->family = x->props.family;
5240         to->ext = 0;
5241
5242         if (to->family == AF_INET6) {
5243                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5244                        sizeof(to->remote_ipv6));
5245         } else {
5246                 to->remote_ipv4 = x->props.saddr.a4;
5247                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5248         }
5249
5250         return 0;
5251 err_clear:
5252         memset(to, 0, size);
5253         return -EINVAL;
5254 }
5255
5256 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5257         .func           = bpf_skb_get_xfrm_state,
5258         .gpl_only       = false,
5259         .ret_type       = RET_INTEGER,
5260         .arg1_type      = ARG_PTR_TO_CTX,
5261         .arg2_type      = ARG_ANYTHING,
5262         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5263         .arg4_type      = ARG_CONST_SIZE,
5264         .arg5_type      = ARG_ANYTHING,
5265 };
5266 #endif
5267
5268 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5269 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5270                                   const struct neighbour *neigh,
5271                                   const struct net_device *dev)
5272 {
5273         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5274         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5275         params->h_vlan_TCI = 0;
5276         params->h_vlan_proto = 0;
5277
5278         return 0;
5279 }
5280 #endif
5281
5282 #if IS_ENABLED(CONFIG_INET)
5283 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5284                                u32 flags, bool check_mtu)
5285 {
5286         struct fib_nh_common *nhc;
5287         struct in_device *in_dev;
5288         struct neighbour *neigh;
5289         struct net_device *dev;
5290         struct fib_result res;
5291         struct flowi4 fl4;
5292         int err;
5293         u32 mtu;
5294
5295         dev = dev_get_by_index_rcu(net, params->ifindex);
5296         if (unlikely(!dev))
5297                 return -ENODEV;
5298
5299         /* verify forwarding is enabled on this interface */
5300         in_dev = __in_dev_get_rcu(dev);
5301         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5302                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5303
5304         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5305                 fl4.flowi4_iif = 1;
5306                 fl4.flowi4_oif = params->ifindex;
5307         } else {
5308                 fl4.flowi4_iif = params->ifindex;
5309                 fl4.flowi4_oif = 0;
5310         }
5311         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5312         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5313         fl4.flowi4_flags = 0;
5314
5315         fl4.flowi4_proto = params->l4_protocol;
5316         fl4.daddr = params->ipv4_dst;
5317         fl4.saddr = params->ipv4_src;
5318         fl4.fl4_sport = params->sport;
5319         fl4.fl4_dport = params->dport;
5320         fl4.flowi4_multipath_hash = 0;
5321
5322         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5323                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5324                 struct fib_table *tb;
5325
5326                 tb = fib_get_table(net, tbid);
5327                 if (unlikely(!tb))
5328                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5329
5330                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5331         } else {
5332                 fl4.flowi4_mark = 0;
5333                 fl4.flowi4_secid = 0;
5334                 fl4.flowi4_tun_key.tun_id = 0;
5335                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5336
5337                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5338         }
5339
5340         if (err) {
5341                 /* map fib lookup errors to RTN_ type */
5342                 if (err == -EINVAL)
5343                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5344                 if (err == -EHOSTUNREACH)
5345                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5346                 if (err == -EACCES)
5347                         return BPF_FIB_LKUP_RET_PROHIBIT;
5348
5349                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5350         }
5351
5352         if (res.type != RTN_UNICAST)
5353                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5354
5355         if (fib_info_num_path(res.fi) > 1)
5356                 fib_select_path(net, &res, &fl4, NULL);
5357
5358         if (check_mtu) {
5359                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5360                 if (params->tot_len > mtu)
5361                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5362         }
5363
5364         nhc = res.nhc;
5365
5366         /* do not handle lwt encaps right now */
5367         if (nhc->nhc_lwtstate)
5368                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5369
5370         dev = nhc->nhc_dev;
5371
5372         params->rt_metric = res.fi->fib_priority;
5373         params->ifindex = dev->ifindex;
5374
5375         /* xdp and cls_bpf programs are run in RCU-bh so
5376          * rcu_read_lock_bh is not needed here
5377          */
5378         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5379                 if (nhc->nhc_gw_family)
5380                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5381
5382                 neigh = __ipv4_neigh_lookup_noref(dev,
5383                                                  (__force u32)params->ipv4_dst);
5384         } else {
5385                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5386
5387                 params->family = AF_INET6;
5388                 *dst = nhc->nhc_gw.ipv6;
5389                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5390         }
5391
5392         if (!neigh)
5393                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5394
5395         return bpf_fib_set_fwd_params(params, neigh, dev);
5396 }
5397 #endif
5398
5399 #if IS_ENABLED(CONFIG_IPV6)
5400 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5401                                u32 flags, bool check_mtu)
5402 {
5403         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5404         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5405         struct fib6_result res = {};
5406         struct neighbour *neigh;
5407         struct net_device *dev;
5408         struct inet6_dev *idev;
5409         struct flowi6 fl6;
5410         int strict = 0;
5411         int oif, err;
5412         u32 mtu;
5413
5414         /* link local addresses are never forwarded */
5415         if (rt6_need_strict(dst) || rt6_need_strict(src))
5416                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5417
5418         dev = dev_get_by_index_rcu(net, params->ifindex);
5419         if (unlikely(!dev))
5420                 return -ENODEV;
5421
5422         idev = __in6_dev_get_safely(dev);
5423         if (unlikely(!idev || !idev->cnf.forwarding))
5424                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5425
5426         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5427                 fl6.flowi6_iif = 1;
5428                 oif = fl6.flowi6_oif = params->ifindex;
5429         } else {
5430                 oif = fl6.flowi6_iif = params->ifindex;
5431                 fl6.flowi6_oif = 0;
5432                 strict = RT6_LOOKUP_F_HAS_SADDR;
5433         }
5434         fl6.flowlabel = params->flowinfo;
5435         fl6.flowi6_scope = 0;
5436         fl6.flowi6_flags = 0;
5437         fl6.mp_hash = 0;
5438
5439         fl6.flowi6_proto = params->l4_protocol;
5440         fl6.daddr = *dst;
5441         fl6.saddr = *src;
5442         fl6.fl6_sport = params->sport;
5443         fl6.fl6_dport = params->dport;
5444
5445         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5446                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5447                 struct fib6_table *tb;
5448
5449                 tb = ipv6_stub->fib6_get_table(net, tbid);
5450                 if (unlikely(!tb))
5451                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5452
5453                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5454                                                    strict);
5455         } else {
5456                 fl6.flowi6_mark = 0;
5457                 fl6.flowi6_secid = 0;
5458                 fl6.flowi6_tun_key.tun_id = 0;
5459                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5460
5461                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5462         }
5463
5464         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5465                      res.f6i == net->ipv6.fib6_null_entry))
5466                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5467
5468         switch (res.fib6_type) {
5469         /* only unicast is forwarded */
5470         case RTN_UNICAST:
5471                 break;
5472         case RTN_BLACKHOLE:
5473                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5474         case RTN_UNREACHABLE:
5475                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5476         case RTN_PROHIBIT:
5477                 return BPF_FIB_LKUP_RET_PROHIBIT;
5478         default:
5479                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5480         }
5481
5482         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5483                                     fl6.flowi6_oif != 0, NULL, strict);
5484
5485         if (check_mtu) {
5486                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5487                 if (params->tot_len > mtu)
5488                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5489         }
5490
5491         if (res.nh->fib_nh_lws)
5492                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5493
5494         if (res.nh->fib_nh_gw_family)
5495                 *dst = res.nh->fib_nh_gw6;
5496
5497         dev = res.nh->fib_nh_dev;
5498         params->rt_metric = res.f6i->fib6_metric;
5499         params->ifindex = dev->ifindex;
5500
5501         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5502          * not needed here.
5503          */
5504         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5505         if (!neigh)
5506                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5507
5508         return bpf_fib_set_fwd_params(params, neigh, dev);
5509 }
5510 #endif
5511
5512 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5513            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5514 {
5515         if (plen < sizeof(*params))
5516                 return -EINVAL;
5517
5518         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5519                 return -EINVAL;
5520
5521         switch (params->family) {
5522 #if IS_ENABLED(CONFIG_INET)
5523         case AF_INET:
5524                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5525                                            flags, true);
5526 #endif
5527 #if IS_ENABLED(CONFIG_IPV6)
5528         case AF_INET6:
5529                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5530                                            flags, true);
5531 #endif
5532         }
5533         return -EAFNOSUPPORT;
5534 }
5535
5536 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5537         .func           = bpf_xdp_fib_lookup,
5538         .gpl_only       = true,
5539         .ret_type       = RET_INTEGER,
5540         .arg1_type      = ARG_PTR_TO_CTX,
5541         .arg2_type      = ARG_PTR_TO_MEM,
5542         .arg3_type      = ARG_CONST_SIZE,
5543         .arg4_type      = ARG_ANYTHING,
5544 };
5545
5546 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5547            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5548 {
5549         struct net *net = dev_net(skb->dev);
5550         int rc = -EAFNOSUPPORT;
5551         bool check_mtu = false;
5552
5553         if (plen < sizeof(*params))
5554                 return -EINVAL;
5555
5556         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5557                 return -EINVAL;
5558
5559         if (params->tot_len)
5560                 check_mtu = true;
5561
5562         switch (params->family) {
5563 #if IS_ENABLED(CONFIG_INET)
5564         case AF_INET:
5565                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5566                 break;
5567 #endif
5568 #if IS_ENABLED(CONFIG_IPV6)
5569         case AF_INET6:
5570                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5571                 break;
5572 #endif
5573         }
5574
5575         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5576                 struct net_device *dev;
5577
5578                 /* When tot_len isn't provided by user, check skb
5579                  * against MTU of FIB lookup resulting net_device
5580                  */
5581                 dev = dev_get_by_index_rcu(net, params->ifindex);
5582                 if (!is_skb_forwardable(dev, skb))
5583                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5584         }
5585
5586         return rc;
5587 }
5588
5589 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5590         .func           = bpf_skb_fib_lookup,
5591         .gpl_only       = true,
5592         .ret_type       = RET_INTEGER,
5593         .arg1_type      = ARG_PTR_TO_CTX,
5594         .arg2_type      = ARG_PTR_TO_MEM,
5595         .arg3_type      = ARG_CONST_SIZE,
5596         .arg4_type      = ARG_ANYTHING,
5597 };
5598
5599 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5600 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5601 {
5602         int err;
5603         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5604
5605         if (!seg6_validate_srh(srh, len, false))
5606                 return -EINVAL;
5607
5608         switch (type) {
5609         case BPF_LWT_ENCAP_SEG6_INLINE:
5610                 if (skb->protocol != htons(ETH_P_IPV6))
5611                         return -EBADMSG;
5612
5613                 err = seg6_do_srh_inline(skb, srh);
5614                 break;
5615         case BPF_LWT_ENCAP_SEG6:
5616                 skb_reset_inner_headers(skb);
5617                 skb->encapsulation = 1;
5618                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5619                 break;
5620         default:
5621                 return -EINVAL;
5622         }
5623
5624         bpf_compute_data_pointers(skb);
5625         if (err)
5626                 return err;
5627
5628         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5629
5630         return seg6_lookup_nexthop(skb, NULL, 0);
5631 }
5632 #endif /* CONFIG_IPV6_SEG6_BPF */
5633
5634 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5635 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5636                              bool ingress)
5637 {
5638         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5639 }
5640 #endif
5641
5642 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5643            u32, len)
5644 {
5645         switch (type) {
5646 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5647         case BPF_LWT_ENCAP_SEG6:
5648         case BPF_LWT_ENCAP_SEG6_INLINE:
5649                 return bpf_push_seg6_encap(skb, type, hdr, len);
5650 #endif
5651 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5652         case BPF_LWT_ENCAP_IP:
5653                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5654 #endif
5655         default:
5656                 return -EINVAL;
5657         }
5658 }
5659
5660 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5661            void *, hdr, u32, len)
5662 {
5663         switch (type) {
5664 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5665         case BPF_LWT_ENCAP_IP:
5666                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5667 #endif
5668         default:
5669                 return -EINVAL;
5670         }
5671 }
5672
5673 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5674         .func           = bpf_lwt_in_push_encap,
5675         .gpl_only       = false,
5676         .ret_type       = RET_INTEGER,
5677         .arg1_type      = ARG_PTR_TO_CTX,
5678         .arg2_type      = ARG_ANYTHING,
5679         .arg3_type      = ARG_PTR_TO_MEM,
5680         .arg4_type      = ARG_CONST_SIZE
5681 };
5682
5683 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5684         .func           = bpf_lwt_xmit_push_encap,
5685         .gpl_only       = false,
5686         .ret_type       = RET_INTEGER,
5687         .arg1_type      = ARG_PTR_TO_CTX,
5688         .arg2_type      = ARG_ANYTHING,
5689         .arg3_type      = ARG_PTR_TO_MEM,
5690         .arg4_type      = ARG_CONST_SIZE
5691 };
5692
5693 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5694 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5695            const void *, from, u32, len)
5696 {
5697         struct seg6_bpf_srh_state *srh_state =
5698                 this_cpu_ptr(&seg6_bpf_srh_states);
5699         struct ipv6_sr_hdr *srh = srh_state->srh;
5700         void *srh_tlvs, *srh_end, *ptr;
5701         int srhoff = 0;
5702
5703         if (srh == NULL)
5704                 return -EINVAL;
5705
5706         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5707         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5708
5709         ptr = skb->data + offset;
5710         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5711                 srh_state->valid = false;
5712         else if (ptr < (void *)&srh->flags ||
5713                  ptr + len > (void *)&srh->segments)
5714                 return -EFAULT;
5715
5716         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5717                 return -EFAULT;
5718         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5719                 return -EINVAL;
5720         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5721
5722         memcpy(skb->data + offset, from, len);
5723         return 0;
5724 }
5725
5726 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5727         .func           = bpf_lwt_seg6_store_bytes,
5728         .gpl_only       = false,
5729         .ret_type       = RET_INTEGER,
5730         .arg1_type      = ARG_PTR_TO_CTX,
5731         .arg2_type      = ARG_ANYTHING,
5732         .arg3_type      = ARG_PTR_TO_MEM,
5733         .arg4_type      = ARG_CONST_SIZE
5734 };
5735
5736 static void bpf_update_srh_state(struct sk_buff *skb)
5737 {
5738         struct seg6_bpf_srh_state *srh_state =
5739                 this_cpu_ptr(&seg6_bpf_srh_states);
5740         int srhoff = 0;
5741
5742         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5743                 srh_state->srh = NULL;
5744         } else {
5745                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5746                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5747                 srh_state->valid = true;
5748         }
5749 }
5750
5751 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5752            u32, action, void *, param, u32, param_len)
5753 {
5754         struct seg6_bpf_srh_state *srh_state =
5755                 this_cpu_ptr(&seg6_bpf_srh_states);
5756         int hdroff = 0;
5757         int err;
5758
5759         switch (action) {
5760         case SEG6_LOCAL_ACTION_END_X:
5761                 if (!seg6_bpf_has_valid_srh(skb))
5762                         return -EBADMSG;
5763                 if (param_len != sizeof(struct in6_addr))
5764                         return -EINVAL;
5765                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5766         case SEG6_LOCAL_ACTION_END_T:
5767                 if (!seg6_bpf_has_valid_srh(skb))
5768                         return -EBADMSG;
5769                 if (param_len != sizeof(int))
5770                         return -EINVAL;
5771                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5772         case SEG6_LOCAL_ACTION_END_DT6:
5773                 if (!seg6_bpf_has_valid_srh(skb))
5774                         return -EBADMSG;
5775                 if (param_len != sizeof(int))
5776                         return -EINVAL;
5777
5778                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5779                         return -EBADMSG;
5780                 if (!pskb_pull(skb, hdroff))
5781                         return -EBADMSG;
5782
5783                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5784                 skb_reset_network_header(skb);
5785                 skb_reset_transport_header(skb);
5786                 skb->encapsulation = 0;
5787
5788                 bpf_compute_data_pointers(skb);
5789                 bpf_update_srh_state(skb);
5790                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5791         case SEG6_LOCAL_ACTION_END_B6:
5792                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5793                         return -EBADMSG;
5794                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5795                                           param, param_len);
5796                 if (!err)
5797                         bpf_update_srh_state(skb);
5798
5799                 return err;
5800         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5801                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5802                         return -EBADMSG;
5803                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5804                                           param, param_len);
5805                 if (!err)
5806                         bpf_update_srh_state(skb);
5807
5808                 return err;
5809         default:
5810                 return -EINVAL;
5811         }
5812 }
5813
5814 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5815         .func           = bpf_lwt_seg6_action,
5816         .gpl_only       = false,
5817         .ret_type       = RET_INTEGER,
5818         .arg1_type      = ARG_PTR_TO_CTX,
5819         .arg2_type      = ARG_ANYTHING,
5820         .arg3_type      = ARG_PTR_TO_MEM,
5821         .arg4_type      = ARG_CONST_SIZE
5822 };
5823
5824 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5825            s32, len)
5826 {
5827         struct seg6_bpf_srh_state *srh_state =
5828                 this_cpu_ptr(&seg6_bpf_srh_states);
5829         struct ipv6_sr_hdr *srh = srh_state->srh;
5830         void *srh_end, *srh_tlvs, *ptr;
5831         struct ipv6hdr *hdr;
5832         int srhoff = 0;
5833         int ret;
5834
5835         if (unlikely(srh == NULL))
5836                 return -EINVAL;
5837
5838         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5839                         ((srh->first_segment + 1) << 4));
5840         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5841                         srh_state->hdrlen);
5842         ptr = skb->data + offset;
5843
5844         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5845                 return -EFAULT;
5846         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5847                 return -EFAULT;
5848
5849         if (len > 0) {
5850                 ret = skb_cow_head(skb, len);
5851                 if (unlikely(ret < 0))
5852                         return ret;
5853
5854                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5855         } else {
5856                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5857         }
5858
5859         bpf_compute_data_pointers(skb);
5860         if (unlikely(ret < 0))
5861                 return ret;
5862
5863         hdr = (struct ipv6hdr *)skb->data;
5864         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5865
5866         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5867                 return -EINVAL;
5868         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5869         srh_state->hdrlen += len;
5870         srh_state->valid = false;
5871         return 0;
5872 }
5873
5874 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5875         .func           = bpf_lwt_seg6_adjust_srh,
5876         .gpl_only       = false,
5877         .ret_type       = RET_INTEGER,
5878         .arg1_type      = ARG_PTR_TO_CTX,
5879         .arg2_type      = ARG_ANYTHING,
5880         .arg3_type      = ARG_ANYTHING,
5881 };
5882 #endif /* CONFIG_IPV6_SEG6_BPF */
5883
5884 #ifdef CONFIG_INET
5885 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5886                               int dif, int sdif, u8 family, u8 proto)
5887 {
5888         bool refcounted = false;
5889         struct sock *sk = NULL;
5890
5891         if (family == AF_INET) {
5892                 __be32 src4 = tuple->ipv4.saddr;
5893                 __be32 dst4 = tuple->ipv4.daddr;
5894
5895                 if (proto == IPPROTO_TCP)
5896                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
5897                                            src4, tuple->ipv4.sport,
5898                                            dst4, tuple->ipv4.dport,
5899                                            dif, sdif, &refcounted);
5900                 else
5901                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
5902                                                dst4, tuple->ipv4.dport,
5903                                                dif, sdif, &udp_table, NULL);
5904 #if IS_ENABLED(CONFIG_IPV6)
5905         } else {
5906                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
5907                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
5908
5909                 if (proto == IPPROTO_TCP)
5910                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
5911                                             src6, tuple->ipv6.sport,
5912                                             dst6, ntohs(tuple->ipv6.dport),
5913                                             dif, sdif, &refcounted);
5914                 else if (likely(ipv6_bpf_stub))
5915                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
5916                                                             src6, tuple->ipv6.sport,
5917                                                             dst6, tuple->ipv6.dport,
5918                                                             dif, sdif,
5919                                                             &udp_table, NULL);
5920 #endif
5921         }
5922
5923         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
5924                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5925                 sk = NULL;
5926         }
5927         return sk;
5928 }
5929
5930 /* bpf_skc_lookup performs the core lookup for different types of sockets,
5931  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
5932  * Returns the socket as an 'unsigned long' to simplify the casting in the
5933  * callers to satisfy BPF_CALL declarations.
