net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/types.h>
  22 #include <linux/mm.h>
  23 #include <linux/fcntl.h>
  24 #include <linux/socket.h>
  25 #include <linux/sock_diag.h>
  26 #include <linux/in.h>
  27 #include <linux/inet.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/if_packet.h>
  30 #include <linux/if_arp.h>
  31 #include <linux/gfp.h>
  32 #include <net/inet_common.h>
  33 #include <net/ip.h>
  34 #include <net/protocol.h>
  35 #include <net/netlink.h>
  36 #include <linux/skbuff.h>
  37 #include <linux/skmsg.h>
  38 #include <net/sock.h>
  39 #include <net/flow_dissector.h>
  40 #include <linux/errno.h>
  41 #include <linux/timer.h>
  42 #include <linux/uaccess.h>
  43 #include <asm/unaligned.h>
  44 #include <asm/cmpxchg.h>
  45 #include <linux/filter.h>
  46 #include <linux/ratelimit.h>
  47 #include <linux/seccomp.h>
  48 #include <linux/if_vlan.h>
  49 #include <linux/bpf.h>
  50 #include <linux/btf.h>
  51 #include <net/sch_generic.h>
  52 #include <net/cls_cgroup.h>
  53 #include <net/dst_metadata.h>
  54 #include <net/dst.h>
  55 #include <net/sock_reuseport.h>
  56 #include <net/busy_poll.h>
  57 #include <net/tcp.h>
  58 #include <net/xfrm.h>
  59 #include <net/udp.h>
  60 #include <linux/bpf_trace.h>
  61 #include <net/xdp_sock.h>
  62 #include <linux/inetdevice.h>
  63 #include <net/inet_hashtables.h>
  64 #include <net/inet6_hashtables.h>
  65 #include <net/ip_fib.h>
  66 #include <net/nexthop.h>
  67 #include <net/flow.h>
  68 #include <net/arp.h>
  69 #include <net/ipv6.h>
  70 #include <net/net_namespace.h>
  71 #include <linux/seg6_local.h>
  72 #include <net/seg6.h>
  73 #include <net/seg6_local.h>
  74 #include <net/lwtunnel.h>
  75 #include <net/ipv6_stubs.h>
  76 #include <net/bpf_sk_storage.h>
  77 #include <net/transp_v6.h>
  78 #include <linux/btf_ids.h>
  79 #include <net/tls.h>
  80
  81 static const struct bpf_func_proto *
  82 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  83
  84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  85 {
  86         if (in_compat_syscall()) {
  87                 struct compat_sock_fprog f32;
  88
  89                 if (len != sizeof(f32))
  90                         return -EINVAL;
  91                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  92                         return -EFAULT;
  93                 memset(dst, 0, sizeof(*dst));
  94                 dst->len = f32.len;
  95                 dst->filter = compat_ptr(f32.filter);
  96         } else {
  97                 if (len != sizeof(*dst))
  98                         return -EINVAL;
  99                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 100                         return -EFAULT;
 101         }
 102
 103         return 0;
 104 }
 105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 106
 107 /**
 108  *      sk_filter_trim_cap - run a packet through a socket filter
 109  *      @sk: sock associated with &sk_buff
 110  *      @skb: buffer to filter
 111  *      @cap: limit on how short the eBPF program may trim the packet
 112  *
 113  * Run the eBPF program and then cut skb->data to correct size returned by
 114  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 115  * than pkt_len we keep whole skb->data. This is the socket level
 116  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
 117  * be accepted or -EPERM if the packet should be tossed.
 118  *
 119  */
 120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 121 {
 122         int err;
 123         struct sk_filter *filter;
 124
 125         /*
 126          * If the skb was allocated from pfmemalloc reserves, only
 127          * allow SOCK_MEMALLOC sockets to use it as this socket is
 128          * helping free memory
 129          */
 130         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 131                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 132                 return -ENOMEM;
 133         }
 134         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 135         if (err)
 136                 return err;
 137
 138         err = security_sock_rcv_skb(sk, skb);
 139         if (err)
 140                 return err;
 141
 142         rcu_read_lock();
 143         filter = rcu_dereference(sk->sk_filter);
 144         if (filter) {
 145                 struct sock *save_sk = skb->sk;
 146                 unsigned int pkt_len;
 147
 148                 skb->sk = sk;
 149                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 150                 skb->sk = save_sk;
 151                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 152         }
 153         rcu_read_unlock();
 154
 155         return err;
 156 }
 157 EXPORT_SYMBOL(sk_filter_trim_cap);
 158
 159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 160 {
 161         return skb_get_poff(skb);
 162 }
 163
 164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 165 {
 166         struct nlattr *nla;
 167
 168         if (skb_is_nonlinear(skb))
 169                 return 0;
 170
 171         if (skb->len < sizeof(struct nlattr))
 172                 return 0;
 173
 174         if (a > skb->len - sizeof(struct nlattr))
 175                 return 0;
 176
 177         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 178         if (nla)
 179                 return (void *) nla - (void *) skb->data;
 180
 181         return 0;
 182 }
 183
 184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 185 {
 186         struct nlattr *nla;
 187
 188         if (skb_is_nonlinear(skb))
 189                 return 0;
 190
 191         if (skb->len < sizeof(struct nlattr))
 192                 return 0;
 193
 194         if (a > skb->len - sizeof(struct nlattr))
 195                 return 0;
 196
 197         nla = (struct nlattr *) &skb->data[a];
 198         if (nla->nla_len > skb->len - a)
 199                 return 0;
 200
 201         nla = nla_find_nested(nla, x);
 202         if (nla)
 203                 return (void *) nla - (void *) skb->data;
 204
 205         return 0;
 206 }
 207
 208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 209            data, int, headlen, int, offset)
 210 {
 211         u8 tmp, *ptr;
 212         const int len = sizeof(tmp);
 213
 214         if (offset >= 0) {
 215                 if (headlen - offset >= len)
 216                         return *(u8 *)(data + offset);
 217                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 218                         return tmp;
 219         } else {
 220                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 221                 if (likely(ptr))
 222                         return *(u8 *)ptr;
 223         }
 224
 225         return -EFAULT;
 226 }
 227
 228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 229            int, offset)
 230 {
 231         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 232                                          offset);
 233 }
 234
 235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 236            data, int, headlen, int, offset)
 237 {
 238         u16 tmp, *ptr;
 239         const int len = sizeof(tmp);
 240
 241         if (offset >= 0) {
 242                 if (headlen - offset >= len)
 243                         return get_unaligned_be16(data + offset);
 244                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 245                         return be16_to_cpu(tmp);
 246         } else {
 247                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 248                 if (likely(ptr))
 249                         return get_unaligned_be16(ptr);
 250         }
 251
 252         return -EFAULT;
 253 }
 254
 255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 256            int, offset)
 257 {
 258         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 259                                           offset);
 260 }
 261
 262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 263            data, int, headlen, int, offset)
 264 {
 265         u32 tmp, *ptr;
 266         const int len = sizeof(tmp);
 267
 268         if (likely(offset >= 0)) {
 269                 if (headlen - offset >= len)
 270                         return get_unaligned_be32(data + offset);
 271                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 272                         return be32_to_cpu(tmp);
 273         } else {
 274                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 275                 if (likely(ptr))
 276                         return get_unaligned_be32(ptr);
 277         }
 278
 279         return -EFAULT;
 280 }
 281
 282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 283            int, offset)
 284 {
 285         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 286                                           offset);
 287 }
 288
 289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 290                               struct bpf_insn *insn_buf)
 291 {
 292         struct bpf_insn *insn = insn_buf;
 293
 294         switch (skb_field) {
 295         case SKF_AD_MARK:
 296                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 297
 298                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 299                                       offsetof(struct sk_buff, mark));
 300                 break;
 301
 302         case SKF_AD_PKTTYPE:
 303                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 304                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 305 #ifdef __BIG_ENDIAN_BITFIELD
 306                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 307 #endif
 308                 break;
 309
 310         case SKF_AD_QUEUE:
 311                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 312
 313                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 314                                       offsetof(struct sk_buff, queue_mapping));
 315                 break;
 316
 317         case SKF_AD_VLAN_TAG:
 318                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 319
 320                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 321                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 322                                       offsetof(struct sk_buff, vlan_tci));
 323                 break;
 324         case SKF_AD_VLAN_TAG_PRESENT:
 325                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET());
 326                 if (PKT_VLAN_PRESENT_BIT)
 327                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 328                 if (PKT_VLAN_PRESENT_BIT < 7)
 329                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 330                 break;
 331         }
 332
 333         return insn - insn_buf;
 334 }
 335
 336 static bool convert_bpf_extensions(struct sock_filter *fp,
 337                                    struct bpf_insn **insnp)
 338 {
 339         struct bpf_insn *insn = *insnp;
 340         u32 cnt;
 341
 342         switch (fp->k) {
 343         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 344                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 345
 346                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 347                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 348                                       offsetof(struct sk_buff, protocol));
 349                 /* A = ntohs(A) [emitting a nop or swap16] */
 350                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 351                 break;
 352
 353         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 354                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 355                 insn += cnt - 1;
 356                 break;
 357
 358         case SKF_AD_OFF + SKF_AD_IFINDEX:
 359         case SKF_AD_OFF + SKF_AD_HATYPE:
 360                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 361                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 362
 363                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 364                                       BPF_REG_TMP, BPF_REG_CTX,
 365                                       offsetof(struct sk_buff, dev));
 366                 /* if (tmp != 0) goto pc + 1 */
 367                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 368                 *insn++ = BPF_EXIT_INSN();
 369                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 370                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 371                                             offsetof(struct net_device, ifindex));
 372                 else
 373                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 374                                             offsetof(struct net_device, type));
 375                 break;
 376
 377         case SKF_AD_OFF + SKF_AD_MARK:
 378                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 379                 insn += cnt - 1;
 380                 break;
 381
 382         case SKF_AD_OFF + SKF_AD_RXHASH:
 383                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 384
 385                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 386                                     offsetof(struct sk_buff, hash));
 387                 break;
 388
 389         case SKF_AD_OFF + SKF_AD_QUEUE:
 390                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 391                 insn += cnt - 1;
 392                 break;
 393
 394         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 395                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 396                                          BPF_REG_A, BPF_REG_CTX, insn);
 397                 insn += cnt - 1;
 398                 break;
 399
 400         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 401                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 402                                          BPF_REG_A, BPF_REG_CTX, insn);
 403                 insn += cnt - 1;
 404                 break;
 405
 406         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 407                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 408
 409                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 410                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 411                                       offsetof(struct sk_buff, vlan_proto));
 412                 /* A = ntohs(A) [emitting a nop or swap16] */
 413                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 414                 break;
 415
 416         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 417         case SKF_AD_OFF + SKF_AD_NLATTR:
 418         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 419         case SKF_AD_OFF + SKF_AD_CPU:
 420         case SKF_AD_OFF + SKF_AD_RANDOM:
 421                 /* arg1 = CTX */
 422                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 423                 /* arg2 = A */
 424                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 425                 /* arg3 = X */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 427                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 428                 switch (fp->k) {
 429                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 430                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 431                         break;
 432                 case SKF_AD_OFF + SKF_AD_NLATTR:
 433                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 434                         break;
 435                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 436                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 437                         break;
 438                 case SKF_AD_OFF + SKF_AD_CPU:
 439                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 440                         break;
 441                 case SKF_AD_OFF + SKF_AD_RANDOM:
 442                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 443                         bpf_user_rnd_init_once();
 444                         break;
 445                 }
 446                 break;
 447
 448         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 449                 /* A ^= X */
 450                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 451                 break;
 452
 453         default:
 454                 /* This is just a dummy call to avoid letting the compiler
 455                  * evict __bpf_call_base() as an optimization. Placed here
 456                  * where no-one bothers.