5934  */
5935 static struct sock *
5936 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5937                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5938                  u64 flags)
5939 {
5940         struct sock *sk = NULL;
5941         u8 family = AF_UNSPEC;
5942         struct net *net;
5943         int sdif;
5944
5945         if (len == sizeof(tuple->ipv4))
5946                 family = AF_INET;
5947         else if (len == sizeof(tuple->ipv6))
5948                 family = AF_INET6;
5949         else
5950                 return NULL;
5951
5952         if (unlikely(family == AF_UNSPEC || flags ||
5953                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
5954                 goto out;
5955
5956         if (family == AF_INET)
5957                 sdif = inet_sdif(skb);
5958         else
5959                 sdif = inet6_sdif(skb);
5960
5961         if ((s32)netns_id < 0) {
5962                 net = caller_net;
5963                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5964         } else {
5965                 net = get_net_ns_by_id(caller_net, netns_id);
5966                 if (unlikely(!net))
5967                         goto out;
5968                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
5969                 put_net(net);
5970         }
5971
5972 out:
5973         return sk;
5974 }
5975
5976 static struct sock *
5977 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
5978                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
5979                 u64 flags)
5980 {
5981         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
5982                                            ifindex, proto, netns_id, flags);
5983
5984         if (sk) {
5985                 struct sock *sk2 = sk_to_full_sk(sk);
5986
5987                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
5988                  * sock refcnt is decremented to prevent a request_sock leak.
5989                  */
5990                 if (!sk_fullsock(sk2))
5991                         sk2 = NULL;
5992                 if (sk2 != sk) {
5993                         sock_gen_put(sk);
5994                         /* Ensure there is no need to bump sk2 refcnt */
5995                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
5996                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
5997                                 return NULL;
5998                         }
5999                         sk = sk2;
6000                 }
6001         }
6002
6003         return sk;
6004 }
6005
6006 static struct sock *
6007 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6008                u8 proto, u64 netns_id, u64 flags)
6009 {
6010         struct net *caller_net;
6011         int ifindex;
6012
6013         if (skb->dev) {
6014                 caller_net = dev_net(skb->dev);
6015                 ifindex = skb->dev->ifindex;
6016         } else {
6017                 caller_net = sock_net(skb->sk);
6018                 ifindex = 0;
6019         }
6020
6021         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6022                                 netns_id, flags);
6023 }
6024
6025 static struct sock *
6026 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6027               u8 proto, u64 netns_id, u64 flags)
6028 {
6029         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6030                                          flags);
6031
6032         if (sk) {
6033                 struct sock *sk2 = sk_to_full_sk(sk);
6034
6035                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6036                  * sock refcnt is decremented to prevent a request_sock leak.
6037                  */
6038                 if (!sk_fullsock(sk2))
6039                         sk2 = NULL;
6040                 if (sk2 != sk) {
6041                         sock_gen_put(sk);
6042                         /* Ensure there is no need to bump sk2 refcnt */
6043                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6044                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6045                                 return NULL;
6046                         }
6047                         sk = sk2;
6048                 }
6049         }
6050
6051         return sk;
6052 }
6053
6054 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6055            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6056 {
6057         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6058                                              netns_id, flags);
6059 }
6060
6061 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6062         .func           = bpf_skc_lookup_tcp,
6063         .gpl_only       = false,
6064         .pkt_access     = true,
6065         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6066         .arg1_type      = ARG_PTR_TO_CTX,
6067         .arg2_type      = ARG_PTR_TO_MEM,
6068         .arg3_type      = ARG_CONST_SIZE,
6069         .arg4_type      = ARG_ANYTHING,
6070         .arg5_type      = ARG_ANYTHING,
6071 };
6072
6073 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6074            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6075 {
6076         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6077                                             netns_id, flags);
6078 }
6079
6080 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6081         .func           = bpf_sk_lookup_tcp,
6082         .gpl_only       = false,
6083         .pkt_access     = true,
6084         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6085         .arg1_type      = ARG_PTR_TO_CTX,
6086         .arg2_type      = ARG_PTR_TO_MEM,
6087         .arg3_type      = ARG_CONST_SIZE,
6088         .arg4_type      = ARG_ANYTHING,
6089         .arg5_type      = ARG_ANYTHING,
6090 };
6091
6092 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6093            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6094 {
6095         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6096                                             netns_id, flags);
6097 }
6098
6099 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6100         .func           = bpf_sk_lookup_udp,
6101         .gpl_only       = false,
6102         .pkt_access     = true,
6103         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6104         .arg1_type      = ARG_PTR_TO_CTX,
6105         .arg2_type      = ARG_PTR_TO_MEM,
6106         .arg3_type      = ARG_CONST_SIZE,
6107         .arg4_type      = ARG_ANYTHING,
6108         .arg5_type      = ARG_ANYTHING,
6109 };
6110
6111 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6112 {
6113         if (sk && sk_is_refcounted(sk))
6114                 sock_gen_put(sk);
6115         return 0;
6116 }
6117
6118 static const struct bpf_func_proto bpf_sk_release_proto = {
6119         .func           = bpf_sk_release,
6120         .gpl_only       = false,
6121         .ret_type       = RET_INTEGER,
6122         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6123 };
6124
6125 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6126            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6127 {
6128         struct net *caller_net = dev_net(ctx->rxq->dev);
6129         int ifindex = ctx->rxq->dev->ifindex;
6130
6131         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6132                                               ifindex, IPPROTO_UDP, netns_id,
6133                                               flags);
6134 }
6135
6136 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6137         .func           = bpf_xdp_sk_lookup_udp,
6138         .gpl_only       = false,
6139         .pkt_access     = true,
6140         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6141         .arg1_type      = ARG_PTR_TO_CTX,
6142         .arg2_type      = ARG_PTR_TO_MEM,
6143         .arg3_type      = ARG_CONST_SIZE,
6144         .arg4_type      = ARG_ANYTHING,
6145         .arg5_type      = ARG_ANYTHING,
6146 };
6147
6148 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6149            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6150 {
6151         struct net *caller_net = dev_net(ctx->rxq->dev);
6152         int ifindex = ctx->rxq->dev->ifindex;
6153
6154         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6155                                                ifindex, IPPROTO_TCP, netns_id,
6156                                                flags);
6157 }
6158
6159 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6160         .func           = bpf_xdp_skc_lookup_tcp,
6161         .gpl_only       = false,
6162         .pkt_access     = true,
6163         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6164         .arg1_type      = ARG_PTR_TO_CTX,
6165         .arg2_type      = ARG_PTR_TO_MEM,
6166         .arg3_type      = ARG_CONST_SIZE,
6167         .arg4_type      = ARG_ANYTHING,
6168         .arg5_type      = ARG_ANYTHING,
6169 };
6170
6171 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6172            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6173 {
6174         struct net *caller_net = dev_net(ctx->rxq->dev);
6175         int ifindex = ctx->rxq->dev->ifindex;
6176
6177         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6178                                               ifindex, IPPROTO_TCP, netns_id,
6179                                               flags);
6180 }
6181
6182 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6183         .func           = bpf_xdp_sk_lookup_tcp,
6184         .gpl_only       = false,
6185         .pkt_access     = true,
6186         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6187         .arg1_type      = ARG_PTR_TO_CTX,
6188         .arg2_type      = ARG_PTR_TO_MEM,
6189         .arg3_type      = ARG_CONST_SIZE,
6190         .arg4_type      = ARG_ANYTHING,
6191         .arg5_type      = ARG_ANYTHING,
6192 };
6193
6194 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6195            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6196 {
6197         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6198                                                sock_net(ctx->sk), 0,
6199                                                IPPROTO_TCP, netns_id, flags);
6200 }
6201
6202 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6203         .func           = bpf_sock_addr_skc_lookup_tcp,
6204         .gpl_only       = false,
6205         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6206         .arg1_type      = ARG_PTR_TO_CTX,
6207         .arg2_type      = ARG_PTR_TO_MEM,
6208         .arg3_type      = ARG_CONST_SIZE,
6209         .arg4_type      = ARG_ANYTHING,
6210         .arg5_type      = ARG_ANYTHING,
6211 };
6212
6213 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6214            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6215 {
6216         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6217                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6218                                               netns_id, flags);
6219 }
6220
6221 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6222         .func           = bpf_sock_addr_sk_lookup_tcp,
6223         .gpl_only       = false,
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_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6233            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6234 {
6235         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6236                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6237                                               netns_id, flags);
6238 }
6239
6240 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6241         .func           = bpf_sock_addr_sk_lookup_udp,
6242         .gpl_only       = false,
6243         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6244         .arg1_type      = ARG_PTR_TO_CTX,
6245         .arg2_type      = ARG_PTR_TO_MEM,
6246         .arg3_type      = ARG_CONST_SIZE,
6247         .arg4_type      = ARG_ANYTHING,
6248         .arg5_type      = ARG_ANYTHING,
6249 };
6250
6251 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6252                                   struct bpf_insn_access_aux *info)
6253 {
6254         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6255                                           icsk_retransmits))
6256                 return false;
6257
6258         if (off % size != 0)
6259                 return false;
6260
6261         switch (off) {
6262         case offsetof(struct bpf_tcp_sock, bytes_received):
6263         case offsetof(struct bpf_tcp_sock, bytes_acked):
6264                 return size == sizeof(__u64);
6265         default:
6266                 return size == sizeof(__u32);
6267         }
6268 }
6269
6270 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6271                                     const struct bpf_insn *si,
6272                                     struct bpf_insn *insn_buf,
6273                                     struct bpf_prog *prog, u32 *target_size)
6274 {
6275         struct bpf_insn *insn = insn_buf;
6276
6277 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6278         do {                                                            \
6279                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6280                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6281                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6282                                       si->dst_reg, si->src_reg,         \
6283                                       offsetof(struct tcp_sock, FIELD)); \
6284         } while (0)
6285
6286 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6287         do {                                                            \
6288                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6289                                           FIELD) >                      \
6290                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6291                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6292                                         struct inet_connection_sock,    \
6293                                         FIELD),                         \
6294                                       si->dst_reg, si->src_reg,         \
6295                                       offsetof(                         \
6296                                         struct inet_connection_sock,    \
6297                                         FIELD));                        \
6298         } while (0)
6299
6300         if (insn > insn_buf)
6301                 return insn - insn_buf;
6302
6303         switch (si->off) {
6304         case offsetof(struct bpf_tcp_sock, rtt_min):
6305                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6306                              sizeof(struct minmax));
6307                 BUILD_BUG_ON(sizeof(struct minmax) <
6308                              sizeof(struct minmax_sample));
6309
6310                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6311                                       offsetof(struct tcp_sock, rtt_min) +
6312                                       offsetof(struct minmax_sample, v));
6313                 break;
6314         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6315                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6316                 break;
6317         case offsetof(struct bpf_tcp_sock, srtt_us):
6318                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6319                 break;
6320         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6321                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6322                 break;
6323         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6324                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6325                 break;
6326         case offsetof(struct bpf_tcp_sock, snd_nxt):
6327                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6328                 break;
6329         case offsetof(struct bpf_tcp_sock, snd_una):
6330                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6331                 break;
6332         case offsetof(struct bpf_tcp_sock, mss_cache):
6333                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6334                 break;
6335         case offsetof(struct bpf_tcp_sock, ecn_flags):
6336                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6337                 break;
6338         case offsetof(struct bpf_tcp_sock, rate_delivered):
6339                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6340                 break;
6341         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6342                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6343                 break;
6344         case offsetof(struct bpf_tcp_sock, packets_out):
6345                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6346                 break;
6347         case offsetof(struct bpf_tcp_sock, retrans_out):
6348                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6349                 break;
6350         case offsetof(struct bpf_tcp_sock, total_retrans):
6351                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6352                 break;
6353         case offsetof(struct bpf_tcp_sock, segs_in):
6354                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6355                 break;
6356         case offsetof(struct bpf_tcp_sock, data_segs_in):
6357                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6358                 break;
6359         case offsetof(struct bpf_tcp_sock, segs_out):
6360                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6361                 break;
6362         case offsetof(struct bpf_tcp_sock, data_segs_out):
6363                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6364                 break;
6365         case offsetof(struct bpf_tcp_sock, lost_out):
6366                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6367                 break;
6368         case offsetof(struct bpf_tcp_sock, sacked_out):
6369                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6370                 break;
6371         case offsetof(struct bpf_tcp_sock, bytes_received):
6372                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6373                 break;
6374         case offsetof(struct bpf_tcp_sock, bytes_acked):
6375                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6376                 break;
6377         case offsetof(struct bpf_tcp_sock, dsack_dups):
6378                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6379                 break;
6380         case offsetof(struct bpf_tcp_sock, delivered):
6381                 BPF_TCP_SOCK_GET_COMMON(delivered);
6382                 break;
6383         case offsetof(struct bpf_tcp_sock, delivered_ce):
6384                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6385                 break;
6386         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6387                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6388                 break;
6389         }
6390
6391         return insn - insn_buf;
6392 }
6393
6394 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6395 {
6396         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6397                 return (unsigned long)sk;
6398
6399         return (unsigned long)NULL;
6400 }
6401
6402 const struct bpf_func_proto bpf_tcp_sock_proto = {
6403         .func           = bpf_tcp_sock,
6404         .gpl_only       = false,
6405         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6406         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6407 };
6408
6409 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6410 {
6411         sk = sk_to_full_sk(sk);
6412
6413         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6414                 return (unsigned long)sk;
6415
6416         return (unsigned long)NULL;
6417 }
6418
6419 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6420         .func           = bpf_get_listener_sock,
6421         .gpl_only       = false,
6422         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6423         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6424 };
6425
6426 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6427 {
6428         unsigned int iphdr_len;
6429
6430         switch (skb_protocol(skb, true)) {
6431         case cpu_to_be16(ETH_P_IP):
6432                 iphdr_len = sizeof(struct iphdr);
6433                 break;
6434         case cpu_to_be16(ETH_P_IPV6):
6435                 iphdr_len = sizeof(struct ipv6hdr);
6436                 break;
6437         default:
6438                 return 0;
6439         }
6440
6441         if (skb_headlen(skb) < iphdr_len)
6442                 return 0;
6443
6444         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6445                 return 0;
6446
6447         return INET_ECN_set_ce(skb);
6448 }
6449
6450 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6451                                   struct bpf_insn_access_aux *info)
6452 {
6453         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6454                 return false;
6455
6456         if (off % size != 0)
6457                 return false;
6458
6459         switch (off) {
6460         default:
6461                 return size == sizeof(__u32);
6462         }
6463 }
6464
6465 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6466                                     const struct bpf_insn *si,
6467                                     struct bpf_insn *insn_buf,
6468                                     struct bpf_prog *prog, u32 *target_size)
6469 {
6470         struct bpf_insn *insn = insn_buf;
6471
6472 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6473         do {                                                            \
6474                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6475                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6476                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6477                                       si->dst_reg, si->src_reg,         \
6478                                       offsetof(struct xdp_sock, FIELD)); \
6479         } while (0)
6480
6481         switch (si->off) {
6482         case offsetof(struct bpf_xdp_sock, queue_id):
6483                 BPF_XDP_SOCK_GET(queue_id);
6484                 break;
6485         }
6486
6487         return insn - insn_buf;
6488 }
6489
6490 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6491         .func           = bpf_skb_ecn_set_ce,
6492         .gpl_only       = false,
6493         .ret_type       = RET_INTEGER,
6494         .arg1_type      = ARG_PTR_TO_CTX,
6495 };
6496
6497 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6498            struct tcphdr *, th, u32, th_len)
6499 {
6500 #ifdef CONFIG_SYN_COOKIES
6501         u32 cookie;
6502         int ret;
6503
6504         if (unlikely(!sk || th_len < sizeof(*th)))
6505                 return -EINVAL;
6506
6507         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6508         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6509                 return -EINVAL;
6510
6511         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6512                 return -EINVAL;
6513
6514         if (!th->ack || th->rst || th->syn)
6515                 return -ENOENT;
6516
6517         if (unlikely(iph_len < sizeof(struct iphdr)))
6518                 return -EINVAL;
6519
6520         if (tcp_synq_no_recent_overflow(sk))
6521                 return -ENOENT;
6522
6523         cookie = ntohl(th->ack_seq) - 1;
6524
6525         /* Both struct iphdr and struct ipv6hdr have the version field at the
6526          * same offset so we can cast to the shorter header (struct iphdr).
6527          */
6528         switch (((struct iphdr *)iph)->version) {
6529         case 4:
6530                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
6531                         return -EINVAL;
6532
6533                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6534                 break;
6535
6536 #if IS_BUILTIN(CONFIG_IPV6)
6537         case 6:
6538                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6539                         return -EINVAL;
6540
6541                 if (sk->sk_family != AF_INET6)
6542                         return -EINVAL;
6543
6544                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6545                 break;
6546 #endif /* CONFIG_IPV6 */
6547
6548         default:
6549                 return -EPROTONOSUPPORT;
6550         }
6551
6552         if (ret > 0)
6553                 return 0;
6554
6555         return -ENOENT;
6556 #else
6557         return -ENOTSUPP;
6558 #endif
6559 }
6560
6561 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6562         .func           = bpf_tcp_check_syncookie,
6563         .gpl_only       = true,
6564         .pkt_access     = true,
6565         .ret_type       = RET_INTEGER,
6566         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6567         .arg2_type      = ARG_PTR_TO_MEM,
6568         .arg3_type      = ARG_CONST_SIZE,
6569         .arg4_type      = ARG_PTR_TO_MEM,
6570         .arg5_type      = ARG_CONST_SIZE,
6571 };
6572
6573 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6574            struct tcphdr *, th, u32, th_len)
6575 {
6576 #ifdef CONFIG_SYN_COOKIES
6577         u32 cookie;
6578         u16 mss;
6579
6580         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6581                 return -EINVAL;
6582
6583         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6584                 return -EINVAL;
6585
6586         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6587                 return -ENOENT;
6588
6589         if (!th->syn || th->ack || th->fin || th->rst)
6590                 return -EINVAL;
6591
6592         if (unlikely(iph_len < sizeof(struct iphdr)))
6593                 return -EINVAL;
6594
6595         /* Both struct iphdr and struct ipv6hdr have the version field at the
6596          * same offset so we can cast to the shorter header (struct iphdr).
6597          */
6598         switch (((struct iphdr *)iph)->version) {
6599         case 4:
6600                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6601                         return -EINVAL;
6602
6603                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6604                 break;
6605
6606 #if IS_BUILTIN(CONFIG_IPV6)
6607         case 6:
6608                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6609                         return -EINVAL;
6610
6611                 if (sk->sk_family != AF_INET6)
6612                         return -EINVAL;
6613
6614                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6615                 break;
6616 #endif /* CONFIG_IPV6 */
6617
6618         default:
6619                 return -EPROTONOSUPPORT;
6620         }
6621         if (mss == 0)
6622                 return -ENOENT;
6623
6624         return cookie | ((u64)mss << 32);
6625 #else
6626         return -EOPNOTSUPP;
6627 #endif /* CONFIG_SYN_COOKIES */
6628 }
6629
6630 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6631         .func           = bpf_tcp_gen_syncookie,
6632         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6633         .pkt_access     = true,
6634         .ret_type       = RET_INTEGER,
6635         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6636         .arg2_type      = ARG_PTR_TO_MEM,
6637         .arg3_type      = ARG_CONST_SIZE,
6638         .arg4_type      = ARG_PTR_TO_MEM,
6639         .arg5_type      = ARG_CONST_SIZE,
6640 };
6641
6642 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6643 {
6644         if (!sk || flags != 0)
6645                 return -EINVAL;
6646         if (!skb_at_tc_ingress(skb))
6647                 return -EOPNOTSUPP;
6648         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6649                 return -ENETUNREACH;
6650         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6651                 return -ESOCKTNOSUPPORT;
6652         if (sk_is_refcounted(sk) &&
6653             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6654                 return -ENOENT;
6655
6656         skb_orphan(skb);
6657         skb->sk = sk;
6658         skb->destructor = sock_pfree;
6659
6660         return 0;
6661 }
6662
6663 static const struct bpf_func_proto bpf_sk_assign_proto = {
6664         .func           = bpf_sk_assign,
6665         .gpl_only       = false,
6666         .ret_type       = RET_INTEGER,
6667         .arg1_type      = ARG_PTR_TO_CTX,
6668         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6669         .arg3_type      = ARG_ANYTHING,
6670 };
6671
6672 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6673                                     u8 search_kind, const u8 *magic,
6674                                     u8 magic_len, bool *eol)
6675 {
6676         u8 kind, kind_len;
6677
6678         *eol = false;
6679
6680         while (op < opend) {
6681                 kind = op[0];
6682
6683                 if (kind == TCPOPT_EOL) {
6684                         *eol = true;
6685                         return ERR_PTR(-ENOMSG);
6686                 } else if (kind == TCPOPT_NOP) {
6687                         op++;
6688                         continue;
6689                 }
6690
6691                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6692                         /* Something is wrong in the received header.
6693                          * Follow the TCP stack's tcp_parse_options()
6694                          * and just bail here.
6695                          */
6696                         return ERR_PTR(-EFAULT);
6697
6698                 kind_len = op[1];
6699                 if (search_kind == kind) {
6700                         if (!magic_len)
6701                                 return op;
6702
6703                         if (magic_len > kind_len - 2)
6704                                 return ERR_PTR(-ENOMSG);
6705
6706                         if (!memcmp(&op[2], magic, magic_len))
6707                                 return op;
6708                 }
6709
6710                 op += kind_len;
6711         }
6712
6713         return ERR_PTR(-ENOMSG);
6714 }
6715
6716 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6717            void *, search_res, u32, len, u64, flags)
6718 {
6719         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6720         const u8 *op, *opend, *magic, *search = search_res;
6721         u8 search_kind, search_len, copy_len, magic_len;
6722         int ret;
6723
6724         /* 2 byte is the minimal option len except TCPOPT_NOP and
6725          * TCPOPT_EOL which are useless for the bpf prog to learn
6726          * and this helper disallow loading them also.