 457                  */
 458                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 459                 return false;
 460         }
 461
 462         *insnp = insn;
 463         return true;
 464 }
 465
 466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 467 {
 468         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 469         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 470         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 471                       BPF_SIZE(fp->code) == BPF_W;
 472         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 473         const int ip_align = NET_IP_ALIGN;
 474         struct bpf_insn *insn = *insnp;
 475         int offset = fp->k;
 476
 477         if (!indirect &&
 478             ((unaligned_ok && offset >= 0) ||
 479              (!unaligned_ok && offset >= 0 &&
 480               offset + ip_align >= 0 &&
 481               offset + ip_align % size == 0))) {
 482                 bool ldx_off_ok = offset <= S16_MAX;
 483
 484                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 485                 if (offset)
 486                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 487                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 488                                       size, 2 + endian + (!ldx_off_ok * 2));
 489                 if (ldx_off_ok) {
 490                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 491                                               BPF_REG_D, offset);
 492                 } else {
 493                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 494                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 495                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 496                                               BPF_REG_TMP, 0);
 497                 }
 498                 if (endian)
 499                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 500                 *insn++ = BPF_JMP_A(8);
 501         }
 502
 503         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 504         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 505         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 506         if (!indirect) {
 507                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 508         } else {
 509                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 510                 if (fp->k)
 511                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 512         }
 513
 514         switch (BPF_SIZE(fp->code)) {
 515         case BPF_B:
 516                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 517                 break;
 518         case BPF_H:
 519                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 520                 break;
 521         case BPF_W:
 522                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 523                 break;
 524         default:
 525                 return false;
 526         }
 527
 528         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 529         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 530         *insn   = BPF_EXIT_INSN();
 531
 532         *insnp = insn;
 533         return true;
 534 }
 535
 536 /**
 537  *      bpf_convert_filter - convert filter program
 538  *      @prog: the user passed filter program
 539  *      @len: the length of the user passed filter program
 540  *      @new_prog: allocated 'struct bpf_prog' or NULL
 541  *      @new_len: pointer to store length of converted program
 542  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 543  *
 544  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 545  * style extended BPF (eBPF).
 546  * Conversion workflow:
 547  *
 548  * 1) First pass for calculating the new program length:
 549  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 550  *
 551  * 2) 2nd pass to remap in two passes: 1st pass finds new
 552  *    jump offsets, 2nd pass remapping:
 553  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 554  */
 555 static int bpf_convert_filter(struct sock_filter *prog, int len,
 556                               struct bpf_prog *new_prog, int *new_len,
 557                               bool *seen_ld_abs)
 558 {
 559         int new_flen = 0, pass = 0, target, i, stack_off;
 560         struct bpf_insn *new_insn, *first_insn = NULL;
 561         struct sock_filter *fp;
 562         int *addrs = NULL;
 563         u8 bpf_src;
 564
 565         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 566         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 567
 568         if (len <= 0 || len > BPF_MAXINSNS)
 569                 return -EINVAL;
 570
 571         if (new_prog) {
 572                 first_insn = new_prog->insnsi;
 573                 addrs = kcalloc(len, sizeof(*addrs),
 574                                 GFP_KERNEL | __GFP_NOWARN);
 575                 if (!addrs)
 576                         return -ENOMEM;
 577         }
 578
 579 do_pass:
 580         new_insn = first_insn;
 581         fp = prog;
 582
 583         /* Classic BPF related prologue emission. */
 584         if (new_prog) {
 585                 /* Classic BPF expects A and X to be reset first. These need
 586                  * to be guaranteed to be the first two instructions.
 587                  */
 588                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 589                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 590
 591                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 592                  * In eBPF case it's done by the compiler, here we need to
 593                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 594                  */
 595                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 596                 if (*seen_ld_abs) {
 597                         /* For packet access in classic BPF, cache skb->data
 598                          * in callee-saved BPF R8 and skb->len - skb->data_len
 599                          * (headlen) in BPF R9. Since classic BPF is read-only
 600                          * on CTX, we only need to cache it once.
 601                          */
 602                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 603                                                   BPF_REG_D, BPF_REG_CTX,
 604                                                   offsetof(struct sk_buff, data));
 605                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 606                                                   offsetof(struct sk_buff, len));
 607                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, data_len));
 609                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 610                 }
 611         } else {
 612                 new_insn += 3;
 613         }
 614
 615         for (i = 0; i < len; fp++, i++) {
 616                 struct bpf_insn tmp_insns[32] = { };
 617                 struct bpf_insn *insn = tmp_insns;
 618
 619                 if (addrs)
 620                         addrs[i] = new_insn - first_insn;
 621
 622                 switch (fp->code) {
 623                 /* All arithmetic insns and skb loads map as-is. */
 624                 case BPF_ALU | BPF_ADD | BPF_X:
 625                 case BPF_ALU | BPF_ADD | BPF_K:
 626                 case BPF_ALU | BPF_SUB | BPF_X:
 627                 case BPF_ALU | BPF_SUB | BPF_K:
 628                 case BPF_ALU | BPF_AND | BPF_X:
 629                 case BPF_ALU | BPF_AND | BPF_K:
 630                 case BPF_ALU | BPF_OR | BPF_X:
 631                 case BPF_ALU | BPF_OR | BPF_K:
 632                 case BPF_ALU | BPF_LSH | BPF_X:
 633                 case BPF_ALU | BPF_LSH | BPF_K:
 634                 case BPF_ALU | BPF_RSH | BPF_X:
 635                 case BPF_ALU | BPF_RSH | BPF_K:
 636                 case BPF_ALU | BPF_XOR | BPF_X:
 637                 case BPF_ALU | BPF_XOR | BPF_K:
 638                 case BPF_ALU | BPF_MUL | BPF_X:
 639                 case BPF_ALU | BPF_MUL | BPF_K:
 640                 case BPF_ALU | BPF_DIV | BPF_X:
 641                 case BPF_ALU | BPF_DIV | BPF_K:
 642                 case BPF_ALU | BPF_MOD | BPF_X:
 643                 case BPF_ALU | BPF_MOD | BPF_K:
 644                 case BPF_ALU | BPF_NEG:
 645                 case BPF_LD | BPF_ABS | BPF_W:
 646                 case BPF_LD | BPF_ABS | BPF_H:
 647                 case BPF_LD | BPF_ABS | BPF_B:
 648                 case BPF_LD | BPF_IND | BPF_W:
 649                 case BPF_LD | BPF_IND | BPF_H:
 650                 case BPF_LD | BPF_IND | BPF_B:
 651                         /* Check for overloaded BPF extension and
 652                          * directly convert it if found, otherwise
 653                          * just move on with mapping.
 654                          */
 655                         if (BPF_CLASS(fp->code) == BPF_LD &&
 656                             BPF_MODE(fp->code) == BPF_ABS &&
 657                             convert_bpf_extensions(fp, &insn))
 658                                 break;
 659                         if (BPF_CLASS(fp->code) == BPF_LD &&
 660                             convert_bpf_ld_abs(fp, &insn)) {
 661                                 *seen_ld_abs = true;
 662                                 break;
 663                         }
 664
 665                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 666                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 667                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 668                                 /* Error with exception code on div/mod by 0.
 669                                  * For cBPF programs, this was always return 0.
 670                                  */
 671                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 672                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 673                                 *insn++ = BPF_EXIT_INSN();
 674                         }
 675
 676                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 677                         break;
 678
 679                 /* Jump transformation cannot use BPF block macros
 680                  * everywhere as offset calculation and target updates
 681                  * require a bit more work than the rest, i.e. jump
 682                  * opcodes map as-is, but offsets need adjustment.
 683                  */
 684
 685 #define BPF_EMIT_JMP                                                    \
 686         do {                                                            \
 687                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 688                 s32 off;                                                \
 689                                                                         \
 690                 if (target >= len || target < 0)                        \
 691                         goto err;                                       \
 692                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 693                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 694                 off -= insn - tmp_insns;                                \
 695                 /* Reject anything not fitting into insn->off. */       \
 696                 if (off < off_min || off > off_max)                     \
 697                         goto err;                                       \
 698                 insn->off = off;                                        \
 699         } while (0)
 700
 701                 case BPF_JMP | BPF_JA:
 702                         target = i + fp->k + 1;
 703                         insn->code = fp->code;
 704                         BPF_EMIT_JMP;
 705                         break;
 706
 707                 case BPF_JMP | BPF_JEQ | BPF_K:
 708                 case BPF_JMP | BPF_JEQ | BPF_X:
 709                 case BPF_JMP | BPF_JSET | BPF_K:
 710                 case BPF_JMP | BPF_JSET | BPF_X:
 711                 case BPF_JMP | BPF_JGT | BPF_K:
 712                 case BPF_JMP | BPF_JGT | BPF_X:
 713                 case BPF_JMP | BPF_JGE | BPF_K:
 714                 case BPF_JMP | BPF_JGE | BPF_X:
 715                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 716                                 /* BPF immediates are signed, zero extend
 717                                  * immediate into tmp register and use it
 718                                  * in compare insn.