6727          */
6728         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6729                 return -EINVAL;
6730
6731         search_kind = search[0];
6732         search_len = search[1];
6733
6734         if (search_len > len || search_kind == TCPOPT_NOP ||
6735             search_kind == TCPOPT_EOL)
6736                 return -EINVAL;
6737
6738         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6739                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6740                 if (search_len != 4 && search_len != 6)
6741                         return -EINVAL;
6742                 magic = &search[2];
6743                 magic_len = search_len - 2;
6744         } else {
6745                 if (search_len)
6746                         return -EINVAL;
6747                 magic = NULL;
6748                 magic_len = 0;
6749         }
6750
6751         if (load_syn) {
6752                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6753                 if (ret < 0)
6754                         return ret;
6755
6756                 opend = op + ret;
6757                 op += sizeof(struct tcphdr);
6758         } else {
6759                 if (!bpf_sock->skb ||
6760                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6761                         /* This bpf_sock->op cannot call this helper */
6762                         return -EPERM;
6763
6764                 opend = bpf_sock->skb_data_end;
6765                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6766         }
6767
6768         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6769                                 &eol);
6770         if (IS_ERR(op))
6771                 return PTR_ERR(op);
6772
6773         copy_len = op[1];
6774         ret = copy_len;
6775         if (copy_len > len) {
6776                 ret = -ENOSPC;
6777                 copy_len = len;
6778         }
6779
6780         memcpy(search_res, op, copy_len);
6781         return ret;
6782 }
6783
6784 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6785         .func           = bpf_sock_ops_load_hdr_opt,
6786         .gpl_only       = false,
6787         .ret_type       = RET_INTEGER,
6788         .arg1_type      = ARG_PTR_TO_CTX,
6789         .arg2_type      = ARG_PTR_TO_MEM,
6790         .arg3_type      = ARG_CONST_SIZE,
6791         .arg4_type      = ARG_ANYTHING,
6792 };
6793
6794 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6795            const void *, from, u32, len, u64, flags)
6796 {
6797         u8 new_kind, new_kind_len, magic_len = 0, *opend;
6798         const u8 *op, *new_op, *magic = NULL;
6799         struct sk_buff *skb;
6800         bool eol;
6801
6802         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6803                 return -EPERM;
6804
6805         if (len < 2 || flags)
6806                 return -EINVAL;
6807
6808         new_op = from;
6809         new_kind = new_op[0];
6810         new_kind_len = new_op[1];
6811
6812         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6813             new_kind == TCPOPT_EOL)
6814                 return -EINVAL;
6815
6816         if (new_kind_len > bpf_sock->remaining_opt_len)
6817                 return -ENOSPC;
6818
6819         /* 253 is another experimental kind */
6820         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
6821                 if (new_kind_len < 4)
6822                         return -EINVAL;
6823                 /* Match for the 2 byte magic also.
6824                  * RFC 6994: the magic could be 2 or 4 bytes.
6825                  * Hence, matching by 2 byte only is on the
6826                  * conservative side but it is the right
6827                  * thing to do for the 'search-for-duplication'
6828                  * purpose.
6829                  */
6830                 magic = &new_op[2];
6831                 magic_len = 2;
6832         }
6833
6834         /* Check for duplication */
6835         skb = bpf_sock->skb;
6836         op = skb->data + sizeof(struct tcphdr);
6837         opend = bpf_sock->skb_data_end;
6838
6839         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6840                                 &eol);
6841         if (!IS_ERR(op))
6842                 return -EEXIST;
6843
6844         if (PTR_ERR(op) != -ENOMSG)
6845                 return PTR_ERR(op);
6846
6847         if (eol)
6848                 /* The option has been ended.  Treat it as no more
6849                  * header option can be written.
6850                  */
6851                 return -ENOSPC;
6852
6853         /* No duplication found.  Store the header option. */
6854         memcpy(opend, from, new_kind_len);
6855
6856         bpf_sock->remaining_opt_len -= new_kind_len;
6857         bpf_sock->skb_data_end += new_kind_len;
6858
6859         return 0;
6860 }
6861
6862 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6863         .func           = bpf_sock_ops_store_hdr_opt,
6864         .gpl_only       = false,
6865         .ret_type       = RET_INTEGER,
6866         .arg1_type      = ARG_PTR_TO_CTX,
6867         .arg2_type      = ARG_PTR_TO_MEM,
6868         .arg3_type      = ARG_CONST_SIZE,
6869         .arg4_type      = ARG_ANYTHING,
6870 };
6871
6872 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6873            u32, len, u64, flags)
6874 {
6875         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6876                 return -EPERM;
6877
6878         if (flags || len < 2)
6879                 return -EINVAL;
6880
6881         if (len > bpf_sock->remaining_opt_len)
6882                 return -ENOSPC;
6883
6884         bpf_sock->remaining_opt_len -= len;
6885
6886         return 0;
6887 }
6888
6889 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6890         .func           = bpf_sock_ops_reserve_hdr_opt,
6891         .gpl_only       = false,
6892         .ret_type       = RET_INTEGER,
6893         .arg1_type      = ARG_PTR_TO_CTX,
6894         .arg2_type      = ARG_ANYTHING,
6895         .arg3_type      = ARG_ANYTHING,
6896 };
6897
6898 #endif /* CONFIG_INET */
6899
6900 bool bpf_helper_changes_pkt_data(void *func)
6901 {
6902         if (func == bpf_skb_vlan_push ||
6903             func == bpf_skb_vlan_pop ||
6904             func == bpf_skb_store_bytes ||
6905             func == bpf_skb_change_proto ||
6906             func == bpf_skb_change_head ||
6907             func == sk_skb_change_head ||
6908             func == bpf_skb_change_tail ||
6909             func == sk_skb_change_tail ||
6910             func == bpf_skb_adjust_room ||
6911             func == sk_skb_adjust_room ||
6912             func == bpf_skb_pull_data ||
6913             func == sk_skb_pull_data ||
6914             func == bpf_clone_redirect ||
6915             func == bpf_l3_csum_replace ||
6916             func == bpf_l4_csum_replace ||
6917             func == bpf_xdp_adjust_head ||
6918             func == bpf_xdp_adjust_meta ||
6919             func == bpf_msg_pull_data ||
6920             func == bpf_msg_push_data ||
6921             func == bpf_msg_pop_data ||
6922             func == bpf_xdp_adjust_tail ||
6923 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6924             func == bpf_lwt_seg6_store_bytes ||
6925             func == bpf_lwt_seg6_adjust_srh ||
6926             func == bpf_lwt_seg6_action ||
6927 #endif
6928 #ifdef CONFIG_INET
6929             func == bpf_sock_ops_store_hdr_opt ||
6930 #endif
6931             func == bpf_lwt_in_push_encap ||
6932             func == bpf_lwt_xmit_push_encap)
6933                 return true;
6934
6935         return false;
6936 }
6937
6938 const struct bpf_func_proto bpf_event_output_data_proto __weak;
6939 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
6940
6941 static const struct bpf_func_proto *
6942 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6943 {
6944         switch (func_id) {
6945         /* inet and inet6 sockets are created in a process
6946          * context so there is always a valid uid/gid
6947          */
6948         case BPF_FUNC_get_current_uid_gid:
6949                 return &bpf_get_current_uid_gid_proto;
6950         case BPF_FUNC_get_local_storage:
6951                 return &bpf_get_local_storage_proto;
6952         case BPF_FUNC_get_socket_cookie:
6953                 return &bpf_get_socket_cookie_sock_proto;
6954         case BPF_FUNC_get_netns_cookie:
6955                 return &bpf_get_netns_cookie_sock_proto;
6956         case BPF_FUNC_perf_event_output:
6957                 return &bpf_event_output_data_proto;
6958         case BPF_FUNC_get_current_pid_tgid:
6959                 return &bpf_get_current_pid_tgid_proto;
6960         case BPF_FUNC_get_current_comm:
6961                 return &bpf_get_current_comm_proto;
6962 #ifdef CONFIG_CGROUPS
6963         case BPF_FUNC_get_current_cgroup_id:
6964                 return &bpf_get_current_cgroup_id_proto;
6965         case BPF_FUNC_get_current_ancestor_cgroup_id:
6966                 return &bpf_get_current_ancestor_cgroup_id_proto;
6967 #endif
6968 #ifdef CONFIG_CGROUP_NET_CLASSID
6969         case BPF_FUNC_get_cgroup_classid:
6970                 return &bpf_get_cgroup_classid_curr_proto;
6971 #endif
6972         case BPF_FUNC_sk_storage_get:
6973                 return &bpf_sk_storage_get_cg_sock_proto;
6974         default:
6975                 return bpf_base_func_proto(func_id);
6976         }
6977 }
6978
6979 static const struct bpf_func_proto *
6980 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
6981 {
6982         switch (func_id) {
6983         /* inet and inet6 sockets are created in a process
6984          * context so there is always a valid uid/gid
6985          */
6986         case BPF_FUNC_get_current_uid_gid:
6987                 return &bpf_get_current_uid_gid_proto;
6988         case BPF_FUNC_bind:
6989                 switch (prog->expected_attach_type) {
6990                 case BPF_CGROUP_INET4_CONNECT:
6991                 case BPF_CGROUP_INET6_CONNECT:
6992                         return &bpf_bind_proto;
6993                 default:
6994                         return NULL;
6995                 }
6996         case BPF_FUNC_get_socket_cookie:
6997                 return &bpf_get_socket_cookie_sock_addr_proto;
6998         case BPF_FUNC_get_netns_cookie:
6999                 return &bpf_get_netns_cookie_sock_addr_proto;
7000         case BPF_FUNC_get_local_storage:
7001                 return &bpf_get_local_storage_proto;
7002         case BPF_FUNC_perf_event_output:
7003                 return &bpf_event_output_data_proto;
7004         case BPF_FUNC_get_current_pid_tgid:
7005                 return &bpf_get_current_pid_tgid_proto;
7006         case BPF_FUNC_get_current_comm:
7007                 return &bpf_get_current_comm_proto;
7008 #ifdef CONFIG_CGROUPS
7009         case BPF_FUNC_get_current_cgroup_id:
7010                 return &bpf_get_current_cgroup_id_proto;
7011         case BPF_FUNC_get_current_ancestor_cgroup_id:
7012                 return &bpf_get_current_ancestor_cgroup_id_proto;
7013 #endif
7014 #ifdef CONFIG_CGROUP_NET_CLASSID
7015         case BPF_FUNC_get_cgroup_classid:
7016                 return &bpf_get_cgroup_classid_curr_proto;
7017 #endif
7018 #ifdef CONFIG_INET
7019         case BPF_FUNC_sk_lookup_tcp:
7020                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7021         case BPF_FUNC_sk_lookup_udp:
7022                 return &bpf_sock_addr_sk_lookup_udp_proto;
7023         case BPF_FUNC_sk_release:
7024                 return &bpf_sk_release_proto;
7025         case BPF_FUNC_skc_lookup_tcp:
7026                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7027 #endif /* CONFIG_INET */
7028         case BPF_FUNC_sk_storage_get:
7029                 return &bpf_sk_storage_get_proto;
7030         case BPF_FUNC_sk_storage_delete:
7031                 return &bpf_sk_storage_delete_proto;
7032         case BPF_FUNC_setsockopt:
7033                 switch (prog->expected_attach_type) {
7034                 case BPF_CGROUP_INET4_CONNECT:
7035                 case BPF_CGROUP_INET6_CONNECT:
7036                         return &bpf_sock_addr_setsockopt_proto;
7037                 default:
7038                         return NULL;
7039                 }
7040         case BPF_FUNC_getsockopt:
7041                 switch (prog->expected_attach_type) {
7042                 case BPF_CGROUP_INET4_CONNECT:
7043                 case BPF_CGROUP_INET6_CONNECT:
7044                         return &bpf_sock_addr_getsockopt_proto;
7045                 default:
7046                         return NULL;
7047                 }
7048         default:
7049                 return bpf_sk_base_func_proto(func_id);
7050         }
7051 }
7052
7053 static const struct bpf_func_proto *
7054 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7055 {
7056         switch (func_id) {
7057         case BPF_FUNC_skb_load_bytes:
7058                 return &bpf_skb_load_bytes_proto;
7059         case BPF_FUNC_skb_load_bytes_relative:
7060                 return &bpf_skb_load_bytes_relative_proto;
7061         case BPF_FUNC_get_socket_cookie:
7062                 return &bpf_get_socket_cookie_proto;
7063         case BPF_FUNC_get_socket_uid:
7064                 return &bpf_get_socket_uid_proto;
7065         case BPF_FUNC_perf_event_output:
7066                 return &bpf_skb_event_output_proto;
7067         default:
7068                 return bpf_sk_base_func_proto(func_id);
7069         }
7070 }
7071
7072 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7073 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7074
7075 static const struct bpf_func_proto *
7076 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7077 {
7078         switch (func_id) {
7079         case BPF_FUNC_get_local_storage:
7080                 return &bpf_get_local_storage_proto;
7081         case BPF_FUNC_sk_fullsock:
7082                 return &bpf_sk_fullsock_proto;
7083         case BPF_FUNC_sk_storage_get:
7084                 return &bpf_sk_storage_get_proto;
7085         case BPF_FUNC_sk_storage_delete:
7086                 return &bpf_sk_storage_delete_proto;
7087         case BPF_FUNC_perf_event_output:
7088                 return &bpf_skb_event_output_proto;
7089 #ifdef CONFIG_SOCK_CGROUP_DATA
7090         case BPF_FUNC_skb_cgroup_id:
7091                 return &bpf_skb_cgroup_id_proto;
7092         case BPF_FUNC_skb_ancestor_cgroup_id:
7093                 return &bpf_skb_ancestor_cgroup_id_proto;
7094         case BPF_FUNC_sk_cgroup_id:
7095                 return &bpf_sk_cgroup_id_proto;
7096         case BPF_FUNC_sk_ancestor_cgroup_id:
7097                 return &bpf_sk_ancestor_cgroup_id_proto;
7098 #endif
7099 #ifdef CONFIG_INET
7100         case BPF_FUNC_sk_lookup_tcp:
7101                 return &bpf_sk_lookup_tcp_proto;
7102         case BPF_FUNC_sk_lookup_udp:
7103                 return &bpf_sk_lookup_udp_proto;
7104         case BPF_FUNC_sk_release:
7105                 return &bpf_sk_release_proto;
7106         case BPF_FUNC_skc_lookup_tcp:
7107                 return &bpf_skc_lookup_tcp_proto;
7108         case BPF_FUNC_tcp_sock:
7109                 return &bpf_tcp_sock_proto;
7110         case BPF_FUNC_get_listener_sock:
7111                 return &bpf_get_listener_sock_proto;
7112         case BPF_FUNC_skb_ecn_set_ce:
7113                 return &bpf_skb_ecn_set_ce_proto;
7114 #endif
7115         default:
7116                 return sk_filter_func_proto(func_id, prog);
7117         }
7118 }
7119
7120 static const struct bpf_func_proto *
7121 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7122 {
7123         switch (func_id) {
7124         case BPF_FUNC_skb_store_bytes:
7125                 return &bpf_skb_store_bytes_proto;
7126         case BPF_FUNC_skb_load_bytes:
7127                 return &bpf_skb_load_bytes_proto;
7128         case BPF_FUNC_skb_load_bytes_relative:
7129                 return &bpf_skb_load_bytes_relative_proto;
7130         case BPF_FUNC_skb_pull_data:
7131                 return &bpf_skb_pull_data_proto;
7132         case BPF_FUNC_csum_diff:
7133                 return &bpf_csum_diff_proto;
7134         case BPF_FUNC_csum_update:
7135                 return &bpf_csum_update_proto;
7136         case BPF_FUNC_csum_level:
7137                 return &bpf_csum_level_proto;
7138         case BPF_FUNC_l3_csum_replace:
7139                 return &bpf_l3_csum_replace_proto;
7140         case BPF_FUNC_l4_csum_replace:
7141                 return &bpf_l4_csum_replace_proto;
7142         case BPF_FUNC_clone_redirect:
7143                 return &bpf_clone_redirect_proto;
7144         case BPF_FUNC_get_cgroup_classid:
7145                 return &bpf_get_cgroup_classid_proto;
7146         case BPF_FUNC_skb_vlan_push:
7147                 return &bpf_skb_vlan_push_proto;
7148         case BPF_FUNC_skb_vlan_pop:
7149                 return &bpf_skb_vlan_pop_proto;
7150         case BPF_FUNC_skb_change_proto:
7151                 return &bpf_skb_change_proto_proto;
7152         case BPF_FUNC_skb_change_type:
7153                 return &bpf_skb_change_type_proto;
7154         case BPF_FUNC_skb_adjust_room:
7155                 return &bpf_skb_adjust_room_proto;
7156         case BPF_FUNC_skb_change_tail:
7157                 return &bpf_skb_change_tail_proto;
7158         case BPF_FUNC_skb_change_head:
7159                 return &bpf_skb_change_head_proto;
7160         case BPF_FUNC_skb_get_tunnel_key:
7161                 return &bpf_skb_get_tunnel_key_proto;
7162         case BPF_FUNC_skb_set_tunnel_key:
7163                 return bpf_get_skb_set_tunnel_proto(func_id);
7164         case BPF_FUNC_skb_get_tunnel_opt:
7165                 return &bpf_skb_get_tunnel_opt_proto;
7166         case BPF_FUNC_skb_set_tunnel_opt:
7167                 return bpf_get_skb_set_tunnel_proto(func_id);
7168         case BPF_FUNC_redirect:
7169                 return &bpf_redirect_proto;
7170         case BPF_FUNC_redirect_neigh:
7171                 return &bpf_redirect_neigh_proto;
7172         case BPF_FUNC_redirect_peer:
7173                 return &bpf_redirect_peer_proto;
7174         case BPF_FUNC_get_route_realm:
7175                 return &bpf_get_route_realm_proto;
7176         case BPF_FUNC_get_hash_recalc:
7177                 return &bpf_get_hash_recalc_proto;
7178         case BPF_FUNC_set_hash_invalid:
7179                 return &bpf_set_hash_invalid_proto;
7180         case BPF_FUNC_set_hash:
7181                 return &bpf_set_hash_proto;
7182         case BPF_FUNC_perf_event_output:
7183                 return &bpf_skb_event_output_proto;
7184         case BPF_FUNC_get_smp_processor_id:
7185                 return &bpf_get_smp_processor_id_proto;
7186         case BPF_FUNC_skb_under_cgroup:
7187                 return &bpf_skb_under_cgroup_proto;
7188         case BPF_FUNC_get_socket_cookie:
7189                 return &bpf_get_socket_cookie_proto;
7190         case BPF_FUNC_get_socket_uid:
7191                 return &bpf_get_socket_uid_proto;
7192         case BPF_FUNC_fib_lookup:
7193                 return &bpf_skb_fib_lookup_proto;
7194         case BPF_FUNC_sk_fullsock:
7195                 return &bpf_sk_fullsock_proto;
7196         case BPF_FUNC_sk_storage_get:
7197                 return &bpf_sk_storage_get_proto;
7198         case BPF_FUNC_sk_storage_delete:
7199                 return &bpf_sk_storage_delete_proto;
7200 #ifdef CONFIG_XFRM
7201         case BPF_FUNC_skb_get_xfrm_state:
7202                 return &bpf_skb_get_xfrm_state_proto;
7203 #endif
7204 #ifdef CONFIG_CGROUP_NET_CLASSID
7205         case BPF_FUNC_skb_cgroup_classid:
7206                 return &bpf_skb_cgroup_classid_proto;
7207 #endif
7208 #ifdef CONFIG_SOCK_CGROUP_DATA
7209         case BPF_FUNC_skb_cgroup_id:
7210                 return &bpf_skb_cgroup_id_proto;
7211         case BPF_FUNC_skb_ancestor_cgroup_id:
7212                 return &bpf_skb_ancestor_cgroup_id_proto;
7213 #endif
7214 #ifdef CONFIG_INET
7215         case BPF_FUNC_sk_lookup_tcp:
7216                 return &bpf_sk_lookup_tcp_proto;
7217         case BPF_FUNC_sk_lookup_udp:
7218                 return &bpf_sk_lookup_udp_proto;
7219         case BPF_FUNC_sk_release:
7220                 return &bpf_sk_release_proto;
7221         case BPF_FUNC_tcp_sock:
7222                 return &bpf_tcp_sock_proto;
7223         case BPF_FUNC_get_listener_sock:
7224                 return &bpf_get_listener_sock_proto;
7225         case BPF_FUNC_skc_lookup_tcp:
7226                 return &bpf_skc_lookup_tcp_proto;
7227         case BPF_FUNC_tcp_check_syncookie:
7228                 return &bpf_tcp_check_syncookie_proto;
7229         case BPF_FUNC_skb_ecn_set_ce:
7230                 return &bpf_skb_ecn_set_ce_proto;
7231         case BPF_FUNC_tcp_gen_syncookie:
7232                 return &bpf_tcp_gen_syncookie_proto;
7233         case BPF_FUNC_sk_assign:
7234                 return &bpf_sk_assign_proto;
7235 #endif
7236         default:
7237                 return bpf_sk_base_func_proto(func_id);
7238         }
7239 }
7240
7241 static const struct bpf_func_proto *
7242 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7243 {
7244         switch (func_id) {
7245         case BPF_FUNC_perf_event_output:
7246                 return &bpf_xdp_event_output_proto;
7247         case BPF_FUNC_get_smp_processor_id:
7248                 return &bpf_get_smp_processor_id_proto;
7249         case BPF_FUNC_csum_diff:
7250                 return &bpf_csum_diff_proto;
7251         case BPF_FUNC_xdp_adjust_head:
7252                 return &bpf_xdp_adjust_head_proto;