 719                                  */
 720                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 721
 722                                 insn->dst_reg = BPF_REG_A;
 723                                 insn->src_reg = BPF_REG_TMP;
 724                                 bpf_src = BPF_X;
 725                         } else {
 726                                 insn->dst_reg = BPF_REG_A;
 727                                 insn->imm = fp->k;
 728                                 bpf_src = BPF_SRC(fp->code);
 729                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 730                         }
 731
 732                         /* Common case where 'jump_false' is next insn. */
 733                         if (fp->jf == 0) {
 734                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 735                                 target = i + fp->jt + 1;
 736                                 BPF_EMIT_JMP;
 737                                 break;
 738                         }
 739
 740                         /* Convert some jumps when 'jump_true' is next insn. */
 741                         if (fp->jt == 0) {
 742                                 switch (BPF_OP(fp->code)) {
 743                                 case BPF_JEQ:
 744                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 745                                         break;
 746                                 case BPF_JGT:
 747                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 748                                         break;
 749                                 case BPF_JGE:
 750                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 751                                         break;
 752                                 default:
 753                                         goto jmp_rest;
 754                                 }
 755
 756                                 target = i + fp->jf + 1;
 757                                 BPF_EMIT_JMP;
 758                                 break;
 759                         }
 760 jmp_rest:
 761                         /* Other jumps are mapped into two insns: Jxx and JA. */
 762                         target = i + fp->jt + 1;
 763                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 764                         BPF_EMIT_JMP;
 765                         insn++;
 766
 767                         insn->code = BPF_JMP | BPF_JA;
 768                         target = i + fp->jf + 1;
 769                         BPF_EMIT_JMP;
 770                         break;
 771
 772                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 773                 case BPF_LDX | BPF_MSH | BPF_B: {
 774                         struct sock_filter tmp = {
 775                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 776                                 .k      = fp->k,
 777                         };
 778
 779                         *seen_ld_abs = true;
 780
 781                         /* X = A */
 782                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 783                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 784                         convert_bpf_ld_abs(&tmp, &insn);
 785                         insn++;
 786                         /* A &= 0xf */
 787                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 788                         /* A <<= 2 */
 789                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 790                         /* tmp = X */
 791                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 792                         /* X = A */
 793                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 794                         /* A = tmp */
 795                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 796                         break;
 797                 }
 798                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 799                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 800                  */
 801                 case BPF_RET | BPF_A:
 802                 case BPF_RET | BPF_K:
 803                         if (BPF_RVAL(fp->code) == BPF_K)
 804                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 805                                                         0, fp->k);
 806                         *insn = BPF_EXIT_INSN();
 807                         break;
 808
 809                 /* Store to stack. */
 810                 case BPF_ST:
 811                 case BPF_STX:
 812                         stack_off = fp->k * 4  + 4;
 813                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 814                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 815                                             -stack_off);
 816                         /* check_load_and_stores() verifies that classic BPF can
 817                          * load from stack only after write, so tracking
 818                          * stack_depth for ST|STX insns is enough
 819                          */
 820                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 821                                 new_prog->aux->stack_depth = stack_off;
 822                         break;
 823
 824                 /* Load from stack. */
 825                 case BPF_LD | BPF_MEM:
 826                 case BPF_LDX | BPF_MEM:
 827                         stack_off = fp->k * 4  + 4;
 828                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 829                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 830                                             -stack_off);
 831                         break;
 832
 833                 /* A = K or X = K */
 834                 case BPF_LD | BPF_IMM:
 835                 case BPF_LDX | BPF_IMM:
 836                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 837                                               BPF_REG_A : BPF_REG_X, fp->k);
 838                         break;
 839
 840                 /* X = A */
 841                 case BPF_MISC | BPF_TAX:
 842                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 843                         break;
 844
 845                 /* A = X */
 846                 case BPF_MISC | BPF_TXA:
 847                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 848                         break;
 849
 850                 /* A = skb->len or X = skb->len */
 851                 case BPF_LD | BPF_W | BPF_LEN:
 852                 case BPF_LDX | BPF_W | BPF_LEN:
 853                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 854                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 855                                             offsetof(struct sk_buff, len));
 856                         break;
 857
 858                 /* Access seccomp_data fields. */
 859                 case BPF_LDX | BPF_ABS | BPF_W:
 860                         /* A = *(u32 *) (ctx + K) */
 861                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 862                         break;
 863
 864                 /* Unknown instruction. */
 865                 default:
 866                         goto err;
 867                 }
 868
 869                 insn++;
 870                 if (new_prog)
 871                         memcpy(new_insn, tmp_insns,
 872                                sizeof(*insn) * (insn - tmp_insns));
 873                 new_insn += insn - tmp_insns;
 874         }
 875
 876         if (!new_prog) {
 877                 /* Only calculating new length. */
 878                 *new_len = new_insn - first_insn;
 879                 if (*seen_ld_abs)
 880                         *new_len += 4; /* Prologue bits. */
 881                 return 0;
 882         }
 883
 884         pass++;
 885         if (new_flen != new_insn - first_insn) {
 886                 new_flen = new_insn - first_insn;
 887                 if (pass > 2)
 888                         goto err;
 889                 goto do_pass;
 890         }
 891
 892         kfree(addrs);
 893         BUG_ON(*new_len != new_flen);
 894         return 0;
 895 err:
 896         kfree(addrs);
 897         return -EINVAL;
 898 }
 899
 900 /* Security:
 901  *
 902  * As we dont want to clear mem[] array for each packet going through
 903  * __bpf_prog_run(), we check that filter loaded by user never try to read
 904  * a cell if not previously written, and we check all branches to be sure
 905  * a malicious user doesn't try to abuse us.
 906  */
 907 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 908 {
 909         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 910         int pc, ret = 0;
 911
 912         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 913
 914         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 915         if (!masks)
 916                 return -ENOMEM;
 917
 918         memset(masks, 0xff, flen * sizeof(*masks));
 919
 920         for (pc = 0; pc < flen; pc++) {
 921                 memvalid &= masks[pc];
 922
 923                 switch (filter[pc].code) {
 924                 case BPF_ST:
 925                 case BPF_STX:
 926                         memvalid |= (1 << filter[pc].k);
 927                         break;
 928                 case BPF_LD | BPF_MEM:
 929                 case BPF_LDX | BPF_MEM:
 930                         if (!(memvalid & (1 << filter[pc].k))) {
 931                                 ret = -EINVAL;
 932                                 goto error;
 933                         }
 934                         break;
 935                 case BPF_JMP | BPF_JA:
 936                         /* A jump must set masks on target */
 937                         masks[pc + 1 + filter[pc].k] &= memvalid;
 938                         memvalid = ~0;
 939                         break;
 940                 case BPF_JMP | BPF_JEQ | BPF_K:
 941                 case BPF_JMP | BPF_JEQ | BPF_X:
 942                 case BPF_JMP | BPF_JGE | BPF_K:
 943                 case BPF_JMP | BPF_JGE | BPF_X:
 944                 case BPF_JMP | BPF_JGT | BPF_K:
 945                 case BPF_JMP | BPF_JGT | BPF_X:
 946                 case BPF_JMP | BPF_JSET | BPF_K:
 947                 case BPF_JMP | BPF_JSET | BPF_X:
 948                         /* A jump must set masks on targets */
 949                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 950                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 951                         memvalid = ~0;
 952                         break;
 953                 }
 954         }
 955 error:
 956         kfree(masks);
 957         return ret;
 958 }
 959
 960 static bool chk_code_allowed(u16 code_to_probe)
 961 {
 962         static const bool codes[] = {
 963                 /* 32 bit ALU operations */
 964                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 965                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 966                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 967                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 968                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 969                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 970                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 971                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 972                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 973                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 974                 [BPF_ALU | BPF_AND | BPF_K] = true,
 975                 [BPF_ALU | BPF_AND | BPF_X] = true,
 976                 [BPF_ALU | BPF_OR | BPF_K] = true,
 977                 [BPF_ALU | BPF_OR | BPF_X] = true,
 978                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 979                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 980                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 981                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 982                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 983                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 984                 [BPF_ALU | BPF_NEG] = true,
 985                 /* Load instructions */
 986                 [BPF_LD | BPF_W | BPF_ABS] = true,
 987                 [BPF_LD | BPF_H | BPF_ABS] = true,
 988                 [BPF_LD | BPF_B | BPF_ABS] = true,
 989                 [BPF_LD | BPF_W | BPF_LEN] = true,
 990                 [BPF_LD | BPF_W | BPF_IND] = true,
 991                 [BPF_LD | BPF_H | BPF_IND] = true,
 992                 [BPF_LD | BPF_B | BPF_IND] = true,
 993                 [BPF_LD | BPF_IMM] = true,
 994                 [BPF_LD | BPF_MEM] = true,
 995                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 996                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 997                 [BPF_LDX | BPF_IMM] = true,
 998                 [BPF_LDX | BPF_MEM] = true,
 999                 /* Store instructions */
1000                 [BPF_ST] = true,
1001                 [BPF_STX] = true,
1002                 /* Misc instructions */
1003                 [BPF_MISC | BPF_TAX] = true,
1004                 [BPF_MISC | BPF_TXA] = true,
1005                 /* Return instructions */
1006                 [BPF_RET | BPF_K] = true,
1007                 [BPF_RET | BPF_A] = true,
1008                 /* Jump instructions */
1009                 [BPF_JMP | BPF_JA] = true,
1010                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1011                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1012                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1013                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1014                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1015                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1016                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1017                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1018         };
1019
1020         if (code_to_probe >= ARRAY_SIZE(codes))
1021                 return false;
1022
1023         return codes[code_to_probe];
1024 }
1025
1026 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1027                                 unsigned int flen)
1028 {
1029         if (filter == NULL)
1030                 return false;
1031         if (flen == 0 || flen > BPF_MAXINSNS)
1032                 return false;
1033
1034         return true;
1035 }
1036
1037 /**
1038  *      bpf_check_classic - verify socket filter code
1039  *      @filter: filter to verify
1040  *      @flen: length of filter
1041  *
1042  * Check the user's filter code. If we let some ugly
1043  * filter code slip through kaboom! The filter must contain
1044  * no references or jumps that are out of range, no illegal
1045  * instructions, and must end with a RET instruction.
1046  *
1047  * All jumps are forward as they are not signed.
1048  *
1049  * Returns 0 if the rule set is legal or -EINVAL if not.