7253         case BPF_FUNC_xdp_adjust_meta:
7254                 return &bpf_xdp_adjust_meta_proto;
7255         case BPF_FUNC_redirect:
7256                 return &bpf_xdp_redirect_proto;
7257         case BPF_FUNC_redirect_map:
7258                 return &bpf_xdp_redirect_map_proto;
7259         case BPF_FUNC_xdp_adjust_tail:
7260                 return &bpf_xdp_adjust_tail_proto;
7261         case BPF_FUNC_fib_lookup:
7262                 return &bpf_xdp_fib_lookup_proto;
7263 #ifdef CONFIG_INET
7264         case BPF_FUNC_sk_lookup_udp:
7265                 return &bpf_xdp_sk_lookup_udp_proto;
7266         case BPF_FUNC_sk_lookup_tcp:
7267                 return &bpf_xdp_sk_lookup_tcp_proto;
7268         case BPF_FUNC_sk_release:
7269                 return &bpf_sk_release_proto;
7270         case BPF_FUNC_skc_lookup_tcp:
7271                 return &bpf_xdp_skc_lookup_tcp_proto;
7272         case BPF_FUNC_tcp_check_syncookie:
7273                 return &bpf_tcp_check_syncookie_proto;
7274         case BPF_FUNC_tcp_gen_syncookie:
7275                 return &bpf_tcp_gen_syncookie_proto;
7276 #endif
7277         default:
7278                 return bpf_sk_base_func_proto(func_id);
7279         }
7280 }
7281
7282 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7283 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7284
7285 static const struct bpf_func_proto *
7286 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7287 {
7288         switch (func_id) {
7289         case BPF_FUNC_setsockopt:
7290                 return &bpf_sock_ops_setsockopt_proto;
7291         case BPF_FUNC_getsockopt:
7292                 return &bpf_sock_ops_getsockopt_proto;
7293         case BPF_FUNC_sock_ops_cb_flags_set:
7294                 return &bpf_sock_ops_cb_flags_set_proto;
7295         case BPF_FUNC_sock_map_update:
7296                 return &bpf_sock_map_update_proto;
7297         case BPF_FUNC_sock_hash_update:
7298                 return &bpf_sock_hash_update_proto;
7299         case BPF_FUNC_get_socket_cookie:
7300                 return &bpf_get_socket_cookie_sock_ops_proto;
7301         case BPF_FUNC_get_local_storage:
7302                 return &bpf_get_local_storage_proto;
7303         case BPF_FUNC_perf_event_output:
7304                 return &bpf_event_output_data_proto;
7305         case BPF_FUNC_sk_storage_get:
7306                 return &bpf_sk_storage_get_proto;
7307         case BPF_FUNC_sk_storage_delete:
7308                 return &bpf_sk_storage_delete_proto;
7309 #ifdef CONFIG_INET
7310         case BPF_FUNC_load_hdr_opt:
7311                 return &bpf_sock_ops_load_hdr_opt_proto;
7312         case BPF_FUNC_store_hdr_opt:
7313                 return &bpf_sock_ops_store_hdr_opt_proto;
7314         case BPF_FUNC_reserve_hdr_opt:
7315                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7316         case BPF_FUNC_tcp_sock:
7317                 return &bpf_tcp_sock_proto;
7318 #endif /* CONFIG_INET */
7319         default:
7320                 return bpf_sk_base_func_proto(func_id);
7321         }
7322 }
7323
7324 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7325 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7326
7327 static const struct bpf_func_proto *
7328 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7329 {
7330         switch (func_id) {
7331         case BPF_FUNC_msg_redirect_map:
7332                 return &bpf_msg_redirect_map_proto;
7333         case BPF_FUNC_msg_redirect_hash:
7334                 return &bpf_msg_redirect_hash_proto;
7335         case BPF_FUNC_msg_apply_bytes:
7336                 return &bpf_msg_apply_bytes_proto;
7337         case BPF_FUNC_msg_cork_bytes:
7338                 return &bpf_msg_cork_bytes_proto;
7339         case BPF_FUNC_msg_pull_data:
7340                 return &bpf_msg_pull_data_proto;
7341         case BPF_FUNC_msg_push_data:
7342                 return &bpf_msg_push_data_proto;
7343         case BPF_FUNC_msg_pop_data:
7344                 return &bpf_msg_pop_data_proto;
7345         case BPF_FUNC_perf_event_output:
7346                 return &bpf_event_output_data_proto;
7347         case BPF_FUNC_get_current_uid_gid:
7348                 return &bpf_get_current_uid_gid_proto;
7349         case BPF_FUNC_get_current_pid_tgid:
7350                 return &bpf_get_current_pid_tgid_proto;
7351         case BPF_FUNC_sk_storage_get:
7352                 return &bpf_sk_storage_get_proto;
7353         case BPF_FUNC_sk_storage_delete:
7354                 return &bpf_sk_storage_delete_proto;
7355 #ifdef CONFIG_CGROUPS
7356         case BPF_FUNC_get_current_cgroup_id:
7357                 return &bpf_get_current_cgroup_id_proto;
7358         case BPF_FUNC_get_current_ancestor_cgroup_id:
7359                 return &bpf_get_current_ancestor_cgroup_id_proto;
7360 #endif
7361 #ifdef CONFIG_CGROUP_NET_CLASSID
7362         case BPF_FUNC_get_cgroup_classid:
7363                 return &bpf_get_cgroup_classid_curr_proto;
7364 #endif
7365         default:
7366                 return bpf_sk_base_func_proto(func_id);
7367         }
7368 }
7369
7370 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7371 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7372
7373 static const struct bpf_func_proto *
7374 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7375 {
7376         switch (func_id) {
7377         case BPF_FUNC_skb_store_bytes:
7378                 return &bpf_skb_store_bytes_proto;
7379         case BPF_FUNC_skb_load_bytes:
7380                 return &bpf_skb_load_bytes_proto;
7381         case BPF_FUNC_skb_pull_data:
7382                 return &sk_skb_pull_data_proto;
7383         case BPF_FUNC_skb_change_tail:
7384                 return &sk_skb_change_tail_proto;
7385         case BPF_FUNC_skb_change_head:
7386                 return &sk_skb_change_head_proto;
7387         case BPF_FUNC_skb_adjust_room:
7388                 return &sk_skb_adjust_room_proto;
7389         case BPF_FUNC_get_socket_cookie:
7390                 return &bpf_get_socket_cookie_proto;
7391         case BPF_FUNC_get_socket_uid:
7392                 return &bpf_get_socket_uid_proto;
7393         case BPF_FUNC_sk_redirect_map:
7394                 return &bpf_sk_redirect_map_proto;
7395         case BPF_FUNC_sk_redirect_hash:
7396                 return &bpf_sk_redirect_hash_proto;
7397         case BPF_FUNC_perf_event_output:
7398                 return &bpf_skb_event_output_proto;
7399 #ifdef CONFIG_INET
7400         case BPF_FUNC_sk_lookup_tcp:
7401                 return &bpf_sk_lookup_tcp_proto;
7402         case BPF_FUNC_sk_lookup_udp:
7403                 return &bpf_sk_lookup_udp_proto;
7404         case BPF_FUNC_sk_release:
7405                 return &bpf_sk_release_proto;
7406         case BPF_FUNC_skc_lookup_tcp:
7407                 return &bpf_skc_lookup_tcp_proto;
7408 #endif
7409         default:
7410                 return bpf_sk_base_func_proto(func_id);
7411         }
7412 }
7413
7414 static const struct bpf_func_proto *
7415 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7416 {
7417         switch (func_id) {
7418         case BPF_FUNC_skb_load_bytes:
7419                 return &bpf_flow_dissector_load_bytes_proto;
7420         default:
7421                 return bpf_sk_base_func_proto(func_id);
7422         }
7423 }
7424
7425 static const struct bpf_func_proto *
7426 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7427 {
7428         switch (func_id) {
7429         case BPF_FUNC_skb_load_bytes:
7430                 return &bpf_skb_load_bytes_proto;
7431         case BPF_FUNC_skb_pull_data:
7432                 return &bpf_skb_pull_data_proto;
7433         case BPF_FUNC_csum_diff:
7434                 return &bpf_csum_diff_proto;
7435         case BPF_FUNC_get_cgroup_classid:
7436                 return &bpf_get_cgroup_classid_proto;
7437         case BPF_FUNC_get_route_realm:
7438                 return &bpf_get_route_realm_proto;
7439         case BPF_FUNC_get_hash_recalc:
7440                 return &bpf_get_hash_recalc_proto;
7441         case BPF_FUNC_perf_event_output:
7442                 return &bpf_skb_event_output_proto;
7443         case BPF_FUNC_get_smp_processor_id:
7444                 return &bpf_get_smp_processor_id_proto;
7445         case BPF_FUNC_skb_under_cgroup:
7446                 return &bpf_skb_under_cgroup_proto;
7447         default:
7448                 return bpf_sk_base_func_proto(func_id);
7449         }
7450 }
7451
7452 static const struct bpf_func_proto *
7453 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7454 {
7455         switch (func_id) {
7456         case BPF_FUNC_lwt_push_encap:
7457                 return &bpf_lwt_in_push_encap_proto;
7458         default:
7459                 return lwt_out_func_proto(func_id, prog);
7460         }
7461 }
7462
7463 static const struct bpf_func_proto *
7464 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7465 {
7466         switch (func_id) {
7467         case BPF_FUNC_skb_get_tunnel_key:
7468                 return &bpf_skb_get_tunnel_key_proto;
7469         case BPF_FUNC_skb_set_tunnel_key:
7470                 return bpf_get_skb_set_tunnel_proto(func_id);
7471         case BPF_FUNC_skb_get_tunnel_opt:
7472                 return &bpf_skb_get_tunnel_opt_proto;
7473         case BPF_FUNC_skb_set_tunnel_opt:
7474                 return bpf_get_skb_set_tunnel_proto(func_id);
7475         case BPF_FUNC_redirect:
7476                 return &bpf_redirect_proto;
7477         case BPF_FUNC_clone_redirect:
7478                 return &bpf_clone_redirect_proto;
7479         case BPF_FUNC_skb_change_tail:
7480                 return &bpf_skb_change_tail_proto;
7481         case BPF_FUNC_skb_change_head:
7482                 return &bpf_skb_change_head_proto;
7483         case BPF_FUNC_skb_store_bytes:
7484                 return &bpf_skb_store_bytes_proto;
7485         case BPF_FUNC_csum_update:
7486                 return &bpf_csum_update_proto;
7487         case BPF_FUNC_csum_level:
7488                 return &bpf_csum_level_proto;
7489         case BPF_FUNC_l3_csum_replace:
7490                 return &bpf_l3_csum_replace_proto;
7491         case BPF_FUNC_l4_csum_replace:
7492                 return &bpf_l4_csum_replace_proto;
7493         case BPF_FUNC_set_hash_invalid:
7494                 return &bpf_set_hash_invalid_proto;
7495         case BPF_FUNC_lwt_push_encap:
7496                 return &bpf_lwt_xmit_push_encap_proto;
7497         default:
7498                 return lwt_out_func_proto(func_id, prog);
7499         }
7500 }
7501
7502 static const struct bpf_func_proto *
7503 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7504 {
7505         switch (func_id) {
7506 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7507         case BPF_FUNC_lwt_seg6_store_bytes:
7508                 return &bpf_lwt_seg6_store_bytes_proto;
7509         case BPF_FUNC_lwt_seg6_action:
7510                 return &bpf_lwt_seg6_action_proto;
7511         case BPF_FUNC_lwt_seg6_adjust_srh:
7512                 return &bpf_lwt_seg6_adjust_srh_proto;
7513 #endif
7514         default:
7515                 return lwt_out_func_proto(func_id, prog);
7516         }
7517 }
7518
7519 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7520                                     const struct bpf_prog *prog,
7521                                     struct bpf_insn_access_aux *info)
7522 {
7523         const int size_default = sizeof(__u32);
7524
7525         if (off < 0 || off >= sizeof(struct __sk_buff))
7526                 return false;
7527
7528         /* The verifier guarantees that size > 0. */
7529         if (off % size != 0)
7530                 return false;
7531
7532         switch (off) {
7533         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7534                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7535                         return false;
7536                 break;
7537         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7538         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7539         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7540         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7541         case bpf_ctx_range(struct __sk_buff, data):
7542         case bpf_ctx_range(struct __sk_buff, data_meta):
7543         case bpf_ctx_range(struct __sk_buff, data_end):
7544                 if (size != size_default)
7545                         return false;
7546                 break;
7547         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7548                 return false;
7549         case bpf_ctx_range(struct __sk_buff, tstamp):
7550                 if (size != sizeof(__u64))
7551                         return false;
7552                 break;
7553         case offsetof(struct __sk_buff, sk):
7554                 if (type == BPF_WRITE || size != sizeof(__u64))
7555                         return false;
7556                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7557                 break;
7558         default:
7559                 /* Only narrow read access allowed for now. */
7560                 if (type == BPF_WRITE) {
7561                         if (size != size_default)
7562                                 return false;
7563                 } else {
7564                         bpf_ctx_record_field_size(info, size_default);
7565                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7566                                 return false;
7567                 }
7568         }
7569
7570         return true;
7571 }
7572
7573 static bool sk_filter_is_valid_access(int off, int size,
7574                                       enum bpf_access_type type,
7575                                       const struct bpf_prog *prog,
7576                                       struct bpf_insn_access_aux *info)
7577 {
7578         switch (off) {
7579         case bpf_ctx_range(struct __sk_buff, tc_classid):
7580         case bpf_ctx_range(struct __sk_buff, data):
7581         case bpf_ctx_range(struct __sk_buff, data_meta):
7582         case bpf_ctx_range(struct __sk_buff, data_end):
7583         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7584         case bpf_ctx_range(struct __sk_buff, tstamp):
7585         case bpf_ctx_range(struct __sk_buff, wire_len):
7586                 return false;
7587         }
7588
7589         if (type == BPF_WRITE) {
7590                 switch (off) {
7591                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7592                         break;
7593                 default:
7594                         return false;
7595                 }
7596         }
7597
7598         return bpf_skb_is_valid_access(off, size, type, prog, info);
7599 }
7600
7601 static bool cg_skb_is_valid_access(int off, int size,
7602                                    enum bpf_access_type type,
7603                                    const struct bpf_prog *prog,
7604                                    struct bpf_insn_access_aux *info)
7605 {
7606         switch (off) {
7607         case bpf_ctx_range(struct __sk_buff, tc_classid):
7608         case bpf_ctx_range(struct __sk_buff, data_meta):
7609         case bpf_ctx_range(struct __sk_buff, wire_len):
7610                 return false;
7611         case bpf_ctx_range(struct __sk_buff, data):
7612         case bpf_ctx_range(struct __sk_buff, data_end):
7613                 if (!bpf_capable())
7614                         return false;
7615                 break;
7616         }
7617
7618         if (type == BPF_WRITE) {
7619                 switch (off) {
7620                 case bpf_ctx_range(struct __sk_buff, mark):
7621                 case bpf_ctx_range(struct __sk_buff, priority):
7622                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7623                         break;
7624                 case bpf_ctx_range(struct __sk_buff, tstamp):
7625                         if (!bpf_capable())
7626                                 return false;
7627                         break;
7628                 default:
7629                         return false;
7630                 }
7631         }
7632
7633         switch (off) {
7634         case bpf_ctx_range(struct __sk_buff, data):
7635                 info->reg_type = PTR_TO_PACKET;
7636                 break;
7637         case bpf_ctx_range(struct __sk_buff, data_end):
7638                 info->reg_type = PTR_TO_PACKET_END;
7639                 break;
7640         }
7641
7642         return bpf_skb_is_valid_access(off, size, type, prog, info);
7643 }
7644
7645 static bool lwt_is_valid_access(int off, int size,
7646                                 enum bpf_access_type type,
7647                                 const struct bpf_prog *prog,
7648                                 struct bpf_insn_access_aux *info)
7649 {
7650         switch (off) {
7651         case bpf_ctx_range(struct __sk_buff, tc_classid):
7652         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7653         case bpf_ctx_range(struct __sk_buff, data_meta):
7654         case bpf_ctx_range(struct __sk_buff, tstamp):
7655         case bpf_ctx_range(struct __sk_buff, wire_len):
7656                 return false;
7657         }
7658
7659         if (type == BPF_WRITE) {
7660                 switch (off) {
7661                 case bpf_ctx_range(struct __sk_buff, mark):
7662                 case bpf_ctx_range(struct __sk_buff, priority):
7663                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7664                         break;
7665                 default:
7666                         return false;
7667                 }
7668         }
7669
7670         switch (off) {
7671         case bpf_ctx_range(struct __sk_buff, data):
7672                 info->reg_type = PTR_TO_PACKET;
7673                 break;
7674         case bpf_ctx_range(struct __sk_buff, data_end):
7675                 info->reg_type = PTR_TO_PACKET_END;
7676                 break;
7677         }
7678
7679         return bpf_skb_is_valid_access(off, size, type, prog, info);
7680 }
7681
7682 /* Attach type specific accesses */
7683 static bool __sock_filter_check_attach_type(int off,
7684                                             enum bpf_access_type access_type,
7685                                             enum bpf_attach_type attach_type)
7686 {
7687         switch (off) {
7688         case offsetof(struct bpf_sock, bound_dev_if):
7689         case offsetof(struct bpf_sock, mark):
7690         case offsetof(struct bpf_sock, priority):
7691                 switch (attach_type) {
7692                 case BPF_CGROUP_INET_SOCK_CREATE:
7693                 case BPF_CGROUP_INET_SOCK_RELEASE:
7694                         goto full_access;
7695                 default:
7696                         return false;
7697                 }
7698         case bpf_ctx_range(struct bpf_sock, src_ip4):
7699                 switch (attach_type) {
7700                 case BPF_CGROUP_INET4_POST_BIND:
7701                         goto read_only;
7702                 default:
7703                         return false;
7704                 }
7705         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7706                 switch (attach_type) {
7707                 case BPF_CGROUP_INET6_POST_BIND:
7708                         goto read_only;
7709                 default:
7710                         return false;
7711                 }
7712         case bpf_ctx_range(struct bpf_sock, src_port):
7713                 switch (attach_type) {
7714                 case BPF_CGROUP_INET4_POST_BIND:
7715                 case BPF_CGROUP_INET6_POST_BIND:
7716                         goto read_only;
7717                 default:
7718                         return false;
7719                 }
7720         }
7721 read_only:
7722         return access_type == BPF_READ;
7723 full_access:
7724         return true;
7725 }
7726
7727 bool bpf_sock_common_is_valid_access(int off, int size,
7728                                      enum bpf_access_type type,
7729                                      struct bpf_insn_access_aux *info)
7730 {
7731         switch (off) {
7732         case bpf_ctx_range_till(struct bpf_sock, type, priority):
7733                 return false;
7734         default:
7735                 return bpf_sock_is_valid_access(off, size, type, info);
7736         }
7737 }
7738
7739 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7740                               struct bpf_insn_access_aux *info)
7741 {
7742         const int size_default = sizeof(__u32);
7743         int field_size;
7744
7745         if (off < 0 || off >= sizeof(struct bpf_sock))
7746                 return false;
7747         if (off % size != 0)
7748                 return false;
7749
7750         switch (off) {
7751         case offsetof(struct bpf_sock, state):
7752         case offsetof(struct bpf_sock, family):
7753         case offsetof(struct bpf_sock, type):
7754         case offsetof(struct bpf_sock, protocol):
7755         case offsetof(struct bpf_sock, src_port):
7756         case offsetof(struct bpf_sock, rx_queue_mapping):
7757         case bpf_ctx_range(struct bpf_sock, src_ip4):
7758         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7759         case bpf_ctx_range(struct bpf_sock, dst_ip4):
7760         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7761                 bpf_ctx_record_field_size(info, size_default);
7762                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7763         case bpf_ctx_range(struct bpf_sock, dst_port):
7764                 field_size = size == size_default ?