1050  */
1051 static int bpf_check_classic(const struct sock_filter *filter,
1052                              unsigned int flen)
1053 {
1054         bool anc_found;
1055         int pc;
1056
1057         /* Check the filter code now */
1058         for (pc = 0; pc < flen; pc++) {
1059                 const struct sock_filter *ftest = &filter[pc];
1060
1061                 /* May we actually operate on this code? */
1062                 if (!chk_code_allowed(ftest->code))
1063                         return -EINVAL;
1064
1065                 /* Some instructions need special checks */
1066                 switch (ftest->code) {
1067                 case BPF_ALU | BPF_DIV | BPF_K:
1068                 case BPF_ALU | BPF_MOD | BPF_K:
1069                         /* Check for division by zero */
1070                         if (ftest->k == 0)
1071                                 return -EINVAL;
1072                         break;
1073                 case BPF_ALU | BPF_LSH | BPF_K:
1074                 case BPF_ALU | BPF_RSH | BPF_K:
1075                         if (ftest->k >= 32)
1076                                 return -EINVAL;
1077                         break;
1078                 case BPF_LD | BPF_MEM:
1079                 case BPF_LDX | BPF_MEM:
1080                 case BPF_ST:
1081                 case BPF_STX:
1082                         /* Check for invalid memory addresses */
1083                         if (ftest->k >= BPF_MEMWORDS)
1084                                 return -EINVAL;
1085                         break;
1086                 case BPF_JMP | BPF_JA:
1087                         /* Note, the large ftest->k might cause loops.
1088                          * Compare this with conditional jumps below,
1089                          * where offsets are limited. --ANK (981016)
1090                          */
1091                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1092                                 return -EINVAL;
1093                         break;
1094                 case BPF_JMP | BPF_JEQ | BPF_K:
1095                 case BPF_JMP | BPF_JEQ | BPF_X:
1096                 case BPF_JMP | BPF_JGE | BPF_K:
1097                 case BPF_JMP | BPF_JGE | BPF_X:
1098                 case BPF_JMP | BPF_JGT | BPF_K:
1099                 case BPF_JMP | BPF_JGT | BPF_X:
1100                 case BPF_JMP | BPF_JSET | BPF_K:
1101                 case BPF_JMP | BPF_JSET | BPF_X:
1102                         /* Both conditionals must be safe */
1103                         if (pc + ftest->jt + 1 >= flen ||
1104                             pc + ftest->jf + 1 >= flen)
1105                                 return -EINVAL;
1106                         break;
1107                 case BPF_LD | BPF_W | BPF_ABS:
1108                 case BPF_LD | BPF_H | BPF_ABS:
1109                 case BPF_LD | BPF_B | BPF_ABS:
1110                         anc_found = false;
1111                         if (bpf_anc_helper(ftest) & BPF_ANC)
1112                                 anc_found = true;
1113                         /* Ancillary operation unknown or unsupported */
1114                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1115                                 return -EINVAL;
1116                 }
1117         }
1118
1119         /* Last instruction must be a RET code */
1120         switch (filter[flen - 1].code) {
1121         case BPF_RET | BPF_K:
1122         case BPF_RET | BPF_A:
1123                 return check_load_and_stores(filter, flen);
1124         }
1125
1126         return -EINVAL;
1127 }
1128
1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1130                                       const struct sock_fprog *fprog)
1131 {
1132         unsigned int fsize = bpf_classic_proglen(fprog);
1133         struct sock_fprog_kern *fkprog;
1134
1135         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1136         if (!fp->orig_prog)
1137                 return -ENOMEM;
1138
1139         fkprog = fp->orig_prog;
1140         fkprog->len = fprog->len;
1141
1142         fkprog->filter = kmemdup(fp->insns, fsize,
1143                                  GFP_KERNEL | __GFP_NOWARN);
1144         if (!fkprog->filter) {
1145                 kfree(fp->orig_prog);
1146                 return -ENOMEM;
1147         }
1148
1149         return 0;
1150 }
1151
1152 static void bpf_release_orig_filter(struct bpf_prog *fp)
1153 {
1154         struct sock_fprog_kern *fprog = fp->orig_prog;
1155
1156         if (fprog) {
1157                 kfree(fprog->filter);
1158                 kfree(fprog);
1159         }
1160 }
1161
1162 static void __bpf_prog_release(struct bpf_prog *prog)
1163 {
1164         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1165                 bpf_prog_put(prog);
1166         } else {
1167                 bpf_release_orig_filter(prog);
1168                 bpf_prog_free(prog);
1169         }
1170 }
1171
1172 static void __sk_filter_release(struct sk_filter *fp)
1173 {
1174         __bpf_prog_release(fp->prog);
1175         kfree(fp);
1176 }
1177
1178 /**
1179  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1180  *      @rcu: rcu_head that contains the sk_filter to free
1181  */
1182 static void sk_filter_release_rcu(struct rcu_head *rcu)
1183 {
1184         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1185
1186         __sk_filter_release(fp);
1187 }
1188
1189 /**
1190  *      sk_filter_release - release a socket filter
1191  *      @fp: filter to remove
1192  *
1193  *      Remove a filter from a socket and release its resources.
1194  */
1195 static void sk_filter_release(struct sk_filter *fp)
1196 {
1197         if (refcount_dec_and_test(&fp->refcnt))
1198                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1199 }
1200
1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1202 {
1203         u32 filter_size = bpf_prog_size(fp->prog->len);
1204
1205         atomic_sub(filter_size, &sk->sk_omem_alloc);
1206         sk_filter_release(fp);
1207 }
1208
1209 /* try to charge the socket memory if there is space available
1210  * return true on success
1211  */
1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1213 {
1214         u32 filter_size = bpf_prog_size(fp->prog->len);
1215
1216         /* same check as in sock_kmalloc() */
1217         if (filter_size <= sysctl_optmem_max &&
1218             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1219                 atomic_add(filter_size, &sk->sk_omem_alloc);
1220                 return true;
1221         }
1222         return false;
1223 }
1224
1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1226 {
1227         if (!refcount_inc_not_zero(&fp->refcnt))
1228                 return false;
1229
1230         if (!__sk_filter_charge(sk, fp)) {
1231                 sk_filter_release(fp);
1232                 return false;
1233         }
1234         return true;
1235 }
1236
1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1238 {
1239         struct sock_filter *old_prog;
1240         struct bpf_prog *old_fp;
1241         int err, new_len, old_len = fp->len;
1242         bool seen_ld_abs = false;
1243
1244         /* We are free to overwrite insns et al right here as it
1245          * won't be used at this point in time anymore internally
1246          * after the migration to the internal BPF instruction
1247          * representation.
1248          */
1249         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1250                      sizeof(struct bpf_insn));
1251
1252         /* Conversion cannot happen on overlapping memory areas,
1253          * so we need to keep the user BPF around until the 2nd
1254          * pass. At this time, the user BPF is stored in fp->insns.
1255          */
1256         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1257                            GFP_KERNEL | __GFP_NOWARN);
1258         if (!old_prog) {
1259                 err = -ENOMEM;
1260                 goto out_err;
1261         }
1262
1263         /* 1st pass: calculate the new program length. */
1264         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1265                                  &seen_ld_abs);
1266         if (err)
1267                 goto out_err_free;
1268
1269         /* Expand fp for appending the new filter representation. */
1270         old_fp = fp;
1271         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1272         if (!fp) {
1273                 /* The old_fp is still around in case we couldn't
1274                  * allocate new memory, so uncharge on that one.
1275                  */
1276                 fp = old_fp;
1277                 err = -ENOMEM;
1278                 goto out_err_free;
1279         }
1280
1281         fp->len = new_len;
1282
1283         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1284         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1285                                  &seen_ld_abs);
1286         if (err)
1287                 /* 2nd bpf_convert_filter() can fail only if it fails
1288                  * to allocate memory, remapping must succeed. Note,
1289                  * that at this time old_fp has already been released
1290                  * by krealloc().
1291                  */
1292                 goto out_err_free;
1293
1294         fp = bpf_prog_select_runtime(fp, &err);
1295         if (err)
1296                 goto out_err_free;
1297
1298         kfree(old_prog);
1299         return fp;
1300
1301 out_err_free:
1302         kfree(old_prog);
1303 out_err:
1304         __bpf_prog_release(fp);
1305         return ERR_PTR(err);
1306 }
1307
1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1309                                            bpf_aux_classic_check_t trans)
1310 {
1311         int err;
1312
1313         fp->bpf_func = NULL;
1314         fp->jited = 0;
1315
1316         err = bpf_check_classic(fp->insns, fp->len);
1317         if (err) {
1318                 __bpf_prog_release(fp);
1319                 return ERR_PTR(err);
1320         }
1321
1322         /* There might be additional checks and transformations
1323          * needed on classic filters, f.e. in case of seccomp.
1324          */
1325         if (trans) {
1326                 err = trans(fp->insns, fp->len);
1327                 if (err) {
1328                         __bpf_prog_release(fp);
1329                         return ERR_PTR(err);
1330                 }
1331         }
1332
1333         /* Probe if we can JIT compile the filter and if so, do
1334          * the compilation of the filter.
1335          */
1336         bpf_jit_compile(fp);
1337
1338         /* JIT compiler couldn't process this filter, so do the
1339          * internal BPF translation for the optimized interpreter.
1340          */
1341         if (!fp->jited)
1342                 fp = bpf_migrate_filter(fp);
1343
1344         return fp;
1345 }
1346
1347 /**
1348  *      bpf_prog_create - create an unattached filter
1349  *      @pfp: the unattached filter that is created
1350  *      @fprog: the filter program
1351  *
1352  * Create a filter independent of any socket. We first run some
1353  * sanity checks on it to make sure it does not explode on us later.
1354  * If an error occurs or there is insufficient memory for the filter
1355  * a negative errno code is returned. On success the return is zero.
1356  */
1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1358 {
1359         unsigned int fsize = bpf_classic_proglen(fprog);
1360         struct bpf_prog *fp;
1361
1362         /* Make sure new filter is there and in the right amounts. */
1363         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1364                 return -EINVAL;
1365
1366         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1367         if (!fp)
1368                 return -ENOMEM;
1369
1370         memcpy(fp->insns, fprog->filter, fsize);
1371
1372         fp->len = fprog->len;
1373         /* Since unattached filters are not copied back to user
1374          * space through sk_get_filter(), we do not need to hold
1375          * a copy here, and can spare us the work.
1376          */
1377         fp->orig_prog = NULL;
1378
1379         /* bpf_prepare_filter() already takes care of freeing
1380          * memory in case something goes wrong.
1381          */
1382         fp = bpf_prepare_filter(fp, NULL);
1383         if (IS_ERR(fp))
1384                 return PTR_ERR(fp);
1385
1386         *pfp = fp;
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(bpf_prog_create);
1390
1391 /**
1392  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1393  *      @pfp: the unattached filter that is created
1394  *      @fprog: the filter program
1395  *      @trans: post-classic verifier transformation handler
1396  *      @save_orig: save classic BPF program
1397  *
1398  * This function effectively does the same as bpf_prog_create(), only
1399  * that it builds up its insns buffer from user space provided buffer.
1400  * It also allows for passing a bpf_aux_classic_check_t handler.
1401  */
1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1403                               bpf_aux_classic_check_t trans, bool save_orig)
1404 {
1405         unsigned int fsize = bpf_classic_proglen(fprog);
1406         struct bpf_prog *fp;
1407         int err;
1408
1409         /* Make sure new filter is there and in the right amounts. */
1410         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1411                 return -EINVAL;
1412
1413         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1414         if (!fp)
1415                 return -ENOMEM;
1416
1417         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1418                 __bpf_prog_free(fp);
1419                 return -EFAULT;
1420         }
1421
1422         fp->len = fprog->len;
1423         fp->orig_prog = NULL;
1424
1425         if (save_orig) {
1426                 err = bpf_prog_store_orig_filter(fp, fprog);
1427                 if (err) {
1428                         __bpf_prog_free(fp);
1429                         return -ENOMEM;
1430                 }
1431         }
1432
1433         /* bpf_prepare_filter() already takes care of freeing
1434          * memory in case something goes wrong.
1435          */
1436         fp = bpf_prepare_filter(fp, trans);
1437         if (IS_ERR(fp))
1438                 return PTR_ERR(fp);
1439
1440         *pfp = fp;
1441         return 0;
1442 }
1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1444
1445 void bpf_prog_destroy(struct bpf_prog *fp)
1446 {
1447         __bpf_prog_release(fp);
1448 }
1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1450
1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1452 {
1453         struct sk_filter *fp, *old_fp;
1454
1455         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1456         if (!fp)
1457                 return -ENOMEM;
1458
1459         fp->prog = prog;
1460
1461         if (!__sk_filter_charge(sk, fp)) {
1462                 kfree(fp);
1463                 return -ENOMEM;
1464         }
1465         refcount_set(&fp->refcnt, 1);
1466
1467         old_fp = rcu_dereference_protected(sk->sk_filter,
1468                                            lockdep_sock_is_held(sk));
1469         rcu_assign_pointer(sk->sk_filter, fp);
1470
1471         if (old_fp)
1472                 sk_filter_uncharge(sk, old_fp);
1473
1474         return 0;
1475 }
1476
1477 static
1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1479 {
1480         unsigned int fsize = bpf_classic_proglen(fprog);
1481         struct bpf_prog *prog;
1482         int err;
1483
1484         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1485                 return ERR_PTR(-EPERM);
1486
1487         /* Make sure new filter is there and in the right amounts. */
1488         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1489                 return ERR_PTR(-EINVAL);
1490
1491         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1492         if (!prog)
1493                 return ERR_PTR(-ENOMEM);
1494
1495         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1496                 __bpf_prog_free(prog);
1497                 return ERR_PTR(-EFAULT);
1498         }
1499
1500         prog->len = fprog->len;
1501
1502         err = bpf_prog_store_orig_filter(prog, fprog);
1503         if (err) {
1504                 __bpf_prog_free(prog);
1505                 return ERR_PTR(-ENOMEM);
1506         }
1507
1508         /* bpf_prepare_filter() already takes care of freeing
1509          * memory in case something goes wrong.
1510          */
1511         return bpf_prepare_filter(prog, NULL);
1512 }
1513
1514 /**
1515  *      sk_attach_filter - attach a socket filter
1516  *      @fprog: the filter program
1517  *      @sk: the socket to use
1518  *
1519  * Attach the user's filter code. We first run some sanity checks on
1520  * it to make sure it does not explode on us later. If an error
1521  * occurs or there is insufficient memory for the filter a negative
1522  * errno code is returned. On success the return is zero.
1523  */
1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1525 {
1526         struct bpf_prog *prog = __get_filter(fprog, sk);
1527         int err;
1528
1529         if (IS_ERR(prog))
1530                 return PTR_ERR(prog);
1531
1532         err = __sk_attach_prog(prog, sk);
1533         if (err < 0) {
1534                 __bpf_prog_release(prog);
1535                 return err;
1536         }
1537
1538         return 0;
1539 }
1540 EXPORT_SYMBOL_GPL(sk_attach_filter);
1541
1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1543 {
1544         struct bpf_prog *prog = __get_filter(fprog, sk);
1545         int err;
1546
1547         if (IS_ERR(prog))
1548                 return PTR_ERR(prog);
1549
1550         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1551                 err = -ENOMEM;
1552         else
1553                 err = reuseport_attach_prog(sk, prog);
1554
1555         if (err)
1556                 __bpf_prog_release(prog);
1557
1558         return err;
1559 }
1560
1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1562 {
1563         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1564                 return ERR_PTR(-EPERM);
1565
1566         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1567 }
1568
1569 int sk_attach_bpf(u32 ufd, struct sock *sk)
1570 {
1571         struct bpf_prog *prog = __get_bpf(ufd, sk);
1572         int err;
1573
1574         if (IS_ERR(prog))
1575                 return PTR_ERR(prog);
1576
1577         err = __sk_attach_prog(prog, sk);
1578         if (err < 0) {
1579                 bpf_prog_put(prog);
1580                 return err;
1581         }
1582
1583         return 0;
1584 }
1585
1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1587 {
1588         struct bpf_prog *prog;
1589         int err;
1590
1591         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1592                 return -EPERM;
1593
1594         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1595         if (PTR_ERR(prog) == -EINVAL)
1596                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1597         if (IS_ERR(prog))
1598                 return PTR_ERR(prog);
1599
1600         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1601                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1602                  * bpf prog (e.g. sockmap).  It depends on the
1603                  * limitation imposed by bpf_prog_load().
1604                  * Hence, sysctl_optmem_max is not checked.
1605                  */
1606                 if ((sk->sk_type != SOCK_STREAM &&
1607                      sk->sk_type != SOCK_DGRAM) ||
1608                     (sk->sk_protocol != IPPROTO_UDP &&
1609                      sk->sk_protocol != IPPROTO_TCP) ||
1610                     (sk->sk_family != AF_INET &&
1611                      sk->sk_family != AF_INET6)) {
1612                         err = -ENOTSUPP;
1613                         goto err_prog_put;
1614                 }
1615         } else {
1616                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1617                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1618                         err = -ENOMEM;
1619                         goto err_prog_put;
1620                 }
1621         }
1622
1623         err = reuseport_attach_prog(sk, prog);
1624 err_prog_put:
1625         if (err)
1626                 bpf_prog_put(prog);
1627
1628         return err;
1629 }
1630
1631 void sk_reuseport_prog_free(struct bpf_prog *prog)
1632 {
1633         if (!prog)
1634                 return;
1635
1636         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1637                 bpf_prog_put(prog);
1638         else
1639                 bpf_prog_destroy(prog);
1640 }
1641
1642 struct bpf_scratchpad {
1643         union {
1644                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1645                 u8     buff[MAX_BPF_STACK];
1646         };
1647 };
1648
1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1650
1651 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1652                                           unsigned int write_len)
1653 {
1654         return skb_ensure_writable(skb, write_len);
1655 }
1656
1657 static inline int bpf_try_make_writable(struct sk_buff *skb,
1658                                         unsigned int write_len)
1659 {
1660         int err = __bpf_try_make_writable(skb, write_len);
1661
1662         bpf_compute_data_pointers(skb);
1663         return err;
1664 }
1665
1666 static int bpf_try_make_head_writable(struct sk_buff *skb)
1667 {
1668         return bpf_try_make_writable(skb, skb_headlen(skb));
1669 }
1670
1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1672 {
1673         if (skb_at_tc_ingress(skb))
1674                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1675 }
1676
1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1678 {
1679         if (skb_at_tc_ingress(skb))
1680                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1681 }
1682
1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1684            const void *, from, u32, len, u64, flags)
1685 {
1686         void *ptr;
1687
1688         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1689                 return -EINVAL;
1690         if (unlikely(offset > 0xffff))
1691                 return -EFAULT;
1692         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1693                 return -EFAULT;
1694
1695         ptr = skb->data + offset;
1696         if (flags & BPF_F_RECOMPUTE_CSUM)
1697                 __skb_postpull_rcsum(skb, ptr, len, offset);
1698
1699         memcpy(ptr, from, len);
1700
1701         if (flags & BPF_F_RECOMPUTE_CSUM)
1702                 __skb_postpush_rcsum(skb, ptr, len, offset);
1703         if (flags & BPF_F_INVALIDATE_HASH)
1704                 skb_clear_hash(skb);
1705
1706         return 0;
1707 }
1708
1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1710         .func           = bpf_skb_store_bytes,
1711         .gpl_only       = false,
1712         .ret_type       = RET_INTEGER,
1713         .arg1_type      = ARG_PTR_TO_CTX,
1714         .arg2_type      = ARG_ANYTHING,
1715         .arg3_type      = ARG_PTR_TO_MEM,
1716         .arg4_type      = ARG_CONST_SIZE,
1717         .arg5_type      = ARG_ANYTHING,
1718 };
1719
1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1721            void *, to, u32, len)
1722 {
1723         void *ptr;
1724
1725         if (unlikely(offset > 0xffff))
1726                 goto err_clear;
1727
1728         ptr = skb_header_pointer(skb, offset, len, to);
1729         if (unlikely(!ptr))
1730                 goto err_clear;
1731         if (ptr != to)
1732                 memcpy(to, ptr, len);
1733
1734         return 0;
1735 err_clear:
1736         memset(to, 0, len);
1737         return -EFAULT;
1738 }
1739
1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1741         .func           = bpf_skb_load_bytes,
1742         .gpl_only       = false,
1743         .ret_type       = RET_INTEGER,
1744         .arg1_type      = ARG_PTR_TO_CTX,
1745         .arg2_type      = ARG_ANYTHING,
1746         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1747         .arg4_type      = ARG_CONST_SIZE,
1748 };
1749
1750 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1751            const struct bpf_flow_dissector *, ctx, u32, offset,
1752            void *, to, u32, len)
1753 {
1754         void *ptr;
1755
1756         if (unlikely(offset > 0xffff))
1757                 goto err_clear;
1758
1759         if (unlikely(!ctx->skb))
1760                 goto err_clear;
1761
1762         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1763         if (unlikely(!ptr))
1764                 goto err_clear;
1765         if (ptr != to)
1766                 memcpy(to, ptr, len);
1767
1768         return 0;
1769 err_clear:
1770         memset(to, 0, len);
1771         return -EFAULT;
1772 }
1773
1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1775         .func           = bpf_flow_dissector_load_bytes,
1776         .gpl_only       = false,
1777         .ret_type       = RET_INTEGER,
1778         .arg1_type      = ARG_PTR_TO_CTX,
1779         .arg2_type      = ARG_ANYTHING,
1780         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1781         .arg4_type      = ARG_CONST_SIZE,
1782 };
1783
1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1785            u32, offset, void *, to, u32, len, u32, start_header)
1786 {
1787         u8 *end = skb_tail_pointer(skb);
1788         u8 *start, *ptr;
1789
1790         if (unlikely(offset > 0xffff))
1791                 goto err_clear;
1792
1793         switch (start_header) {
1794         case BPF_HDR_START_MAC:
1795                 if (unlikely(!skb_mac_header_was_set(skb)))
1796                         goto err_clear;
1797                 start = skb_mac_header(skb);
1798                 break;
1799         case BPF_HDR_START_NET:
1800                 start = skb_network_header(skb);
1801                 break;
1802         default:
1803                 goto err_clear;
1804         }
1805
1806         ptr = start + offset;
1807
1808         if (likely(ptr + len <= end)) {
1809                 memcpy(to, ptr, len);
1810                 return 0;
1811         }
1812
1813 err_clear:
1814         memset(to, 0, len);
1815         return -EFAULT;
1816 }
1817
1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1819         .func           = bpf_skb_load_bytes_relative,
1820         .gpl_only       = false,
1821         .ret_type       = RET_INTEGER,
1822         .arg1_type      = ARG_PTR_TO_CTX,
1823         .arg2_type      = ARG_ANYTHING,
1824         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1825         .arg4_type      = ARG_CONST_SIZE,
1826         .arg5_type      = ARG_ANYTHING,
1827 };
1828
1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1830 {
1831         /* Idea is the following: should the needed direct read/write
1832          * test fail during runtime, we can pull in more data and redo
1833          * again, since implicitly, we invalidate previous checks here.