7765                         size_default : sizeof_field(struct bpf_sock, dst_port);
7766                 bpf_ctx_record_field_size(info, field_size);
7767                 return bpf_ctx_narrow_access_ok(off, size, field_size);
7768         case offsetofend(struct bpf_sock, dst_port) ...
7769              offsetof(struct bpf_sock, dst_ip4) - 1:
7770                 return false;
7771         }
7772
7773         return size == size_default;
7774 }
7775
7776 static bool sock_filter_is_valid_access(int off, int size,
7777                                         enum bpf_access_type type,
7778                                         const struct bpf_prog *prog,
7779                                         struct bpf_insn_access_aux *info)
7780 {
7781         if (!bpf_sock_is_valid_access(off, size, type, info))
7782                 return false;
7783         return __sock_filter_check_attach_type(off, type,
7784                                                prog->expected_attach_type);
7785 }
7786
7787 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7788                              const struct bpf_prog *prog)
7789 {
7790         /* Neither direct read nor direct write requires any preliminary
7791          * action.
7792          */
7793         return 0;
7794 }
7795
7796 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7797                                 const struct bpf_prog *prog, int drop_verdict)
7798 {
7799         struct bpf_insn *insn = insn_buf;
7800
7801         if (!direct_write)
7802                 return 0;
7803
7804         /* if (!skb->cloned)
7805          *       goto start;
7806          *
7807          * (Fast-path, otherwise approximation that we might be
7808          *  a clone, do the rest in helper.)
7809          */
7810         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7811         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7812         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7813
7814         /* ret = bpf_skb_pull_data(skb, 0); */
7815         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7816         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7817         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7818                                BPF_FUNC_skb_pull_data);
7819         /* if (!ret)
7820          *      goto restore;
7821          * return TC_ACT_SHOT;
7822          */
7823         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7824         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7825         *insn++ = BPF_EXIT_INSN();
7826
7827         /* restore: */
7828         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7829         /* start: */
7830         *insn++ = prog->insnsi[0];
7831
7832         return insn - insn_buf;
7833 }
7834
7835 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7836                           struct bpf_insn *insn_buf)
7837 {
7838         bool indirect = BPF_MODE(orig->code) == BPF_IND;
7839         struct bpf_insn *insn = insn_buf;
7840
7841         if (!indirect) {
7842                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7843         } else {
7844                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7845                 if (orig->imm)
7846                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7847         }
7848         /* We're guaranteed here that CTX is in R6. */
7849         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7850
7851         switch (BPF_SIZE(orig->code)) {
7852         case BPF_B:
7853                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7854                 break;
7855         case BPF_H:
7856                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7857                 break;
7858         case BPF_W:
7859                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7860                 break;
7861         }
7862
7863         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7864         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7865         *insn++ = BPF_EXIT_INSN();
7866
7867         return insn - insn_buf;
7868 }
7869
7870 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7871                                const struct bpf_prog *prog)
7872 {
7873         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7874 }
7875
7876 static bool tc_cls_act_is_valid_access(int off, int size,
7877                                        enum bpf_access_type type,
7878                                        const struct bpf_prog *prog,
7879                                        struct bpf_insn_access_aux *info)
7880 {
7881         if (type == BPF_WRITE) {
7882                 switch (off) {
7883                 case bpf_ctx_range(struct __sk_buff, mark):
7884                 case bpf_ctx_range(struct __sk_buff, tc_index):
7885                 case bpf_ctx_range(struct __sk_buff, priority):
7886                 case bpf_ctx_range(struct __sk_buff, tc_classid):
7887                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7888                 case bpf_ctx_range(struct __sk_buff, tstamp):
7889                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7890                         break;
7891                 default:
7892                         return false;
7893                 }
7894         }
7895
7896         switch (off) {
7897         case bpf_ctx_range(struct __sk_buff, data):
7898                 info->reg_type = PTR_TO_PACKET;
7899                 break;
7900         case bpf_ctx_range(struct __sk_buff, data_meta):
7901                 info->reg_type = PTR_TO_PACKET_META;
7902                 break;
7903         case bpf_ctx_range(struct __sk_buff, data_end):
7904                 info->reg_type = PTR_TO_PACKET_END;
7905                 break;
7906         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7907                 return false;
7908         }
7909
7910         return bpf_skb_is_valid_access(off, size, type, prog, info);
7911 }
7912
7913 static bool __is_valid_xdp_access(int off, int size)
7914 {
7915         if (off < 0 || off >= sizeof(struct xdp_md))
7916                 return false;
7917         if (off % size != 0)
7918                 return false;
7919         if (size != sizeof(__u32))
7920                 return false;
7921
7922         return true;
7923 }
7924
7925 static bool xdp_is_valid_access(int off, int size,
7926                                 enum bpf_access_type type,
7927                                 const struct bpf_prog *prog,
7928                                 struct bpf_insn_access_aux *info)
7929 {
7930         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
7931                 switch (off) {
7932                 case offsetof(struct xdp_md, egress_ifindex):
7933                         return false;
7934                 }
7935         }
7936
7937         if (type == BPF_WRITE) {
7938                 if (bpf_prog_is_dev_bound(prog->aux)) {
7939                         switch (off) {
7940                         case offsetof(struct xdp_md, rx_queue_index):
7941                                 return __is_valid_xdp_access(off, size);
7942                         }
7943                 }
7944                 return false;
7945         }
7946
7947         switch (off) {
7948         case offsetof(struct xdp_md, data):
7949                 info->reg_type = PTR_TO_PACKET;
7950                 break;
7951         case offsetof(struct xdp_md, data_meta):
7952                 info->reg_type = PTR_TO_PACKET_META;
7953                 break;
7954         case offsetof(struct xdp_md, data_end):
7955                 info->reg_type = PTR_TO_PACKET_END;
7956                 break;
7957         }
7958
7959         return __is_valid_xdp_access(off, size);
7960 }
7961
7962 void bpf_warn_invalid_xdp_action(u32 act)
7963 {
7964         const u32 act_max = XDP_REDIRECT;
7965
7966         pr_warn_once("%s XDP return value %u, expect packet loss!\n",
7967                      act > act_max ? "Illegal" : "Driver unsupported",
7968                      act);
7969 }
7970 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
7971
7972 static bool sock_addr_is_valid_access(int off, int size,
7973                                       enum bpf_access_type type,
7974                                       const struct bpf_prog *prog,
7975                                       struct bpf_insn_access_aux *info)
7976 {
7977         const int size_default = sizeof(__u32);
7978
7979         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
7980                 return false;
7981         if (off % size != 0)
7982                 return false;
7983
7984         /* Disallow access to IPv6 fields from IPv4 contex and vise
7985          * versa.
7986          */
7987         switch (off) {
7988         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
7989                 switch (prog->expected_attach_type) {
7990                 case BPF_CGROUP_INET4_BIND:
7991                 case BPF_CGROUP_INET4_CONNECT:
7992                 case BPF_CGROUP_INET4_GETPEERNAME:
7993                 case BPF_CGROUP_INET4_GETSOCKNAME:
7994                 case BPF_CGROUP_UDP4_SENDMSG:
7995                 case BPF_CGROUP_UDP4_RECVMSG:
7996                         break;
7997                 default:
7998                         return false;
7999                 }
8000                 break;
8001         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8002                 switch (prog->expected_attach_type) {
8003                 case BPF_CGROUP_INET6_BIND:
8004                 case BPF_CGROUP_INET6_CONNECT:
8005                 case BPF_CGROUP_INET6_GETPEERNAME:
8006                 case BPF_CGROUP_INET6_GETSOCKNAME:
8007                 case BPF_CGROUP_UDP6_SENDMSG:
8008                 case BPF_CGROUP_UDP6_RECVMSG:
8009                         break;
8010                 default:
8011                         return false;
8012                 }
8013                 break;
8014         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8015                 switch (prog->expected_attach_type) {
8016                 case BPF_CGROUP_UDP4_SENDMSG:
8017                         break;
8018                 default:
8019                         return false;
8020                 }
8021                 break;
8022         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8023                                 msg_src_ip6[3]):
8024                 switch (prog->expected_attach_type) {
8025                 case BPF_CGROUP_UDP6_SENDMSG:
8026                         break;
8027                 default:
8028                         return false;
8029                 }
8030                 break;
8031         }
8032
8033         switch (off) {
8034         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8035         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8036         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8037         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8038                                 msg_src_ip6[3]):
8039         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8040                 if (type == BPF_READ) {
8041                         bpf_ctx_record_field_size(info, size_default);
8042
8043                         if (bpf_ctx_wide_access_ok(off, size,
8044                                                    struct bpf_sock_addr,
8045                                                    user_ip6))
8046                                 return true;
8047
8048                         if (bpf_ctx_wide_access_ok(off, size,
8049                                                    struct bpf_sock_addr,
8050                                                    msg_src_ip6))
8051                                 return true;
8052
8053                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8054                                 return false;
8055                 } else {
8056                         if (bpf_ctx_wide_access_ok(off, size,
8057                                                    struct bpf_sock_addr,
8058                                                    user_ip6))
8059                                 return true;
8060
8061                         if (bpf_ctx_wide_access_ok(off, size,
8062                                                    struct bpf_sock_addr,
8063                                                    msg_src_ip6))
8064                                 return true;
8065
8066                         if (size != size_default)
8067                                 return false;
8068                 }
8069                 break;
8070         case offsetof(struct bpf_sock_addr, sk):
8071                 if (type != BPF_READ)
8072                         return false;
8073                 if (size != sizeof(__u64))
8074                         return false;
8075                 info->reg_type = PTR_TO_SOCKET;
8076                 break;
8077         default:
8078                 if (type == BPF_READ) {
8079                         if (size != size_default)
8080                                 return false;
8081                 } else {
8082                         return false;
8083                 }
8084         }
8085
8086         return true;
8087 }
8088
8089 static bool sock_ops_is_valid_access(int off, int size,
8090                                      enum bpf_access_type type,
8091                                      const struct bpf_prog *prog,
8092                                      struct bpf_insn_access_aux *info)
8093 {
8094         const int size_default = sizeof(__u32);
8095
8096         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8097                 return false;
8098
8099         /* The verifier guarantees that size > 0. */
8100         if (off % size != 0)
8101                 return false;
8102
8103         if (type == BPF_WRITE) {
8104                 switch (off) {
8105                 case offsetof(struct bpf_sock_ops, reply):
8106                 case offsetof(struct bpf_sock_ops, sk_txhash):
8107                         if (size != size_default)
8108                                 return false;
8109                         break;
8110                 default:
8111                         return false;
8112                 }
8113         } else {
8114                 switch (off) {
8115                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8116                                         bytes_acked):
8117                         if (size != sizeof(__u64))
8118                                 return false;
8119                         break;
8120                 case offsetof(struct bpf_sock_ops, sk):
8121                         if (size != sizeof(__u64))
8122                                 return false;
8123                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8124                         break;
8125                 case offsetof(struct bpf_sock_ops, skb_data):
8126                         if (size != sizeof(__u64))
8127                                 return false;
8128                         info->reg_type = PTR_TO_PACKET;
8129                         break;
8130                 case offsetof(struct bpf_sock_ops, skb_data_end):
8131                         if (size != sizeof(__u64))
8132                                 return false;
8133                         info->reg_type = PTR_TO_PACKET_END;
8134                         break;
8135                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8136                         bpf_ctx_record_field_size(info, size_default);
8137                         return bpf_ctx_narrow_access_ok(off, size,
8138                                                         size_default);
8139                 default:
8140                         if (size != size_default)
8141                                 return false;
8142                         break;
8143                 }
8144         }
8145
8146         return true;
8147 }
8148
8149 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8150                            const struct bpf_prog *prog)
8151 {
8152         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8153 }
8154
8155 static bool sk_skb_is_valid_access(int off, int size,
8156                                    enum bpf_access_type type,
8157                                    const struct bpf_prog *prog,
8158                                    struct bpf_insn_access_aux *info)
8159 {
8160         switch (off) {
8161         case bpf_ctx_range(struct __sk_buff, tc_classid):
8162         case bpf_ctx_range(struct __sk_buff, data_meta):
8163         case bpf_ctx_range(struct __sk_buff, tstamp):
8164         case bpf_ctx_range(struct __sk_buff, wire_len):
8165                 return false;
8166         }
8167
8168         if (type == BPF_WRITE) {
8169                 switch (off) {
8170                 case bpf_ctx_range(struct __sk_buff, tc_index):
8171                 case bpf_ctx_range(struct __sk_buff, priority):
8172                         break;
8173                 default:
8174                         return false;
8175                 }
8176         }
8177
8178         switch (off) {
8179         case bpf_ctx_range(struct __sk_buff, mark):
8180                 return false;
8181         case bpf_ctx_range(struct __sk_buff, data):
8182                 info->reg_type = PTR_TO_PACKET;
8183                 break;
8184         case bpf_ctx_range(struct __sk_buff, data_end):
8185                 info->reg_type = PTR_TO_PACKET_END;
8186                 break;
8187         }
8188
8189         return bpf_skb_is_valid_access(off, size, type, prog, info);
8190 }
8191
8192 static bool sk_msg_is_valid_access(int off, int size,
8193                                    enum bpf_access_type type,
8194                                    const struct bpf_prog *prog,
8195                                    struct bpf_insn_access_aux *info)
8196 {
8197         if (type == BPF_WRITE)
8198                 return false;
8199
8200         if (off % size != 0)
8201                 return false;
8202
8203         switch (off) {
8204         case offsetof(struct sk_msg_md, data):
8205                 info->reg_type = PTR_TO_PACKET;
8206                 if (size != sizeof(__u64))
8207                         return false;
8208                 break;
8209         case offsetof(struct sk_msg_md, data_end):
8210                 info->reg_type = PTR_TO_PACKET_END;
8211                 if (size != sizeof(__u64))
8212                         return false;
8213                 break;
8214         case offsetof(struct sk_msg_md, sk):
8215                 if (size != sizeof(__u64))
8216                         return false;
8217                 info->reg_type = PTR_TO_SOCKET;
8218                 break;
8219         case bpf_ctx_range(struct sk_msg_md, family):
8220         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8221         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8222         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8223         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8224         case bpf_ctx_range(struct sk_msg_md, remote_port):
8225         case bpf_ctx_range(struct sk_msg_md, local_port):
8226         case bpf_ctx_range(struct sk_msg_md, size):
8227                 if (size != sizeof(__u32))
8228                         return false;
8229                 break;
8230         default:
8231                 return false;
8232         }
8233         return true;
8234 }
8235
8236 static bool flow_dissector_is_valid_access(int off, int size,
8237                                            enum bpf_access_type type,
8238                                            const struct bpf_prog *prog,
8239                                            struct bpf_insn_access_aux *info)
8240 {
8241         const int size_default = sizeof(__u32);
8242
8243         if (off < 0 || off >= sizeof(struct __sk_buff))
8244                 return false;
8245
8246         if (type == BPF_WRITE)
8247                 return false;
8248
8249         switch (off) {
8250         case bpf_ctx_range(struct __sk_buff, data):
8251                 if (size != size_default)
8252                         return false;
8253                 info->reg_type = PTR_TO_PACKET;
8254                 return true;
8255         case bpf_ctx_range(struct __sk_buff, data_end):
8256                 if (size != size_default)
8257                         return false;
8258                 info->reg_type = PTR_TO_PACKET_END;
8259                 return true;
8260         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8261                 if (size != sizeof(__u64))
8262                         return false;
8263                 info->reg_type = PTR_TO_FLOW_KEYS;
8264                 return true;
8265         default:
8266                 return false;
8267         }
8268 }
8269
8270 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8271                                              const struct bpf_insn *si,
8272                                              struct bpf_insn *insn_buf,
8273                                              struct bpf_prog *prog,
8274                                              u32 *target_size)
8275
8276 {
8277         struct bpf_insn *insn = insn_buf;
8278
8279         switch (si->off) {
8280         case offsetof(struct __sk_buff, data):
8281                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8282                                       si->dst_reg, si->src_reg,
8283                                       offsetof(struct bpf_flow_dissector, data));
8284                 break;
8285
8286         case offsetof(struct __sk_buff, data_end):
8287                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8288                                       si->dst_reg, si->src_reg,
8289                                       offsetof(struct bpf_flow_dissector, data_end));
8290                 break;
8291
8292         case offsetof(struct __sk_buff, flow_keys):
8293                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8294                                       si->dst_reg, si->src_reg,
8295                                       offsetof(struct bpf_flow_dissector, flow_keys));
8296                 break;
8297         }
8298
8299         return insn - insn_buf;
8300 }
8301
8302 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8303                                                   struct bpf_insn *insn)
8304 {
8305         /* si->dst_reg = skb_shinfo(SKB); */
8306 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8307         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8308                               BPF_REG_AX, si->src_reg,
8309                               offsetof(struct sk_buff, end));
8310         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8311                               si->dst_reg, si->src_reg,
8312                               offsetof(struct sk_buff, head));
8313         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8314 #else
8315         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8316                               si->dst_reg, si->src_reg,
8317                               offsetof(struct sk_buff, end));
8318 #endif
8319
8320         return insn;
8321 }
8322
8323 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8324                                   const struct bpf_insn *si,
8325                                   struct bpf_insn *insn_buf,
8326                                   struct bpf_prog *prog, u32 *target_size)
8327 {
8328         struct bpf_insn *insn = insn_buf;
8329         int off;
8330
8331         switch (si->off) {
8332         case offsetof(struct __sk_buff, len):
8333                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8334                                       bpf_target_off(struct sk_buff, len, 4,
8335                                                      target_size));
8336                 break;
8337
8338         case offsetof(struct __sk_buff, protocol):
8339                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8340                                       bpf_target_off(struct sk_buff, protocol, 2,
8341                                                      target_size));
8342                 break;
8343
8344         case offsetof(struct __sk_buff, vlan_proto):
8345                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8346                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8347                                                      target_size));
8348                 break;
8349
8350         case offsetof(struct __sk_buff, priority):
8351                 if (type == BPF_WRITE)
8352                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8353                                               bpf_target_off(struct sk_buff, priority, 4,
8354                                                              target_size));
8355                 else
8356                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8357                                               bpf_target_off(struct sk_buff, priority, 4,
8358                                                              target_size));
8359                 break;
8360
8361         case offsetof(struct __sk_buff, ingress_ifindex):
8362                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8363                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8364                                                      target_size));
8365                 break;
8366
8367         case offsetof(struct __sk_buff, ifindex):
8368                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8369                                       si->dst_reg, si->src_reg,
8370                                       offsetof(struct sk_buff, dev));
8371                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8372                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8373                                       bpf_target_off(struct net_device, ifindex, 4,
8374                                                      target_size));
8375                 break;
8376
8377         case offsetof(struct __sk_buff, hash):
8378                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8379                                       bpf_target_off(struct sk_buff, hash, 4,
8380                                                      target_size));
8381                 break;
8382
8383         case offsetof(struct __sk_buff, mark):
8384                 if (type == BPF_WRITE)
8385                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8386                                               bpf_target_off(struct sk_buff, mark, 4,
8387                                                              target_size));
8388                 else
8389                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8390                                               bpf_target_off(struct sk_buff, mark, 4,
8391                                                              target_size));
8392                 break;
8393
8394         case offsetof(struct __sk_buff, pkt_type):
8395                 *target_size = 1;
8396                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8397                                       PKT_TYPE_OFFSET());
8398                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8399 #ifdef __BIG_ENDIAN_BITFIELD
8400                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8401 #endif
8402                 break;
8403
8404         case offsetof(struct __sk_buff, queue_mapping):
8405                 if (type == BPF_WRITE) {
8406                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8407                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8408                                               bpf_target_off(struct sk_buff,
8409                                                              queue_mapping,
8410                                                              2, target_size));
8411                 } else {
8412                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8413                                               bpf_target_off(struct sk_buff,
8414                                                              queue_mapping,
8415                                                              2, target_size));
8416                 }
8417                 break;
8418
8419         case offsetof(struct __sk_buff, vlan_present):
8420                 *target_size = 1;
8421                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8422                                       PKT_VLAN_PRESENT_OFFSET());
8423                 if (PKT_VLAN_PRESENT_BIT)
8424                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8425                 if (PKT_VLAN_PRESENT_BIT < 7)
8426                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8427                 break;
8428
8429         case offsetof(struct __sk_buff, vlan_tci):
8430                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8431                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8432                                                      target_size));
8433                 break;
8434
8435         case offsetof(struct __sk_buff, cb[0]) ...
8436              offsetofend(struct __sk_buff, cb[4]) - 1:
8437                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8438                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8439                               offsetof(struct qdisc_skb_cb, data)) %
8440                              sizeof(__u64));
8441
8442                 prog->cb_access = 1;
8443                 off  = si->off;
8444                 off -= offsetof(struct __sk_buff, cb[0]);
8445                 off += offsetof(struct sk_buff, cb);
8446                 off += offsetof(struct qdisc_skb_cb, data);
8447                 if (type == BPF_WRITE)
8448                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8449                                               si->src_reg, off);
8450                 else
8451                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8452                                               si->src_reg, off);
8453                 break;
8454
8455         case offsetof(struct __sk_buff, tc_classid):
8456                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8457
8458                 off  = si->off;
8459                 off -= offsetof(struct __sk_buff, tc_classid);
8460                 off += offsetof(struct sk_buff, cb);
8461                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8462                 *target_size = 2;
8463                 if (type == BPF_WRITE)
8464                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8465                                               si->src_reg, off);
8466                 else
8467                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8468                                               si->src_reg, off);
8469                 break;
8470
8471         case offsetof(struct __sk_buff, data):
8472                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8473                                       si->dst_reg, si->src_reg,
8474                                       offsetof(struct sk_buff, data));
8475                 break;
8476
8477         case offsetof(struct __sk_buff, data_meta):
8478                 off  = si->off;
8479                 off -= offsetof(struct __sk_buff, data_meta);
8480                 off += offsetof(struct sk_buff, cb);
8481                 off += offsetof(struct bpf_skb_data_end, data_meta);
8482                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8483                                       si->src_reg, off);
8484                 break;
8485
8486         case offsetof(struct __sk_buff, data_end):
8487                 off  = si->off;
8488                 off -= offsetof(struct __sk_buff, data_end);
8489                 off += offsetof(struct sk_buff, cb);
8490                 off += offsetof(struct bpf_skb_data_end, data_end);
8491                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8492                                       si->src_reg, off);
8493                 break;
8494
8495         case offsetof(struct __sk_buff, tc_index):
8496 #ifdef CONFIG_NET_SCHED
8497                 if (type == BPF_WRITE)
8498                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8499                                               bpf_target_off(struct sk_buff, tc_index, 2,
8500                                                              target_size));
8501                 else
8502                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8503                                               bpf_target_off(struct sk_buff, tc_index, 2,
8504                                                              target_size));
8505 #else
8506                 *target_size = 2;
8507                 if (type == BPF_WRITE)
8508                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8509                 else
8510                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8511 #endif
8512                 break;
8513
8514         case offsetof(struct __sk_buff, napi_id):
8515 #if defined(CONFIG_NET_RX_BUSY_POLL)
8516                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8517                                       bpf_target_off(struct sk_buff, napi_id, 4,
8518                                                      target_size));
8519                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8520                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8521 #else
8522                 *target_size = 4;
8523                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8524 #endif
8525                 break;
8526         case offsetof(struct __sk_buff, family):
8527                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8528
8529                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8530                                       si->dst_reg, si->src_reg,
8531                                       offsetof(struct sk_buff, sk));
8532                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8533                                       bpf_target_off(struct sock_common,
8534                                                      skc_family,
8535                                                      2, target_size));
8536                 break;
8537         case offsetof(struct __sk_buff, remote_ip4):
8538                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8539
8540                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8541                                       si->dst_reg, si->src_reg,
8542                                       offsetof(struct sk_buff, sk));
8543                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8544                                       bpf_target_off(struct sock_common,
8545                                                      skc_daddr,
8546                                                      4, target_size));
8547                 break;
8548         case offsetof(struct __sk_buff, local_ip4):
8549                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8550                                           skc_rcv_saddr) != 4);
8551
8552                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8553                                       si->dst_reg, si->src_reg,
8554                                       offsetof(struct sk_buff, sk));
8555                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8556                                       bpf_target_off(struct sock_common,
8557                                                      skc_rcv_saddr,
8558                                                      4, target_size));
8559                 break;
8560         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8561              offsetof(struct __sk_buff, remote_ip6[3]):
8562 #if IS_ENABLED(CONFIG_IPV6)
8563                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8564                                           skc_v6_daddr.s6_addr32[0]) != 4);
8565
8566                 off = si->off;
8567                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8568
8569                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8570                                       si->dst_reg, si->src_reg,
8571                                       offsetof(struct sk_buff, sk));
8572                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8573                                       offsetof(struct sock_common,
8574                                                skc_v6_daddr.s6_addr32[0]) +
8575                                       off);
8576 #else
8577                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8578 #endif
8579                 break;
8580         case offsetof(struct __sk_buff, local_ip6[0]) ...
8581              offsetof(struct __sk_buff, local_ip6[3]):
8582 #if IS_ENABLED(CONFIG_IPV6)
8583                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8584                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8585
8586                 off = si->off;
8587                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8588
8589                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8590                                       si->dst_reg, si->src_reg,
8591                                       offsetof(struct sk_buff, sk));
8592                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8593                                       offsetof(struct sock_common,
8594                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8595                                       off);
8596 #else
8597                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8598 #endif
8599                 break;
8600
8601         case offsetof(struct __sk_buff, remote_port):
8602                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8603
8604                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8605                                       si->dst_reg, si->src_reg,
8606                                       offsetof(struct sk_buff, sk));
8607                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8608                                       bpf_target_off(struct sock_common,
8609                                                      skc_dport,
8610                                                      2, target_size));
8611 #ifndef __BIG_ENDIAN_BITFIELD
8612                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8613 #endif
8614                 break;
8615
8616         case offsetof(struct __sk_buff, local_port):
8617                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8618
8619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8620                                       si->dst_reg, si->src_reg,
8621                                       offsetof(struct sk_buff, sk));
8622                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8623                                       bpf_target_off(struct sock_common,
8624                                                      skc_num, 2, target_size));
8625                 break;
8626
8627         case offsetof(struct __sk_buff, tstamp):
8628                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8629
8630                 if (type == BPF_WRITE)
8631                         *insn++ = BPF_STX_MEM(BPF_DW,
8632                                               si->dst_reg, si->src_reg,
8633                                               bpf_target_off(struct sk_buff,
8634                                                              tstamp, 8,
8635                                                              target_size));
8636                 else
8637                         *insn++ = BPF_LDX_MEM(BPF_DW,
8638                                               si->dst_reg, si->src_reg,
8639                                               bpf_target_off(struct sk_buff,
8640                                                              tstamp, 8,
8641                                                              target_size));
8642                 break;
8643
8644         case offsetof(struct __sk_buff, gso_segs):
8645                 insn = bpf_convert_shinfo_access(si, insn);
8646                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8647                                       si->dst_reg, si->dst_reg,
8648                                       bpf_target_off(struct skb_shared_info,
8649                                                      gso_segs, 2,
8650                                                      target_size));
8651                 break;
8652         case offsetof(struct __sk_buff, gso_size):
8653                 insn = bpf_convert_shinfo_access(si, insn);
8654                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8655                                       si->dst_reg, si->dst_reg,
8656                                       bpf_target_off(struct skb_shared_info,
8657                                                      gso_size, 2,
8658                                                      target_size));
8659                 break;
8660         case offsetof(struct __sk_buff, wire_len):
8661                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8662
8663                 off = si->off;
8664                 off -= offsetof(struct __sk_buff, wire_len);
8665                 off += offsetof(struct sk_buff, cb);
8666                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8667                 *target_size = 4;
8668                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8669                 break;
8670
8671         case offsetof(struct __sk_buff, sk):
8672                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8673                                       si->dst_reg, si->src_reg,
8674                                       offsetof(struct sk_buff, sk));
8675                 break;
8676         }
8677
8678         return insn - insn_buf;
8679 }
8680
8681 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8682                                 const struct bpf_insn *si,
8683                                 struct bpf_insn *insn_buf,
8684                                 struct bpf_prog *prog, u32 *target_size)
8685 {
8686         struct bpf_insn *insn = insn_buf;
8687         int off;
8688
8689         switch (si->off) {
8690         case offsetof(struct bpf_sock, bound_dev_if):
8691                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8692
8693                 if (type == BPF_WRITE)
8694                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8695                                         offsetof(struct sock, sk_bound_dev_if));
8696                 else
8697                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8698                                       offsetof(struct sock, sk_bound_dev_if));
8699                 break;
8700
8701         case offsetof(struct bpf_sock, mark):
8702                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8703
8704                 if (type == BPF_WRITE)
8705                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8706                                         offsetof(struct sock, sk_mark));
8707                 else
8708                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8709                                       offsetof(struct sock, sk_mark));
8710                 break;
8711
8712         case offsetof(struct bpf_sock, priority):
8713                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8714
8715                 if (type == BPF_WRITE)
8716                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8717                                         offsetof(struct sock, sk_priority));
8718                 else
8719                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8720                                       offsetof(struct sock, sk_priority));
8721                 break;
8722
8723         case offsetof(struct bpf_sock, family):
8724                 *insn++ = BPF_LDX_MEM(
8725                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8726                         si->dst_reg, si->src_reg,
8727                         bpf_target_off(struct sock_common,
8728                                        skc_family,
8729                                        sizeof_field(struct sock_common,
8730                                                     skc_family),
8731                                        target_size));
8732                 break;
8733
8734         case offsetof(struct bpf_sock, type):
8735                 *insn++ = BPF_LDX_MEM(
8736                         BPF_FIELD_SIZEOF(struct sock, sk_type),
8737                         si->dst_reg, si->src_reg,
8738                         bpf_target_off(struct sock, sk_type,
8739                                        sizeof_field(struct sock, sk_type),
8740                                        target_size));
8741                 break;
8742
8743         case offsetof(struct bpf_sock, protocol):
8744                 *insn++ = BPF_LDX_MEM(
8745                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8746                         si->dst_reg, si->src_reg,
8747                         bpf_target_off(struct sock, sk_protocol,
8748                                        sizeof_field(struct sock, sk_protocol),
8749                                        target_size));
8750                 break;
8751
8752         case offsetof(struct bpf_sock, src_ip4):
8753                 *insn++ = BPF_LDX_MEM(
8754                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8755                         bpf_target_off(struct sock_common, skc_rcv_saddr,
8756                                        sizeof_field(struct sock_common,
8757                                                     skc_rcv_saddr),
8758                                        target_size));
8759                 break;
8760
8761         case offsetof(struct bpf_sock, dst_ip4):
8762                 *insn++ = BPF_LDX_MEM(
8763                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8764                         bpf_target_off(struct sock_common, skc_daddr,
8765                                        sizeof_field(struct sock_common,
8766                                                     skc_daddr),
8767                                        target_size));
8768                 break;
8769
8770         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8771 #if IS_ENABLED(CONFIG_IPV6)
8772                 off = si->off;
8773                 off -= offsetof(struct bpf_sock, src_ip6[0]);
8774                 *insn++ = BPF_LDX_MEM(
8775                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8776                         bpf_target_off(
8777                                 struct sock_common,
8778                                 skc_v6_rcv_saddr.s6_addr32[0],
8779                                 sizeof_field(struct sock_common,
8780                                              skc_v6_rcv_saddr.s6_addr32[0]),
8781                                 target_size) + off);
8782 #else
8783                 (void)off;
8784                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8785 #endif
8786                 break;
8787
8788         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8789 #if IS_ENABLED(CONFIG_IPV6)
8790                 off = si->off;
8791                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8792                 *insn++ = BPF_LDX_MEM(
8793                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8794                         bpf_target_off(struct sock_common,
8795                                        skc_v6_daddr.s6_addr32[0],
8796                                        sizeof_field(struct sock_common,
8797                                                     skc_v6_daddr.s6_addr32[0]),
8798                                        target_size) + off);
8799 #else
8800                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8801                 *target_size = 4;
8802 #endif
8803                 break;
8804
8805         case offsetof(struct bpf_sock, src_port):
8806                 *insn++ = BPF_LDX_MEM(
8807                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8808                         si->dst_reg, si->src_reg,
8809                         bpf_target_off(struct sock_common, skc_num,
8810                                        sizeof_field(struct sock_common,
8811                                                     skc_num),
8812                                        target_size));
8813                 break;
8814
8815         case offsetof(struct bpf_sock, dst_port):
8816                 *insn++ = BPF_LDX_MEM(
8817                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8818                         si->dst_reg, si->src_reg,
8819                         bpf_target_off(struct sock_common, skc_dport,
8820                                        sizeof_field(struct sock_common,
8821                                                     skc_dport),
8822                                        target_size));
8823                 break;
8824
8825         case offsetof(struct bpf_sock, state):
8826                 *insn++ = BPF_LDX_MEM(
8827                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8828                         si->dst_reg, si->src_reg,
8829                         bpf_target_off(struct sock_common, skc_state,
8830                                        sizeof_field(struct sock_common,
8831                                                     skc_state),
8832                                        target_size));
8833                 break;
8834         case offsetof(struct bpf_sock, rx_queue_mapping):
8835 #ifdef CONFIG_XPS
8836                 *insn++ = BPF_LDX_MEM(
8837                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8838                         si->dst_reg, si->src_reg,
8839                         bpf_target_off(struct sock, sk_rx_queue_mapping,
8840                                        sizeof_field(struct sock,
8841                                                     sk_rx_queue_mapping),
8842                                        target_size));
8843                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8844                                       1);
8845                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8846 #else
8847                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8848                 *target_size = 2;
8849 #endif
8850                 break;
8851         }
8852
8853         return insn - insn_buf;
8854 }
8855
8856 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8857                                          const struct bpf_insn *si,
8858                                          struct bpf_insn *insn_buf,
8859                                          struct bpf_prog *prog, u32 *target_size)
8860 {
8861         struct bpf_insn *insn = insn_buf;
8862
8863         switch (si->off) {
8864         case offsetof(struct __sk_buff, ifindex):
8865                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8866                                       si->dst_reg, si->src_reg,
8867                                       offsetof(struct sk_buff, dev));
8868                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8869                                       bpf_target_off(struct net_device, ifindex, 4,
8870                                                      target_size));
8871                 break;
8872         default:
8873                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8874                                               target_size);
8875         }
8876
8877         return insn - insn_buf;
8878 }
8879
8880 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8881                                   const struct bpf_insn *si,
8882                                   struct bpf_insn *insn_buf,
8883                                   struct bpf_prog *prog, u32 *target_size)
8884 {
8885         struct bpf_insn *insn = insn_buf;
8886
8887         switch (si->off) {
8888         case offsetof(struct xdp_md, data):
8889                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8890                                       si->dst_reg, si->src_reg,
8891                                       offsetof(struct xdp_buff, data));
8892                 break;
8893         case offsetof(struct xdp_md, data_meta):
8894                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8895                                       si->dst_reg, si->src_reg,
8896                                       offsetof(struct xdp_buff, data_meta));
8897                 break;
8898         case offsetof(struct xdp_md, data_end):
8899                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8900                                       si->dst_reg, si->src_reg,
8901                                       offsetof(struct xdp_buff, data_end));
8902                 break;
8903         case offsetof(struct xdp_md, ingress_ifindex):
8904                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8905                                       si->dst_reg, si->src_reg,
8906                                       offsetof(struct xdp_buff, rxq));
8907                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
8908                                       si->dst_reg, si->dst_reg,
8909                                       offsetof(struct xdp_rxq_info, dev));
8910                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8911                                       offsetof(struct net_device, ifindex));
8912                 break;
8913         case offsetof(struct xdp_md, rx_queue_index):
8914                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8915                                       si->dst_reg, si->src_reg,
8916                                       offsetof(struct xdp_buff, rxq));
8917                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8918                                       offsetof(struct xdp_rxq_info,
8919                                                queue_index));
8920                 break;
8921         case offsetof(struct xdp_md, egress_ifindex):
8922                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
8923                                       si->dst_reg, si->src_reg,
8924                                       offsetof(struct xdp_buff, txq));
8925                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
8926                                       si->dst_reg, si->dst_reg,
8927                                       offsetof(struct xdp_txq_info, dev));
8928                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8929                                       offsetof(struct net_device, ifindex));
8930                 break;
8931         }
8932
8933         return insn - insn_buf;
8934 }
8935
8936 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
8937  * context Structure, F is Field in context structure that contains a pointer
8938  * to Nested Structure of type NS that has the field NF.
8939  *
8940  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
8941  * sure that SIZE is not greater than actual size of S.F.NF.
8942  *
8943  * If offset OFF is provided, the load happens from that offset relative to
8944  * offset of NF.
8945  */
8946 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
8947         do {                                                                   \
8948                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
8949                                       si->src_reg, offsetof(S, F));            \
8950                 *insn++ = BPF_LDX_MEM(                                         \
8951                         SIZE, si->dst_reg, si->dst_reg,                        \
8952                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
8953                                        target_size)                            \
8954                                 + OFF);                                        \
8955         } while (0)
8956
8957 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
8958         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
8959                                              BPF_FIELD_SIZEOF(NS, NF), 0)
8960
8961 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
8962  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
8963  *
8964  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
8965  * "register" since two registers available in convert_ctx_access are not
8966  * enough: we can't override neither SRC, since it contains value to store, nor
8967  * DST since it contains pointer to context that may be used by later
8968  * instructions. But we need a temporary place to save pointer to nested
8969  * structure whose field we want to store to.
8970  */
8971 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
8972         do {                                                                   \
8973                 int tmp_reg = BPF_REG_9;                                       \
8974                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
8975                         --tmp_reg;                                             \
8976                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
8977                         --tmp_reg;                                             \
8978                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
8979                                       offsetof(S, TF));                        \
8980                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
8981                                       si->dst_reg, offsetof(S, F));            \
8982                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
8983                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
8984                                        target_size)                            \
8985                                 + OFF);                                        \
8986                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
8987                                       offsetof(S, TF));                        \
8988         } while (0)
8989
8990 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
8991                                                       TF)                      \
8992         do {                                                                   \
8993                 if (type == BPF_WRITE) {                                       \
8994                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
8995                                                          OFF, TF);             \
8996                 } else {                                                       \
8997                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
8998                                 S, NS, F, NF, SIZE, OFF);  \
8999                 }                                                              \
9000         } while (0)
9001
9002 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9003         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9004                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9005
9006 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9007                                         const struct bpf_insn *si,
9008                                         struct bpf_insn *insn_buf,
9009                                         struct bpf_prog *prog, u32 *target_size)
9010 {
9011         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9012         struct bpf_insn *insn = insn_buf;
9013
9014         switch (si->off) {
9015         case offsetof(struct bpf_sock_addr, user_family):
9016                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9017                                             struct sockaddr, uaddr, sa_family);
9018                 break;
9019
9020         case offsetof(struct bpf_sock_addr, user_ip4):
9021                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9022                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9023                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9024                 break;
9025
9026         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9027                 off = si->off;
9028                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9029                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9030                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9031                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9032                         tmp_reg);
9033                 break;
9034
9035         case offsetof(struct bpf_sock_addr, user_port):
9036                 /* To get port we need to know sa_family first and then treat
9037                  * sockaddr as either sockaddr_in or sockaddr_in6.
9038                  * Though we can simplify since port field has same offset and
9039                  * size in both structures.
9040                  * Here we check this invariant and use just one of the
9041                  * structures if it's true.