1834          *
1835          * Or, since we know how much we need to make read/writeable,
1836          * this can be done once at the program beginning for direct
1837          * access case. By this we overcome limitations of only current
1838          * headroom being accessible.
1839          */
1840         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1841 }
1842
1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1844         .func           = bpf_skb_pull_data,
1845         .gpl_only       = false,
1846         .ret_type       = RET_INTEGER,
1847         .arg1_type      = ARG_PTR_TO_CTX,
1848         .arg2_type      = ARG_ANYTHING,
1849 };
1850
1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1852 {
1853         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1854 }
1855
1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1857         .func           = bpf_sk_fullsock,
1858         .gpl_only       = false,
1859         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1860         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1861 };
1862
1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1864                                            unsigned int write_len)
1865 {
1866         return __bpf_try_make_writable(skb, write_len);
1867 }
1868
1869 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1870 {
1871         /* Idea is the following: should the needed direct read/write
1872          * test fail during runtime, we can pull in more data and redo
1873          * again, since implicitly, we invalidate previous checks here.
1874          *
1875          * Or, since we know how much we need to make read/writeable,
1876          * this can be done once at the program beginning for direct
1877          * access case. By this we overcome limitations of only current
1878          * headroom being accessible.
1879          */
1880         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1881 }
1882
1883 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1884         .func           = sk_skb_pull_data,
1885         .gpl_only       = false,
1886         .ret_type       = RET_INTEGER,
1887         .arg1_type      = ARG_PTR_TO_CTX,
1888         .arg2_type      = ARG_ANYTHING,
1889 };
1890
1891 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1892            u64, from, u64, to, u64, flags)
1893 {
1894         __sum16 *ptr;
1895
1896         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1897                 return -EINVAL;
1898         if (unlikely(offset > 0xffff || offset & 1))
1899                 return -EFAULT;
1900         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1901                 return -EFAULT;
1902
1903         ptr = (__sum16 *)(skb->data + offset);
1904         switch (flags & BPF_F_HDR_FIELD_MASK) {
1905         case 0:
1906                 if (unlikely(from != 0))
1907                         return -EINVAL;
1908
1909                 csum_replace_by_diff(ptr, to);
1910                 break;
1911         case 2:
1912                 csum_replace2(ptr, from, to);
1913                 break;
1914         case 4:
1915                 csum_replace4(ptr, from, to);
1916                 break;
1917         default:
1918                 return -EINVAL;
1919         }
1920
1921         return 0;
1922 }
1923
1924 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1925         .func           = bpf_l3_csum_replace,
1926         .gpl_only       = false,
1927         .ret_type       = RET_INTEGER,
1928         .arg1_type      = ARG_PTR_TO_CTX,
1929         .arg2_type      = ARG_ANYTHING,
1930         .arg3_type      = ARG_ANYTHING,
1931         .arg4_type      = ARG_ANYTHING,
1932         .arg5_type      = ARG_ANYTHING,
1933 };
1934
1935 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1936            u64, from, u64, to, u64, flags)
1937 {
1938         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1939         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1940         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1941         __sum16 *ptr;
1942
1943         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1944                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1945                 return -EINVAL;
1946         if (unlikely(offset > 0xffff || offset & 1))
1947                 return -EFAULT;
1948         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1949                 return -EFAULT;
1950
1951         ptr = (__sum16 *)(skb->data + offset);
1952         if (is_mmzero && !do_mforce && !*ptr)
1953                 return 0;
1954
1955         switch (flags & BPF_F_HDR_FIELD_MASK) {
1956         case 0:
1957                 if (unlikely(from != 0))
1958                         return -EINVAL;
1959
1960                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1961                 break;
1962         case 2:
1963                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1964                 break;
1965         case 4:
1966                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1967                 break;
1968         default:
1969                 return -EINVAL;
1970         }
1971
1972         if (is_mmzero && !*ptr)
1973                 *ptr = CSUM_MANGLED_0;
1974         return 0;
1975 }
1976
1977 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1978         .func           = bpf_l4_csum_replace,
1979         .gpl_only       = false,
1980         .ret_type       = RET_INTEGER,
1981         .arg1_type      = ARG_PTR_TO_CTX,
1982         .arg2_type      = ARG_ANYTHING,
1983         .arg3_type      = ARG_ANYTHING,
1984         .arg4_type      = ARG_ANYTHING,
1985         .arg5_type      = ARG_ANYTHING,
1986 };
1987
1988 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1989            __be32 *, to, u32, to_size, __wsum, seed)
1990 {
1991         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1992         u32 diff_size = from_size + to_size;
1993         int i, j = 0;
1994
1995         /* This is quite flexible, some examples:
1996          *
1997          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1998          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1999          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2000          *
2001          * Even for diffing, from_size and to_size don't need to be equal.
2002          */
2003         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2004                      diff_size > sizeof(sp->diff)))
2005                 return -EINVAL;
2006
2007         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2008                 sp->diff[j] = ~from[i];
2009         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2010                 sp->diff[j] = to[i];
2011
2012         return csum_partial(sp->diff, diff_size, seed);
2013 }
2014
2015 static const struct bpf_func_proto bpf_csum_diff_proto = {
2016         .func           = bpf_csum_diff,
2017         .gpl_only       = false,
2018         .pkt_access     = true,
2019         .ret_type       = RET_INTEGER,
2020         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
2021         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2022         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
2023         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2024         .arg5_type      = ARG_ANYTHING,
2025 };
2026
2027 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2028 {
2029         /* The interface is to be used in combination with bpf_csum_diff()
2030          * for direct packet writes. csum rotation for alignment as well
2031          * as emulating csum_sub() can be done from the eBPF program.
2032          */
2033         if (skb->ip_summed == CHECKSUM_COMPLETE)
2034                 return (skb->csum = csum_add(skb->csum, csum));
2035
2036         return -ENOTSUPP;
2037 }
2038
2039 static const struct bpf_func_proto bpf_csum_update_proto = {
2040         .func           = bpf_csum_update,
2041         .gpl_only       = false,
2042         .ret_type       = RET_INTEGER,
2043         .arg1_type      = ARG_PTR_TO_CTX,
2044         .arg2_type      = ARG_ANYTHING,
2045 };
2046
2047 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2048 {
2049         /* The interface is to be used in combination with bpf_skb_adjust_room()
2050          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2051          * is passed as flags, for example.
2052          */
2053         switch (level) {
2054         case BPF_CSUM_LEVEL_INC:
2055                 __skb_incr_checksum_unnecessary(skb);
2056                 break;
2057         case BPF_CSUM_LEVEL_DEC:
2058                 __skb_decr_checksum_unnecessary(skb);
2059                 break;
2060         case BPF_CSUM_LEVEL_RESET:
2061                 __skb_reset_checksum_unnecessary(skb);
2062                 break;
2063         case BPF_CSUM_LEVEL_QUERY:
2064                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2065                        skb->csum_level : -EACCES;
2066         default:
2067                 return -EINVAL;
2068         }
2069
2070         return 0;
2071 }
2072
2073 static const struct bpf_func_proto bpf_csum_level_proto = {
2074         .func           = bpf_csum_level,
2075         .gpl_only       = false,
2076         .ret_type       = RET_INTEGER,
2077         .arg1_type      = ARG_PTR_TO_CTX,
2078         .arg2_type      = ARG_ANYTHING,
2079 };
2080
2081 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2082 {
2083         return dev_forward_skb_nomtu(dev, skb);
2084 }
2085
2086 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2087                                       struct sk_buff *skb)
2088 {
2089         int ret = ____dev_forward_skb(dev, skb, false);
2090
2091         if (likely(!ret)) {
2092                 skb->dev = dev;
2093                 ret = netif_rx(skb);
2094         }
2095
2096         return ret;
2097 }
2098
2099 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2100 {
2101         int ret;
2102
2103         if (dev_xmit_recursion()) {
2104                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2105                 kfree_skb(skb);
2106                 return -ENETDOWN;
2107         }
2108
2109         skb->dev = dev;
2110         skb->tstamp = 0;
2111
2112         dev_xmit_recursion_inc();
2113         ret = dev_queue_xmit(skb);
2114         dev_xmit_recursion_dec();
2115
2116         return ret;
2117 }
2118
2119 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2120                                  u32 flags)
2121 {
2122         unsigned int mlen = skb_network_offset(skb);
2123
2124         if (mlen) {
2125                 __skb_pull(skb, mlen);
2126
2127                 /* At ingress, the mac header has already been pulled once.