9042                  */
9043                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9044                              offsetof(struct sockaddr_in6, sin6_port));
9045                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9046                              sizeof_field(struct sockaddr_in6, sin6_port));
9047                 /* Account for sin6_port being smaller than user_port. */
9048                 port_size = min(port_size, BPF_LDST_BYTES(si));
9049                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9050                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9051                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9052                 break;
9053
9054         case offsetof(struct bpf_sock_addr, family):
9055                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9056                                             struct sock, sk, sk_family);
9057                 break;
9058
9059         case offsetof(struct bpf_sock_addr, type):
9060                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9061                                             struct sock, sk, sk_type);
9062                 break;
9063
9064         case offsetof(struct bpf_sock_addr, protocol):
9065                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9066                                             struct sock, sk, sk_protocol);
9067                 break;
9068
9069         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9070                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9071                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9072                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9073                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9074                 break;
9075
9076         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9077                                 msg_src_ip6[3]):
9078                 off = si->off;
9079                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9080                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9081                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9082                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9083                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9084                 break;
9085         case offsetof(struct bpf_sock_addr, sk):
9086                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9087                                       si->dst_reg, si->src_reg,
9088                                       offsetof(struct bpf_sock_addr_kern, sk));
9089                 break;
9090         }
9091
9092         return insn - insn_buf;
9093 }
9094
9095 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9096                                        const struct bpf_insn *si,
9097                                        struct bpf_insn *insn_buf,
9098                                        struct bpf_prog *prog,
9099                                        u32 *target_size)
9100 {
9101         struct bpf_insn *insn = insn_buf;
9102         int off;
9103
9104 /* Helper macro for adding read access to tcp_sock or sock fields. */
9105 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9106         do {                                                                  \
9107                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9108                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9109                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9110                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9111                         reg--;                                                \
9112                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9113                         reg--;                                                \
9114                 if (si->dst_reg == si->src_reg) {                             \
9115                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9116                                           offsetof(struct bpf_sock_ops_kern,  \
9117                                           temp));                             \
9118                         fullsock_reg = reg;                                   \
9119                         jmp += 2;                                             \
9120                 }                                                             \
9121                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9122                                                 struct bpf_sock_ops_kern,     \
9123                                                 is_fullsock),                 \
9124                                       fullsock_reg, si->src_reg,              \
9125                                       offsetof(struct bpf_sock_ops_kern,      \
9126                                                is_fullsock));                 \
9127                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9128                 if (si->dst_reg == si->src_reg)                               \
9129                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9130                                       offsetof(struct bpf_sock_ops_kern,      \
9131                                       temp));                                 \
9132                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9133                                                 struct bpf_sock_ops_kern, sk),\
9134                                       si->dst_reg, si->src_reg,               \
9135                                       offsetof(struct bpf_sock_ops_kern, sk));\
9136                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9137                                                        OBJ_FIELD),            \
9138                                       si->dst_reg, si->dst_reg,               \
9139                                       offsetof(OBJ, OBJ_FIELD));              \
9140                 if (si->dst_reg == si->src_reg) {                             \
9141                         *insn++ = BPF_JMP_A(1);                               \
9142                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9143                                       offsetof(struct bpf_sock_ops_kern,      \
9144                                       temp));                                 \
9145                 }                                                             \
9146         } while (0)
9147
9148 #define SOCK_OPS_GET_SK()                                                             \
9149         do {                                                                  \
9150                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9151                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9152                         reg--;                                                \
9153                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9154                         reg--;                                                \
9155                 if (si->dst_reg == si->src_reg) {                             \
9156                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9157                                           offsetof(struct bpf_sock_ops_kern,  \
9158                                           temp));                             \
9159                         fullsock_reg = reg;                                   \
9160                         jmp += 2;                                             \
9161                 }                                                             \
9162                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9163                                                 struct bpf_sock_ops_kern,     \
9164                                                 is_fullsock),                 \
9165                                       fullsock_reg, si->src_reg,              \
9166                                       offsetof(struct bpf_sock_ops_kern,      \
9167                                                is_fullsock));                 \
9168                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9169                 if (si->dst_reg == si->src_reg)                               \
9170                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9171                                       offsetof(struct bpf_sock_ops_kern,      \
9172                                       temp));                                 \
9173                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9174                                                 struct bpf_sock_ops_kern, sk),\
9175                                       si->dst_reg, si->src_reg,               \
9176                                       offsetof(struct bpf_sock_ops_kern, sk));\
9177                 if (si->dst_reg == si->src_reg) {                             \
9178                         *insn++ = BPF_JMP_A(1);                               \
9179                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9180                                       offsetof(struct bpf_sock_ops_kern,      \
9181                                       temp));                                 \
9182                 }                                                             \
9183         } while (0)
9184
9185 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9186                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9187
9188 /* Helper macro for adding write access to tcp_sock or sock fields.
9189  * The macro is called with two registers, dst_reg which contains a pointer
9190  * to ctx (context) and src_reg which contains the value that should be
9191  * stored. However, we need an additional register since we cannot overwrite
9192  * dst_reg because it may be used later in the program.
9193  * Instead we "borrow" one of the other register. We first save its value
9194  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9195  * it at the end of the macro.
9196  */
9197 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9198         do {                                                                  \
9199                 int reg = BPF_REG_9;                                          \
9200                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9201                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9202                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9203                         reg--;                                                \
9204                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9205                         reg--;                                                \
9206                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9207                                       offsetof(struct bpf_sock_ops_kern,      \
9208                                                temp));                        \
9209                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9210                                                 struct bpf_sock_ops_kern,     \
9211                                                 is_fullsock),                 \
9212                                       reg, si->dst_reg,                       \
9213                                       offsetof(struct bpf_sock_ops_kern,      \
9214                                                is_fullsock));                 \
9215                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9216                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9217                                                 struct bpf_sock_ops_kern, sk),\
9218                                       reg, si->dst_reg,                       \
9219                                       offsetof(struct bpf_sock_ops_kern, sk));\
9220                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9221                                       reg, si->src_reg,                       \
9222                                       offsetof(OBJ, OBJ_FIELD));              \
9223                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9224                                       offsetof(struct bpf_sock_ops_kern,      \
9225                                                temp));                        \
9226         } while (0)
9227
9228 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9229         do {                                                                  \
9230                 if (TYPE == BPF_WRITE)                                        \
9231                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9232                 else                                                          \
9233                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9234         } while (0)
9235
9236         if (insn > insn_buf)
9237                 return insn - insn_buf;
9238
9239         switch (si->off) {
9240         case offsetof(struct bpf_sock_ops, op):
9241                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9242                                                        op),
9243                                       si->dst_reg, si->src_reg,
9244                                       offsetof(struct bpf_sock_ops_kern, op));
9245                 break;
9246
9247         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9248              offsetof(struct bpf_sock_ops, replylong[3]):
9249                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9250                              sizeof_field(struct bpf_sock_ops_kern, reply));
9251                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9252                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9253                 off = si->off;
9254                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9255                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9256                 if (type == BPF_WRITE)
9257                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9258                                               off);
9259                 else
9260                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9261                                               off);
9262                 break;
9263
9264         case offsetof(struct bpf_sock_ops, family):
9265                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9266
9267                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9268                                               struct bpf_sock_ops_kern, sk),
9269                                       si->dst_reg, si->src_reg,
9270                                       offsetof(struct bpf_sock_ops_kern, sk));
9271                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9272                                       offsetof(struct sock_common, skc_family));
9273                 break;
9274
9275         case offsetof(struct bpf_sock_ops, remote_ip4):
9276                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9277
9278                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9279                                                 struct bpf_sock_ops_kern, sk),
9280                                       si->dst_reg, si->src_reg,
9281                                       offsetof(struct bpf_sock_ops_kern, sk));
9282                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9283                                       offsetof(struct sock_common, skc_daddr));
9284                 break;
9285
9286         case offsetof(struct bpf_sock_ops, local_ip4):
9287                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9288                                           skc_rcv_saddr) != 4);
9289
9290                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9291                                               struct bpf_sock_ops_kern, sk),
9292                                       si->dst_reg, si->src_reg,
9293                                       offsetof(struct bpf_sock_ops_kern, sk));
9294                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9295                                       offsetof(struct sock_common,
9296                                                skc_rcv_saddr));
9297                 break;
9298
9299         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9300              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9301 #if IS_ENABLED(CONFIG_IPV6)
9302                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9303                                           skc_v6_daddr.s6_addr32[0]) != 4);
9304
9305                 off = si->off;
9306                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9307                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9308                                                 struct bpf_sock_ops_kern, sk),
9309                                       si->dst_reg, si->src_reg,
9310                                       offsetof(struct bpf_sock_ops_kern, sk));
9311                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9312                                       offsetof(struct sock_common,
9313                                                skc_v6_daddr.s6_addr32[0]) +
9314                                       off);
9315 #else
9316                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9317 #endif
9318                 break;
9319
9320         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9321              offsetof(struct bpf_sock_ops, local_ip6[3]):
9322 #if IS_ENABLED(CONFIG_IPV6)
9323                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9324                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9325
9326                 off = si->off;
9327                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9328                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9329                                                 struct bpf_sock_ops_kern, sk),
9330                                       si->dst_reg, si->src_reg,
9331                                       offsetof(struct bpf_sock_ops_kern, sk));
9332                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9333                                       offsetof(struct sock_common,
9334                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9335                                       off);
9336 #else
9337                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9338 #endif
9339                 break;
9340
9341         case offsetof(struct bpf_sock_ops, remote_port):
9342                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9343
9344                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9345                                                 struct bpf_sock_ops_kern, sk),
9346                                       si->dst_reg, si->src_reg,
9347                                       offsetof(struct bpf_sock_ops_kern, sk));
9348                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9349                                       offsetof(struct sock_common, skc_dport));
9350 #ifndef __BIG_ENDIAN_BITFIELD
9351                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9352 #endif
9353                 break;
9354
9355         case offsetof(struct bpf_sock_ops, local_port):
9356                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 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_num));
9364                 break;
9365
9366         case offsetof(struct bpf_sock_ops, is_fullsock):
9367                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9368                                                 struct bpf_sock_ops_kern,
9369                                                 is_fullsock),
9370                                       si->dst_reg, si->src_reg,
9371                                       offsetof(struct bpf_sock_ops_kern,
9372                                                is_fullsock));
9373                 break;
9374
9375         case offsetof(struct bpf_sock_ops, state):
9376                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9377
9378                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9379                                                 struct bpf_sock_ops_kern, sk),
9380                                       si->dst_reg, si->src_reg,
9381                                       offsetof(struct bpf_sock_ops_kern, sk));
9382                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9383                                       offsetof(struct sock_common, skc_state));
9384                 break;
9385
9386         case offsetof(struct bpf_sock_ops, rtt_min):
9387                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9388                              sizeof(struct minmax));
9389                 BUILD_BUG_ON(sizeof(struct minmax) <
9390                              sizeof(struct minmax_sample));
9391
9392                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9393                                                 struct bpf_sock_ops_kern, sk),
9394                                       si->dst_reg, si->src_reg,
9395                                       offsetof(struct bpf_sock_ops_kern, sk));
9396                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9397                                       offsetof(struct tcp_sock, rtt_min) +
9398                                       sizeof_field(struct minmax_sample, t));
9399                 break;
9400
9401         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9402                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9403                                    struct tcp_sock);
9404                 break;
9405
9406         case offsetof(struct bpf_sock_ops, sk_txhash):
9407                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9408                                           struct sock, type);
9409                 break;
9410         case offsetof(struct bpf_sock_ops, snd_cwnd):
9411                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9412                 break;
9413         case offsetof(struct bpf_sock_ops, srtt_us):
9414                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9415                 break;
9416         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9417                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9418                 break;
9419         case offsetof(struct bpf_sock_ops, rcv_nxt):
9420                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9421                 break;
9422         case offsetof(struct bpf_sock_ops, snd_nxt):
9423                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9424                 break;
9425         case offsetof(struct bpf_sock_ops, snd_una):
9426                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9427                 break;
9428         case offsetof(struct bpf_sock_ops, mss_cache):
9429                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9430                 break;
9431         case offsetof(struct bpf_sock_ops, ecn_flags):
9432                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9433                 break;
9434         case offsetof(struct bpf_sock_ops, rate_delivered):
9435                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9436                 break;
9437         case offsetof(struct bpf_sock_ops, rate_interval_us):
9438                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9439                 break;
9440         case offsetof(struct bpf_sock_ops, packets_out):
9441                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9442                 break;
9443         case offsetof(struct bpf_sock_ops, retrans_out):
9444                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9445                 break;
9446         case offsetof(struct bpf_sock_ops, total_retrans):
9447                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9448                 break;
9449         case offsetof(struct bpf_sock_ops, segs_in):
9450                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9451                 break;
9452         case offsetof(struct bpf_sock_ops, data_segs_in):
9453                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9454                 break;
9455         case offsetof(struct bpf_sock_ops, segs_out):
9456                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9457                 break;
9458         case offsetof(struct bpf_sock_ops, data_segs_out):
9459                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9460                 break;
9461         case offsetof(struct bpf_sock_ops, lost_out):
9462                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9463                 break;
9464         case offsetof(struct bpf_sock_ops, sacked_out):
9465                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9466                 break;
9467         case offsetof(struct bpf_sock_ops, bytes_received):
9468                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9469                 break;
9470         case offsetof(struct bpf_sock_ops, bytes_acked):
9471                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9472                 break;
9473         case offsetof(struct bpf_sock_ops, sk):
9474                 SOCK_OPS_GET_SK();
9475                 break;
9476         case offsetof(struct bpf_sock_ops, skb_data_end):
9477                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9478                                                        skb_data_end),
9479                                       si->dst_reg, si->src_reg,
9480                                       offsetof(struct bpf_sock_ops_kern,
9481                                                skb_data_end));
9482                 break;
9483         case offsetof(struct bpf_sock_ops, skb_data):
9484                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9485                                                        skb),
9486                                       si->dst_reg, si->src_reg,
9487                                       offsetof(struct bpf_sock_ops_kern,
9488                                                skb));
9489                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9490                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9491                                       si->dst_reg, si->dst_reg,
9492                                       offsetof(struct sk_buff, data));
9493                 break;
9494         case offsetof(struct bpf_sock_ops, skb_len):
9495                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9496                                                        skb),
9497                                       si->dst_reg, si->src_reg,
9498                                       offsetof(struct bpf_sock_ops_kern,
9499                                                skb));
9500                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9501                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9502                                       si->dst_reg, si->dst_reg,
9503                                       offsetof(struct sk_buff, len));
9504                 break;
9505         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9506                 off = offsetof(struct sk_buff, cb);
9507                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9508                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9509                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9510                                                        skb),
9511                                       si->dst_reg, si->src_reg,
9512                                       offsetof(struct bpf_sock_ops_kern,
9513                                                skb));
9514                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9515                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9516                                                        tcp_flags),
9517                                       si->dst_reg, si->dst_reg, off);
9518                 break;
9519         }
9520         return insn - insn_buf;
9521 }
9522
9523 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9524                                      const struct bpf_insn *si,
9525                                      struct bpf_insn *insn_buf,
9526                                      struct bpf_prog *prog, u32 *target_size)
9527 {
9528         struct bpf_insn *insn = insn_buf;
9529         int off;
9530
9531         switch (si->off) {
9532         case offsetof(struct __sk_buff, data_end):
9533                 off  = si->off;
9534                 off -= offsetof(struct __sk_buff, data_end);
9535                 off += offsetof(struct sk_buff, cb);
9536                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
9537                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9538                                       si->src_reg, off);
9539                 break;
9540         case offsetof(struct __sk_buff, cb[0]) ...
9541              offsetofend(struct __sk_buff, cb[4]) - 1:
9542                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
9543                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9544                               offsetof(struct sk_skb_cb, data)) %
9545                              sizeof(__u64));
9546
9547                 prog->cb_access = 1;
9548                 off  = si->off;
9549                 off -= offsetof(struct __sk_buff, cb[0]);
9550                 off += offsetof(struct sk_buff, cb);
9551                 off += offsetof(struct sk_skb_cb, data);
9552                 if (type == BPF_WRITE)
9553                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9554                                               si->src_reg, off);
9555                 else
9556                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9557                                               si->src_reg, off);
9558                 break;
9559
9560
9561         default:
9562                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9563                                               target_size);
9564         }
9565
9566         return insn - insn_buf;
9567 }
9568
9569 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9570                                      const struct bpf_insn *si,
9571                                      struct bpf_insn *insn_buf,
9572                                      struct bpf_prog *prog, u32 *target_size)
9573 {
9574         struct bpf_insn *insn = insn_buf;
9575 #if IS_ENABLED(CONFIG_IPV6)
9576         int off;
9577 #endif
9578
9579         /* convert ctx uses the fact sg element is first in struct */
9580         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9581
9582         switch (si->off) {
9583         case offsetof(struct sk_msg_md, data):
9584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9585                                       si->dst_reg, si->src_reg,
9586                                       offsetof(struct sk_msg, data));
9587                 break;
9588         case offsetof(struct sk_msg_md, data_end):
9589                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9590                                       si->dst_reg, si->src_reg,
9591                                       offsetof(struct sk_msg, data_end));
9592                 break;
9593         case offsetof(struct sk_msg_md, family):
9594                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9595
9596                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9597                                               struct sk_msg, sk),
9598                                       si->dst_reg, si->src_reg,
9599                                       offsetof(struct sk_msg, sk));
9600                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9601                                       offsetof(struct sock_common, skc_family));
9602                 break;
9603
9604         case offsetof(struct sk_msg_md, remote_ip4):
9605                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9606
9607                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9608                                                 struct sk_msg, sk),
9609                                       si->dst_reg, si->src_reg,
9610                                       offsetof(struct sk_msg, sk));
9611                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9612                                       offsetof(struct sock_common, skc_daddr));
9613                 break;
9614
9615         case offsetof(struct sk_msg_md, local_ip4):
9616                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9617                                           skc_rcv_saddr) != 4);
9618
9619                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9620                                               struct sk_msg, sk),
9621                                       si->dst_reg, si->src_reg,
9622                                       offsetof(struct sk_msg, sk));
9623                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9624                                       offsetof(struct sock_common,
9625                                                skc_rcv_saddr));
9626                 break;
9627
9628         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9629              offsetof(struct sk_msg_md, remote_ip6[3]):
9630 #if IS_ENABLED(CONFIG_IPV6)
9631                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9632                                           skc_v6_daddr.s6_addr32[0]) != 4);
9633
9634                 off = si->off;
9635                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9636                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9637                                                 struct sk_msg, sk),
9638                                       si->dst_reg, si->src_reg,
9639                                       offsetof(struct sk_msg, sk));
9640                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9641                                       offsetof(struct sock_common,
9642                                                skc_v6_daddr.s6_addr32[0]) +
9643                                       off);
9644 #else
9645                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9646 #endif
9647                 break;
9648
9649         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9650              offsetof(struct sk_msg_md, local_ip6[3]):
9651 #if IS_ENABLED(CONFIG_IPV6)
9652                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9653                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9654
9655                 off = si->off;
9656                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9657                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9658                                                 struct sk_msg, sk),
9659                                       si->dst_reg, si->src_reg,
9660                                       offsetof(struct sk_msg, sk));
9661                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9662                                       offsetof(struct sock_common,
9663                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9664                                       off);
9665 #else
9666                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9667 #endif
9668                 break;
9669
9670         case offsetof(struct sk_msg_md, remote_port):
9671                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9672
9673                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9674                                                 struct sk_msg, sk),
9675                                       si->dst_reg, si->src_reg,
9676                                       offsetof(struct sk_msg, sk));
9677                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9678                                       offsetof(struct sock_common, skc_dport));
9679 #ifndef __BIG_ENDIAN_BITFIELD
9680                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9681 #endif
9682                 break;
9683
9684         case offsetof(struct sk_msg_md, local_port):
9685                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9686
9687                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9688                                                 struct sk_msg, sk),
9689                                       si->dst_reg, si->src_reg,
9690                                       offsetof(struct sk_msg, sk));
9691                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9692                                       offsetof(struct sock_common, skc_num));
9693                 break;
9694
9695         case offsetof(struct sk_msg_md, size):
9696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9697                                       si->dst_reg, si->src_reg,
9698                                       offsetof(struct sk_msg_sg, size));
9699                 break;
9700
9701         case offsetof(struct sk_msg_md, sk):
9702                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9703                                       si->dst_reg, si->src_reg,
9704                                       offsetof(struct sk_msg, sk));
9705                 break;
9706         }
9707
9708         return insn - insn_buf;
9709 }
9710
9711 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9712         .get_func_proto         = sk_filter_func_proto,
9713         .is_valid_access        = sk_filter_is_valid_access,
9714         .convert_ctx_access     = bpf_convert_ctx_access,
9715         .gen_ld_abs             = bpf_gen_ld_abs,
9716 };
9717
9718 const struct bpf_prog_ops sk_filter_prog_ops = {
9719         .test_run               = bpf_prog_test_run_skb,
9720 };
9721
9722 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9723         .get_func_proto         = tc_cls_act_func_proto,
9724         .is_valid_access        = tc_cls_act_is_valid_access,
9725         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
9726         .gen_prologue           = tc_cls_act_prologue,
9727         .gen_ld_abs             = bpf_gen_ld_abs,
9728 };
9729
9730 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9731         .test_run               = bpf_prog_test_run_skb,
9732 };
9733
9734 const struct bpf_verifier_ops xdp_verifier_ops = {
9735         .get_func_proto         = xdp_func_proto,
9736         .is_valid_access        = xdp_is_valid_access,
9737         .convert_ctx_access     = xdp_convert_ctx_access,
9738         .gen_prologue           = bpf_noop_prologue,
9739 };
9740
9741 const struct bpf_prog_ops xdp_prog_ops = {
9742         .test_run               = bpf_prog_test_run_xdp,
9743 };
9744
9745 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9746         .get_func_proto         = cg_skb_func_proto,
9747         .is_valid_access        = cg_skb_is_valid_access,
9748         .convert_ctx_access     = bpf_convert_ctx_access,
9749 };
9750
9751 const struct bpf_prog_ops cg_skb_prog_ops = {
9752         .test_run               = bpf_prog_test_run_skb,
9753 };
9754
9755 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9756         .get_func_proto         = lwt_in_func_proto,
9757         .is_valid_access        = lwt_is_valid_access,
9758         .convert_ctx_access     = bpf_convert_ctx_access,
9759 };
9760
9761 const struct bpf_prog_ops lwt_in_prog_ops = {
9762         .test_run               = bpf_prog_test_run_skb,
9763 };
9764
9765 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9766         .get_func_proto         = lwt_out_func_proto,
9767         .is_valid_access        = lwt_is_valid_access,
9768         .convert_ctx_access     = bpf_convert_ctx_access,
9769 };
9770
9771 const struct bpf_prog_ops lwt_out_prog_ops = {
9772         .test_run               = bpf_prog_test_run_skb,
9773 };
9774
9775 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9776         .get_func_proto         = lwt_xmit_func_proto,
9777         .is_valid_access        = lwt_is_valid_access,
9778         .convert_ctx_access     = bpf_convert_ctx_access,
9779         .gen_prologue           = tc_cls_act_prologue,
9780 };
9781
9782 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9783         .test_run               = bpf_prog_test_run_skb,
9784 };
9785
9786 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9787         .get_func_proto         = lwt_seg6local_func_proto,
9788         .is_valid_access        = lwt_is_valid_access,
9789         .convert_ctx_access     = bpf_convert_ctx_access,
9790 };
9791
9792 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9793         .test_run               = bpf_prog_test_run_skb,
9794 };
9795
9796 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9797         .get_func_proto         = sock_filter_func_proto,
9798         .is_valid_access        = sock_filter_is_valid_access,
9799         .convert_ctx_access     = bpf_sock_convert_ctx_access,
9800 };
9801
9802 const struct bpf_prog_ops cg_sock_prog_ops = {
9803 };
9804
9805 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9806         .get_func_proto         = sock_addr_func_proto,
9807         .is_valid_access        = sock_addr_is_valid_access,
9808         .convert_ctx_access     = sock_addr_convert_ctx_access,
9809 };
9810
9811 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9812 };
9813
9814 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9815         .get_func_proto         = sock_ops_func_proto,
9816         .is_valid_access        = sock_ops_is_valid_access,
9817         .convert_ctx_access     = sock_ops_convert_ctx_access,
9818 };
9819
9820 const struct bpf_prog_ops sock_ops_prog_ops = {
9821 };
9822
9823 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9824         .get_func_proto         = sk_skb_func_proto,
9825         .is_valid_access        = sk_skb_is_valid_access,
9826         .convert_ctx_access     = sk_skb_convert_ctx_access,
9827         .gen_prologue           = sk_skb_prologue,
9828 };
9829
9830 const struct bpf_prog_ops sk_skb_prog_ops = {
9831 };
9832
9833 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9834         .get_func_proto         = sk_msg_func_proto,
9835         .is_valid_access        = sk_msg_is_valid_access,
9836         .convert_ctx_access     = sk_msg_convert_ctx_access,
9837         .gen_prologue           = bpf_noop_prologue,
9838 };
9839
9840 const struct bpf_prog_ops sk_msg_prog_ops = {
9841 };
9842
9843 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9844         .get_func_proto         = flow_dissector_func_proto,
9845         .is_valid_access        = flow_dissector_is_valid_access,
9846         .convert_ctx_access     = flow_dissector_convert_ctx_access,
9847 };
9848
9849 const struct bpf_prog_ops flow_dissector_prog_ops = {
9850         .test_run               = bpf_prog_test_run_flow_dissector,
9851 };
9852
9853 int sk_detach_filter(struct sock *sk)
9854 {
9855         int ret = -ENOENT;
9856         struct sk_filter *filter;
9857
9858         if (sock_flag(sk, SOCK_FILTER_LOCKED))
9859                 return -EPERM;
9860
9861         filter = rcu_dereference_protected(sk->sk_filter,
9862                                            lockdep_sock_is_held(sk));
9863         if (filter) {
9864                 RCU_INIT_POINTER(sk->sk_filter, NULL);
9865                 sk_filter_uncharge(sk, filter);
9866                 ret = 0;
9867         }
9868
9869         return ret;
9870 }
9871 EXPORT_SYMBOL_GPL(sk_detach_filter);
9872
9873 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9874                   unsigned int len)
9875 {
9876         struct sock_fprog_kern *fprog;
9877         struct sk_filter *filter;
9878         int ret = 0;
9879
9880         lock_sock(sk);
9881         filter = rcu_dereference_protected(sk->sk_filter,
9882                                            lockdep_sock_is_held(sk));
9883         if (!filter)
9884                 goto out;
9885
9886         /* We're copying the filter that has been originally attached,
9887          * so no conversion/decode needed anymore. eBPF programs that
9888          * have no original program cannot be dumped through this.