2128                  * At egress, skb_pospull_rcsum has to be done in case that
2129                  * the skb is originated from ingress (i.e. a forwarded skb)
2130                  * to ensure that rcsum starts at net header.
2131                  */
2132                 if (!skb_at_tc_ingress(skb))
2133                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2134         }
2135         skb_pop_mac_header(skb);
2136         skb_reset_mac_len(skb);
2137         return flags & BPF_F_INGRESS ?
2138                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2139 }
2140
2141 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2142                                  u32 flags)
2143 {
2144         /* Verify that a link layer header is carried */
2145         if (unlikely(skb->mac_header >= skb->network_header)) {
2146                 kfree_skb(skb);
2147                 return -ERANGE;
2148         }
2149
2150         bpf_push_mac_rcsum(skb);
2151         return flags & BPF_F_INGRESS ?
2152                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2153 }
2154
2155 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2156                           u32 flags)
2157 {
2158         if (dev_is_mac_header_xmit(dev))
2159                 return __bpf_redirect_common(skb, dev, flags);
2160         else
2161                 return __bpf_redirect_no_mac(skb, dev, flags);
2162 }
2163
2164 #if IS_ENABLED(CONFIG_IPV6)
2165 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2166                             struct net_device *dev, struct bpf_nh_params *nh)
2167 {
2168         u32 hh_len = LL_RESERVED_SPACE(dev);
2169         const struct in6_addr *nexthop;
2170         struct dst_entry *dst = NULL;
2171         struct neighbour *neigh;
2172
2173         if (dev_xmit_recursion()) {
2174                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2175                 goto out_drop;
2176         }
2177
2178         skb->dev = dev;
2179         skb->tstamp = 0;
2180
2181         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2182                 struct sk_buff *skb2;
2183
2184                 skb2 = skb_realloc_headroom(skb, hh_len);
2185                 if (unlikely(!skb2)) {
2186                         kfree_skb(skb);
2187                         return -ENOMEM;
2188                 }
2189                 if (skb->sk)
2190                         skb_set_owner_w(skb2, skb->sk);
2191                 consume_skb(skb);
2192                 skb = skb2;
2193         }
2194
2195         rcu_read_lock_bh();
2196         if (!nh) {
2197                 dst = skb_dst(skb);
2198                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2199                                       &ipv6_hdr(skb)->daddr);
2200         } else {
2201                 nexthop = &nh->ipv6_nh;
2202         }
2203         neigh = ip_neigh_gw6(dev, nexthop);
2204         if (likely(!IS_ERR(neigh))) {
2205                 int ret;
2206
2207                 sock_confirm_neigh(skb, neigh);
2208                 dev_xmit_recursion_inc();
2209                 ret = neigh_output(neigh, skb, false);
2210                 dev_xmit_recursion_dec();
2211                 rcu_read_unlock_bh();
2212                 return ret;
2213         }
2214         rcu_read_unlock_bh();
2215         if (dst)
2216                 IP6_INC_STATS(dev_net(dst->dev),
2217                               ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2218 out_drop:
2219         kfree_skb(skb);
2220         return -ENETDOWN;
2221 }
2222
2223 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2224                                    struct bpf_nh_params *nh)
2225 {
2226         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2227         struct net *net = dev_net(dev);
2228         int err, ret = NET_XMIT_DROP;
2229
2230         if (!nh) {
2231                 struct dst_entry *dst;
2232                 struct flowi6 fl6 = {
2233                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2234                         .flowi6_mark  = skb->mark,
2235                         .flowlabel    = ip6_flowinfo(ip6h),
2236                         .flowi6_oif   = dev->ifindex,
2237                         .flowi6_proto = ip6h->nexthdr,
2238                         .daddr        = ip6h->daddr,
2239                         .saddr        = ip6h->saddr,
2240                 };
2241
2242                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2243                 if (IS_ERR(dst))
2244                         goto out_drop;
2245
2246                 skb_dst_set(skb, dst);
2247         } else if (nh->nh_family != AF_INET6) {
2248                 goto out_drop;
2249         }
2250
2251         err = bpf_out_neigh_v6(net, skb, dev, nh);
2252         if (unlikely(net_xmit_eval(err)))
2253                 dev->stats.tx_errors++;
2254         else
2255                 ret = NET_XMIT_SUCCESS;
2256         goto out_xmit;
2257 out_drop:
2258         dev->stats.tx_errors++;
2259         kfree_skb(skb);
2260 out_xmit:
2261         return ret;
2262 }
2263 #else
2264 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2265                                    struct bpf_nh_params *nh)
2266 {
2267         kfree_skb(skb);
2268         return NET_XMIT_DROP;
2269 }
2270 #endif /* CONFIG_IPV6 */
2271
2272 #if IS_ENABLED(CONFIG_INET)
2273 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2274                             struct net_device *dev, struct bpf_nh_params *nh)
2275 {
2276         u32 hh_len = LL_RESERVED_SPACE(dev);
2277         struct neighbour *neigh;
2278         bool is_v6gw = false;
2279
2280         if (dev_xmit_recursion()) {
2281                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2282                 goto out_drop;
2283         }
2284
2285         skb->dev = dev;
2286         skb->tstamp = 0;
2287
2288         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2289                 struct sk_buff *skb2;
2290
2291                 skb2 = skb_realloc_headroom(skb, hh_len);
2292                 if (unlikely(!skb2)) {
2293                         kfree_skb(skb);
2294                         return -ENOMEM;
2295                 }
2296                 if (skb->sk)
2297                         skb_set_owner_w(skb2, skb->sk);
2298                 consume_skb(skb);
2299                 skb = skb2;
2300         }
2301
2302         rcu_read_lock_bh();
2303         if (!nh) {
2304                 struct dst_entry *dst = skb_dst(skb);
2305                 struct rtable *rt = container_of(dst, struct rtable, dst);
2306
2307                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2308         } else if (nh->nh_family == AF_INET6) {
2309                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2310                 is_v6gw = true;
2311         } else if (nh->nh_family == AF_INET) {
2312                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2313         } else {
2314                 rcu_read_unlock_bh();
2315                 goto out_drop;
2316         }
2317
2318         if (likely(!IS_ERR(neigh))) {
2319                 int ret;
2320
2321                 sock_confirm_neigh(skb, neigh);
2322                 dev_xmit_recursion_inc();
2323                 ret = neigh_output(neigh, skb, is_v6gw);
2324                 dev_xmit_recursion_dec();
2325                 rcu_read_unlock_bh();
2326                 return ret;
2327         }
2328         rcu_read_unlock_bh();
2329 out_drop:
2330         kfree_skb(skb);
2331         return -ENETDOWN;
2332 }
2333
2334 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2335                                    struct bpf_nh_params *nh)
2336 {
2337         const struct iphdr *ip4h = ip_hdr(skb);
2338         struct net *net = dev_net(dev);
2339         int err, ret = NET_XMIT_DROP;
2340
2341         if (!nh) {
2342                 struct flowi4 fl4 = {
2343                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2344                         .flowi4_mark  = skb->mark,
2345                         .flowi4_tos   = RT_TOS(ip4h->tos),
2346                         .flowi4_oif   = dev->ifindex,
2347                         .flowi4_proto = ip4h->protocol,
2348                         .daddr        = ip4h->daddr,
2349                         .saddr        = ip4h->saddr,
2350                 };
2351                 struct rtable *rt;
2352
2353                 rt = ip_route_output_flow(net, &fl4, NULL);
2354                 if (IS_ERR(rt))
2355                         goto out_drop;
2356                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2357                         ip_rt_put(rt);
2358                         goto out_drop;
2359                 }
2360
2361                 skb_dst_set(skb, &rt->dst);
2362         }
2363
2364         err = bpf_out_neigh_v4(net, skb, dev, nh);
2365         if (unlikely(net_xmit_eval(err)))
2366                 dev->stats.tx_errors++;
2367         else
2368                 ret = NET_XMIT_SUCCESS;
2369         goto out_xmit;
2370 out_drop:
2371         dev->stats.tx_errors++;
2372         kfree_skb(skb);
2373 out_xmit:
2374         return ret;
2375 }
2376 #else
2377 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2378                                    struct bpf_nh_params *nh)
2379 {
2380         kfree_skb(skb);
2381         return NET_XMIT_DROP;
2382 }
2383 #endif /* CONFIG_INET */
2384
2385 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2386                                 struct bpf_nh_params *nh)
2387 {
2388         struct ethhdr *ethh = eth_hdr(skb);
2389
2390         if (unlikely(skb->mac_header >= skb->network_header))
2391                 goto out;
2392         bpf_push_mac_rcsum(skb);
2393         if (is_multicast_ether_addr(ethh->h_dest))
2394                 goto out;
2395
2396         skb_pull(skb, sizeof(*ethh));
2397         skb_unset_mac_header(skb);
2398         skb_reset_network_header(skb);
2399
2400         if (skb->protocol == htons(ETH_P_IP))
2401                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2402         else if (skb->protocol == htons(ETH_P_IPV6))
2403                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2404 out:
2405         kfree_skb(skb);
2406         return -ENOTSUPP;
2407 }
2408
2409 /* Internal, non-exposed redirect flags. */
2410 enum {
2411         BPF_F_NEIGH     = (1ULL << 1),
2412         BPF_F_PEER      = (1ULL << 2),
2413         BPF_F_NEXTHOP   = (1ULL << 3),
2414 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2415 };
2416
2417 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2418 {
2419         struct net_device *dev;
2420         struct sk_buff *clone;
2421         int ret;
2422
2423         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2424                 return -EINVAL;
2425
2426         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2427         if (unlikely(!dev))
2428                 return -EINVAL;
2429
2430         clone = skb_clone(skb, GFP_ATOMIC);
2431         if (unlikely(!clone))
2432                 return -ENOMEM;
2433
2434         /* For direct write, we need to keep the invariant that the skbs
2435          * we're dealing with need to be uncloned. Should uncloning fail
2436          * here, we need to free the just generated clone to unclone once
2437          * again.