9889          */
9890         ret = -EACCES;
9891         fprog = filter->prog->orig_prog;
9892         if (!fprog)
9893                 goto out;
9894
9895         ret = fprog->len;
9896         if (!len)
9897                 /* User space only enquires number of filter blocks. */
9898                 goto out;
9899
9900         ret = -EINVAL;
9901         if (len < fprog->len)
9902                 goto out;
9903
9904         ret = -EFAULT;
9905         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9906                 goto out;
9907
9908         /* Instead of bytes, the API requests to return the number
9909          * of filter blocks.
9910          */
9911         ret = fprog->len;
9912 out:
9913         release_sock(sk);
9914         return ret;
9915 }
9916
9917 #ifdef CONFIG_INET
9918 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
9919                                     struct sock_reuseport *reuse,
9920                                     struct sock *sk, struct sk_buff *skb,
9921                                     u32 hash)
9922 {
9923         reuse_kern->skb = skb;
9924         reuse_kern->sk = sk;
9925         reuse_kern->selected_sk = NULL;
9926         reuse_kern->data_end = skb->data + skb_headlen(skb);
9927         reuse_kern->hash = hash;
9928         reuse_kern->reuseport_id = reuse->reuseport_id;
9929         reuse_kern->bind_inany = reuse->bind_inany;
9930 }
9931
9932 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
9933                                   struct bpf_prog *prog, struct sk_buff *skb,
9934                                   u32 hash)
9935 {
9936         struct sk_reuseport_kern reuse_kern;
9937         enum sk_action action;
9938
9939         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
9940         action = BPF_PROG_RUN(prog, &reuse_kern);
9941
9942         if (action == SK_PASS)
9943                 return reuse_kern.selected_sk;
9944         else
9945                 return ERR_PTR(-ECONNREFUSED);
9946 }
9947
9948 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
9949            struct bpf_map *, map, void *, key, u32, flags)
9950 {
9951         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
9952         struct sock_reuseport *reuse;
9953         struct sock *selected_sk;
9954
9955         selected_sk = map->ops->map_lookup_elem(map, key);
9956         if (!selected_sk)
9957                 return -ENOENT;
9958
9959         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
9960         if (!reuse) {
9961                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
9962                 if (sk_is_refcounted(selected_sk))
9963                         sock_put(selected_sk);
9964
9965                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
9966                  * The only (!reuse) case here is - the sk has already been
9967                  * unhashed (e.g. by close()), so treat it as -ENOENT.
9968                  *
9969                  * Other maps (e.g. sock_map) do not provide this guarantee and
9970                  * the sk may never be in the reuseport group to begin with.
9971                  */
9972                 return is_sockarray ? -ENOENT : -EINVAL;
9973         }
9974
9975         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
9976                 struct sock *sk = reuse_kern->sk;
9977
9978                 if (sk->sk_protocol != selected_sk->sk_protocol)
9979                         return -EPROTOTYPE;
9980                 else if (sk->sk_family != selected_sk->sk_family)
9981                         return -EAFNOSUPPORT;
9982
9983                 /* Catch all. Likely bound to a different sockaddr. */
9984                 return -EBADFD;
9985         }
9986
9987         reuse_kern->selected_sk = selected_sk;
9988
9989         return 0;
9990 }
9991
9992 static const struct bpf_func_proto sk_select_reuseport_proto = {
9993         .func           = sk_select_reuseport,
9994         .gpl_only       = false,
9995         .ret_type       = RET_INTEGER,
9996         .arg1_type      = ARG_PTR_TO_CTX,
9997         .arg2_type      = ARG_CONST_MAP_PTR,
9998         .arg3_type      = ARG_PTR_TO_MAP_KEY,
9999         .arg4_type      = ARG_ANYTHING,
10000 };
10001
10002 BPF_CALL_4(sk_reuseport_load_bytes,
10003            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10004            void *, to, u32, len)
10005 {
10006         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10007 }
10008
10009 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10010         .func           = sk_reuseport_load_bytes,
10011         .gpl_only       = false,
10012         .ret_type       = RET_INTEGER,
10013         .arg1_type      = ARG_PTR_TO_CTX,
10014         .arg2_type      = ARG_ANYTHING,
10015         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10016         .arg4_type      = ARG_CONST_SIZE,
10017 };
10018
10019 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10020            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10021            void *, to, u32, len, u32, start_header)
10022 {
10023         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10024                                                len, start_header);
10025 }
10026
10027 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10028         .func           = sk_reuseport_load_bytes_relative,
10029         .gpl_only       = false,
10030         .ret_type       = RET_INTEGER,
10031         .arg1_type      = ARG_PTR_TO_CTX,
10032         .arg2_type      = ARG_ANYTHING,
10033         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10034         .arg4_type      = ARG_CONST_SIZE,
10035         .arg5_type      = ARG_ANYTHING,
10036 };
10037
10038 static const struct bpf_func_proto *
10039 sk_reuseport_func_proto(enum bpf_func_id func_id,
10040                         const struct bpf_prog *prog)
10041 {
10042         switch (func_id) {
10043         case BPF_FUNC_sk_select_reuseport:
10044                 return &sk_select_reuseport_proto;
10045         case BPF_FUNC_skb_load_bytes:
10046                 return &sk_reuseport_load_bytes_proto;
10047         case BPF_FUNC_skb_load_bytes_relative:
10048                 return &sk_reuseport_load_bytes_relative_proto;
10049         default:
10050                 return bpf_base_func_proto(func_id);
10051         }
10052 }
10053
10054 static bool
10055 sk_reuseport_is_valid_access(int off, int size,
10056                              enum bpf_access_type type,
10057                              const struct bpf_prog *prog,
10058                              struct bpf_insn_access_aux *info)
10059 {
10060         const u32 size_default = sizeof(__u32);
10061
10062         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10063             off % size || type != BPF_READ)
10064                 return false;
10065
10066         switch (off) {
10067         case offsetof(struct sk_reuseport_md, data):
10068                 info->reg_type = PTR_TO_PACKET;
10069                 return size == sizeof(__u64);
10070
10071         case offsetof(struct sk_reuseport_md, data_end):
10072                 info->reg_type = PTR_TO_PACKET_END;
10073                 return size == sizeof(__u64);
10074
10075         case offsetof(struct sk_reuseport_md, hash):
10076                 return size == size_default;
10077
10078         /* Fields that allow narrowing */
10079         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10080                 if (size < sizeof_field(struct sk_buff, protocol))
10081                         return false;
10082                 fallthrough;
10083         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10084         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10085         case bpf_ctx_range(struct sk_reuseport_md, len):
10086                 bpf_ctx_record_field_size(info, size_default);
10087                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10088
10089         default:
10090                 return false;
10091         }
10092 }
10093
10094 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10095         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10096                               si->dst_reg, si->src_reg,                 \
10097                               bpf_target_off(struct sk_reuseport_kern, F, \
10098                                              sizeof_field(struct sk_reuseport_kern, F), \
10099                                              target_size));             \
10100         })
10101
10102 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10103         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10104                                     struct sk_buff,                     \
10105                                     skb,                                \
10106                                     SKB_FIELD)
10107
10108 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10109         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10110                                     struct sock,                        \
10111                                     sk,                                 \
10112                                     SK_FIELD)
10113
10114 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10115                                            const struct bpf_insn *si,
10116                                            struct bpf_insn *insn_buf,
10117                                            struct bpf_prog *prog,
10118                                            u32 *target_size)
10119 {
10120         struct bpf_insn *insn = insn_buf;
10121
10122         switch (si->off) {
10123         case offsetof(struct sk_reuseport_md, data):
10124                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10125                 break;
10126
10127         case offsetof(struct sk_reuseport_md, len):
10128                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10129                 break;
10130
10131         case offsetof(struct sk_reuseport_md, eth_protocol):
10132                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10133                 break;
10134
10135         case offsetof(struct sk_reuseport_md, ip_protocol):
10136                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10137                 break;
10138
10139         case offsetof(struct sk_reuseport_md, data_end):
10140                 SK_REUSEPORT_LOAD_FIELD(data_end);
10141                 break;
10142
10143         case offsetof(struct sk_reuseport_md, hash):
10144                 SK_REUSEPORT_LOAD_FIELD(hash);
10145                 break;
10146
10147         case offsetof(struct sk_reuseport_md, bind_inany):
10148                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10149                 break;
10150         }
10151
10152         return insn - insn_buf;
10153 }
10154
10155 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10156         .get_func_proto         = sk_reuseport_func_proto,
10157         .is_valid_access        = sk_reuseport_is_valid_access,
10158         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10159 };
10160
10161 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10162 };
10163
10164 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10165 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10166
10167 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10168            struct sock *, sk, u64, flags)
10169 {
10170         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10171                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10172                 return -EINVAL;
10173         if (unlikely(sk && sk_is_refcounted(sk)))
10174                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10175         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10176                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10177
10178         /* Check if socket is suitable for packet L3/L4 protocol */
10179         if (sk && sk->sk_protocol != ctx->protocol)
10180                 return -EPROTOTYPE;
10181         if (sk && sk->sk_family != ctx->family &&
10182             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10183                 return -EAFNOSUPPORT;
10184
10185         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10186                 return -EEXIST;
10187
10188         /* Select socket as lookup result */
10189         ctx->selected_sk = sk;
10190         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10191         return 0;
10192 }
10193
10194 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10195         .func           = bpf_sk_lookup_assign,
10196         .gpl_only       = false,
10197         .ret_type       = RET_INTEGER,
10198         .arg1_type      = ARG_PTR_TO_CTX,
10199         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10200         .arg3_type      = ARG_ANYTHING,
10201 };
10202
10203 static const struct bpf_func_proto *
10204 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10205 {
10206         switch (func_id) {
10207         case BPF_FUNC_perf_event_output:
10208                 return &bpf_event_output_data_proto;
10209         case BPF_FUNC_sk_assign:
10210                 return &bpf_sk_lookup_assign_proto;
10211         case BPF_FUNC_sk_release:
10212                 return &bpf_sk_release_proto;
10213         default:
10214                 return bpf_sk_base_func_proto(func_id);
10215         }
10216 }
10217
10218 static bool sk_lookup_is_valid_access(int off, int size,
10219                                       enum bpf_access_type type,
10220                                       const struct bpf_prog *prog,
10221                                       struct bpf_insn_access_aux *info)
10222 {
10223         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10224                 return false;
10225         if (off % size != 0)
10226                 return false;
10227         if (type != BPF_READ)
10228                 return false;
10229
10230         switch (off) {
10231         case offsetof(struct bpf_sk_lookup, sk):
10232                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10233                 return size == sizeof(__u64);
10234
10235         case bpf_ctx_range(struct bpf_sk_lookup, family):
10236         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10237         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10238         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10239         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10240         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10241         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10242         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10243                 bpf_ctx_record_field_size(info, sizeof(__u32));
10244                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10245
10246         default:
10247                 return false;
10248         }
10249 }
10250
10251 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10252                                         const struct bpf_insn *si,
10253                                         struct bpf_insn *insn_buf,
10254                                         struct bpf_prog *prog,
10255                                         u32 *target_size)
10256 {
10257         struct bpf_insn *insn = insn_buf;
10258
10259         switch (si->off) {
10260         case offsetof(struct bpf_sk_lookup, sk):
10261                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10262                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10263                 break;
10264
10265         case offsetof(struct bpf_sk_lookup, family):
10266                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10267                                       bpf_target_off(struct bpf_sk_lookup_kern,
10268                                                      family, 2, target_size));
10269                 break;
10270
10271         case offsetof(struct bpf_sk_lookup, protocol):
10272                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10273                                       bpf_target_off(struct bpf_sk_lookup_kern,
10274                                                      protocol, 2, target_size));
10275                 break;
10276
10277         case offsetof(struct bpf_sk_lookup, remote_ip4):
10278                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10279                                       bpf_target_off(struct bpf_sk_lookup_kern,
10280                                                      v4.saddr, 4, target_size));
10281                 break;
10282
10283         case offsetof(struct bpf_sk_lookup, local_ip4):
10284                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10285                                       bpf_target_off(struct bpf_sk_lookup_kern,
10286                                                      v4.daddr, 4, target_size));
10287                 break;
10288
10289         case bpf_ctx_range_till(struct bpf_sk_lookup,
10290                                 remote_ip6[0], remote_ip6[3]): {
10291 #if IS_ENABLED(CONFIG_IPV6)
10292                 int off = si->off;
10293
10294                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10295                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10296                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10297                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10298                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10299                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10300 #else
10301                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10302 #endif
10303                 break;
10304         }
10305         case bpf_ctx_range_till(struct bpf_sk_lookup,
10306                                 local_ip6[0], local_ip6[3]): {
10307 #if IS_ENABLED(CONFIG_IPV6)
10308                 int off = si->off;
10309
10310                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10311                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10312                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10313                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10314                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10315                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10316 #else
10317                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10318 #endif
10319                 break;
10320         }
10321         case offsetof(struct bpf_sk_lookup, remote_port):
10322                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10323                                       bpf_target_off(struct bpf_sk_lookup_kern,
10324                                                      sport, 2, target_size));
10325                 break;
10326
10327         case offsetof(struct bpf_sk_lookup, local_port):
10328                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10329                                       bpf_target_off(struct bpf_sk_lookup_kern,
10330                                                      dport, 2, target_size));
10331                 break;
10332         }
10333
10334         return insn - insn_buf;
10335 }
10336
10337 const struct bpf_prog_ops sk_lookup_prog_ops = {
10338         .test_run = bpf_prog_test_run_sk_lookup,
10339 };
10340
10341 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10342         .get_func_proto         = sk_lookup_func_proto,
10343         .is_valid_access        = sk_lookup_is_valid_access,
10344         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10345 };
10346
10347 #endif /* CONFIG_INET */
10348
10349 DEFINE_BPF_DISPATCHER(xdp)
10350
10351 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10352 {
10353         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10354 }
10355
10356 #ifdef CONFIG_DEBUG_INFO_BTF
10357 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10358 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10359 BTF_SOCK_TYPE_xxx
10360 #undef BTF_SOCK_TYPE
10361 #else
10362 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10363 #endif
10364
10365 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10366 {
10367         /* tcp6_sock type is not generated in dwarf and hence btf,
10368          * trigger an explicit type generation here.
10369          */
10370         BTF_TYPE_EMIT(struct tcp6_sock);
10371         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10372             sk->sk_family == AF_INET6)
10373                 return (unsigned long)sk;
10374
10375         return (unsigned long)NULL;
10376 }
10377
10378 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10379         .func                   = bpf_skc_to_tcp6_sock,
10380         .gpl_only               = false,
10381         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10382         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10383         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10384 };
10385
10386 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10387 {
10388         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10389                 return (unsigned long)sk;
10390
10391         return (unsigned long)NULL;
10392 }
10393
10394 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10395         .func                   = bpf_skc_to_tcp_sock,
10396         .gpl_only               = false,
10397         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10398         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10399         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10400 };
10401
10402 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10403 {
10404         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10405          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10406          */
10407         BTF_TYPE_EMIT(struct inet_timewait_sock);
10408         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10409
10410 #ifdef CONFIG_INET
10411         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10412                 return (unsigned long)sk;
10413 #endif
10414
10415 #if IS_BUILTIN(CONFIG_IPV6)
10416         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10417                 return (unsigned long)sk;
10418 #endif
10419
10420         return (unsigned long)NULL;
10421 }
10422
10423 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10424         .func                   = bpf_skc_to_tcp_timewait_sock,
10425         .gpl_only               = false,
10426         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10427         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10428         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10429 };
10430
10431 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10432 {
10433 #ifdef CONFIG_INET
10434         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10435                 return (unsigned long)sk;
10436 #endif
10437
10438 #if IS_BUILTIN(CONFIG_IPV6)
10439         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10440                 return (unsigned long)sk;
10441 #endif
10442
10443         return (unsigned long)NULL;
10444 }
10445
10446 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10447         .func                   = bpf_skc_to_tcp_request_sock,
10448         .gpl_only               = false,
10449         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10450         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10451         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10452 };
10453
10454 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10455 {
10456         /* udp6_sock type is not generated in dwarf and hence btf,
10457          * trigger an explicit type generation here.
10458          */
10459         BTF_TYPE_EMIT(struct udp6_sock);
10460         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10461             sk->sk_type == SOCK_DGRAM && 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_udp6_sock_proto = {
10468         .func                   = bpf_skc_to_udp6_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_UDP6],
10473 };
10474
10475 static const struct bpf_func_proto *
10476 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10477 {
10478         const struct bpf_func_proto *func;
10479
10480         switch (func_id) {
10481         case BPF_FUNC_skc_to_tcp6_sock:
10482                 func = &bpf_skc_to_tcp6_sock_proto;
10483                 break;
10484         case BPF_FUNC_skc_to_tcp_sock:
10485                 func = &bpf_skc_to_tcp_sock_proto;
10486                 break;
10487         case BPF_FUNC_skc_to_tcp_timewait_sock:
10488                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10489                 break;
10490         case BPF_FUNC_skc_to_tcp_request_sock:
10491                 func = &bpf_skc_to_tcp_request_sock_proto;
10492                 break;
10493         case BPF_FUNC_skc_to_udp6_sock:
10494                 func = &bpf_skc_to_udp6_sock_proto;
10495                 break;
10496         default:
10497                 return bpf_base_func_proto(func_id);
10498         }
10499
10500         if (!perfmon_capable())
10501                 return NULL;
10502
10503         return func;
10504 }