2438          */
2439         ret = bpf_try_make_head_writable(skb);
2440         if (unlikely(ret)) {
2441                 kfree_skb(clone);
2442                 return -ENOMEM;
2443         }
2444
2445         return __bpf_redirect(clone, dev, flags);
2446 }
2447
2448 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2449         .func           = bpf_clone_redirect,
2450         .gpl_only       = false,
2451         .ret_type       = RET_INTEGER,
2452         .arg1_type      = ARG_PTR_TO_CTX,
2453         .arg2_type      = ARG_ANYTHING,
2454         .arg3_type      = ARG_ANYTHING,
2455 };
2456
2457 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2458 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2459
2460 int skb_do_redirect(struct sk_buff *skb)
2461 {
2462         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2463         struct net *net = dev_net(skb->dev);
2464         struct net_device *dev;
2465         u32 flags = ri->flags;
2466
2467         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2468         ri->tgt_index = 0;
2469         ri->flags = 0;
2470         if (unlikely(!dev))
2471                 goto out_drop;
2472         if (flags & BPF_F_PEER) {
2473                 const struct net_device_ops *ops = dev->netdev_ops;
2474
2475                 if (unlikely(!ops->ndo_get_peer_dev ||
2476                              !skb_at_tc_ingress(skb)))
2477                         goto out_drop;
2478                 dev = ops->ndo_get_peer_dev(dev);
2479                 if (unlikely(!dev ||
2480                              !(dev->flags & IFF_UP) ||
2481                              net_eq(net, dev_net(dev))))
2482                         goto out_drop;
2483                 skb->dev = dev;
2484                 return -EAGAIN;
2485         }
2486         return flags & BPF_F_NEIGH ?
2487                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2488                                     &ri->nh : NULL) :
2489                __bpf_redirect(skb, dev, flags);
2490 out_drop:
2491         kfree_skb(skb);
2492         return -EINVAL;
2493 }
2494
2495 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2496 {
2497         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2498
2499         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2500                 return TC_ACT_SHOT;
2501
2502         ri->flags = flags;
2503         ri->tgt_index = ifindex;
2504
2505         return TC_ACT_REDIRECT;
2506 }
2507
2508 static const struct bpf_func_proto bpf_redirect_proto = {
2509         .func           = bpf_redirect,
2510         .gpl_only       = false,
2511         .ret_type       = RET_INTEGER,
2512         .arg1_type      = ARG_ANYTHING,
2513         .arg2_type      = ARG_ANYTHING,
2514 };
2515
2516 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2517 {
2518         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2519
2520         if (unlikely(flags))
2521                 return TC_ACT_SHOT;
2522
2523         ri->flags = BPF_F_PEER;
2524         ri->tgt_index = ifindex;
2525
2526         return TC_ACT_REDIRECT;
2527 }
2528
2529 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2530         .func           = bpf_redirect_peer,
2531         .gpl_only       = false,
2532         .ret_type       = RET_INTEGER,
2533         .arg1_type      = ARG_ANYTHING,
2534         .arg2_type      = ARG_ANYTHING,
2535 };
2536
2537 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2538            int, plen, u64, flags)
2539 {
2540         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2541
2542         if (unlikely((plen && plen < sizeof(*params)) || flags))
2543                 return TC_ACT_SHOT;
2544
2545         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2546         ri->tgt_index = ifindex;
2547
2548         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2549         if (plen)
2550                 memcpy(&ri->nh, params, sizeof(ri->nh));
2551
2552         return TC_ACT_REDIRECT;
2553 }
2554
2555 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2556         .func           = bpf_redirect_neigh,
2557         .gpl_only       = false,
2558         .ret_type       = RET_INTEGER,
2559         .arg1_type      = ARG_ANYTHING,
2560         .arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
2561         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2562         .arg4_type      = ARG_ANYTHING,
2563 };
2564
2565 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2566 {
2567         msg->apply_bytes = bytes;
2568         return 0;
2569 }
2570
2571 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2572         .func           = bpf_msg_apply_bytes,
2573         .gpl_only       = false,
2574         .ret_type       = RET_INTEGER,
2575         .arg1_type      = ARG_PTR_TO_CTX,
2576         .arg2_type      = ARG_ANYTHING,
2577 };
2578
2579 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2580 {
2581         msg->cork_bytes = bytes;
2582         return 0;
2583 }
2584
2585 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2586         .func           = bpf_msg_cork_bytes,
2587         .gpl_only       = false,
2588         .ret_type       = RET_INTEGER,
2589         .arg1_type      = ARG_PTR_TO_CTX,
2590         .arg2_type      = ARG_ANYTHING,
2591 };
2592
2593 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2594            u32, end, u64, flags)
2595 {
2596         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2597         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2598         struct scatterlist *sge;
2599         u8 *raw, *to, *from;
2600         struct page *page;
2601
2602         if (unlikely(flags || end <= start))
2603                 return -EINVAL;
2604
2605         /* First find the starting scatterlist element */
2606         i = msg->sg.start;
2607         do {
2608                 offset += len;
2609                 len = sk_msg_elem(msg, i)->length;
2610                 if (start < offset + len)
2611                         break;
2612                 sk_msg_iter_var_next(i);
2613         } while (i != msg->sg.end);
2614
2615         if (unlikely(start >= offset + len))
2616                 return -EINVAL;
2617
2618         first_sge = i;
2619         /* The start may point into the sg element so we need to also
2620          * account for the headroom.
2621          */
2622         bytes_sg_total = start - offset + bytes;
2623         if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len)
2624                 goto out;
2625
2626         /* At this point we need to linearize multiple scatterlist
2627          * elements or a single shared page. Either way we need to
2628          * copy into a linear buffer exclusively owned by BPF. Then
2629          * place the buffer in the scatterlist and fixup the original
2630          * entries by removing the entries now in the linear buffer
2631          * and shifting the remaining entries. For now we do not try
2632          * to copy partial entries to avoid complexity of running out
2633          * of sg_entry slots. The downside is reading a single byte
2634          * will copy the entire sg entry.
2635          */
2636         do {
2637                 copy += sk_msg_elem(msg, i)->length;
2638                 sk_msg_iter_var_next(i);
2639                 if (bytes_sg_total <= copy)
2640                         break;
2641         } while (i != msg->sg.end);
2642         last_sge = i;
2643
2644         if (unlikely(bytes_sg_total > copy))
2645                 return -EINVAL;
2646
2647         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2648                            get_order(copy));
2649         if (unlikely(!page))
2650                 return -ENOMEM;
2651
2652         raw = page_address(page);
2653         i = first_sge;
2654         do {
2655                 sge = sk_msg_elem(msg, i);
2656                 from = sg_virt(sge);
2657                 len = sge->length;
2658                 to = raw + poffset;
2659
2660                 memcpy(to, from, len);
2661                 poffset += len;
2662                 sge->length = 0;
2663                 put_page(sg_page(sge));
2664
2665                 sk_msg_iter_var_next(i);
2666         } while (i != last_sge);
2667
2668         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2669
2670         /* To repair sg ring we need to shift entries. If we only
2671          * had a single entry though we can just replace it and
2672          * be done. Otherwise walk the ring and shift the entries.
2673          */
2674         WARN_ON_ONCE(last_sge == first_sge);
2675         shift = last_sge > first_sge ?
2676                 last_sge - first_sge - 1 :
2677                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2678         if (!shift)
2679                 goto out;
2680
2681         i = first_sge;
2682         sk_msg_iter_var_next(i);
2683         do {
2684                 u32 move_from;
2685
2686                 if (i + shift >= NR_MSG_FRAG_IDS)
2687                         move_from = i + shift - NR_MSG_FRAG_IDS;
2688                 else
2689                         move_from = i + shift;
2690                 if (move_from == msg->sg.end)
2691                         break;
2692
2693                 msg->sg.data[i] = msg->sg.data[move_from];
2694                 msg->sg.data[move_from].length = 0;
2695                 msg->sg.data[move_from].page_link = 0;
2696                 msg->sg.data[move_from].offset = 0;
2697                 sk_msg_iter_var_next(i);
2698         } while (1);
2699
2700         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2701                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2702                       msg->sg.end - shift;
2703 out:
2704         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2705         msg->data_end = msg->data + bytes;
2706         return 0;
2707 }
2708
2709 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2710         .func           = bpf_msg_pull_data,
2711         .gpl_only       = false,
2712         .ret_type       = RET_INTEGER,
2713         .arg1_type      = ARG_PTR_TO_CTX,
2714         .arg2_type      = ARG_ANYTHING,
2715         .arg3_type      = ARG_ANYTHING,
2716         .arg4_type      = ARG_ANYTHING,
2717 };
2718
2719 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2720            u32, len, u64, flags)
2721 {
2722         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2723         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2724         u8 *raw, *to, *from;
2725         struct page *page;
2726
2727         if (unlikely(flags))
2728                 return -EINVAL;
2729
2730         /* First find the starting scatterlist element */
2731         i = msg->sg.start;
2732         do {
2733                 offset += l;
2734                 l = sk_msg_elem(msg, i)->length;
2735
2736                 if (start < offset + l)
2737                         break;
2738                 sk_msg_iter_var_next(i);
2739         } while (i != msg->sg.end);
2740
2741         if (start >= offset + l)
2742                 return -EINVAL;
2743
2744         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2745
2746         /* If no space available will fallback to copy, we need at
2747          * least one scatterlist elem available to push data into
2748          * when start aligns to the beginning of an element or two
2749          * when it falls inside an element. We handle the start equals
2750          * offset case because its the common case for inserting a
2751          * header.
2752          */
2753         if (!space || (space == 1 && start != offset))
2754                 copy = msg->sg.data[i].length;
2755
2756         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2757                            get_order(copy + len));
2758         if (unlikely(!page))
2759                 return -ENOMEM;
2760
2761         if (copy) {
2762                 int front, back;
2763
2764                 raw = page_address(page);
2765
2766                 psge = sk_msg_elem(msg, i);
2767                 front = start - offset;
2768                 back = psge->length - front;
2769                 from = sg_virt(psge);
2770
2771                 if (front)
2772                         memcpy(raw, from, front);
2773
2774                 if (back) {
2775                         from += front;
2776                         to = raw + front + len;
2777
2778                         memcpy(to, from, back);
2779                 }
2780
2781                 put_page(sg_page(psge));
2782         } else if (start - offset) {
2783                 psge = sk_msg_elem(msg, i);
2784                 rsge = sk_msg_elem_cpy(msg, i);
2785
2786                 psge->length = start - offset;
2787                 rsge.length -= psge->length;
2788                 rsge.offset += start;
2789
2790                 sk_msg_iter_var_next(i);
2791                 sg_unmark_end(psge);
2792                 sg_unmark_end(&rsge);
2793                 sk_msg_iter_next(msg, end);
2794         }
2795
2796         /* Slot(s) to place newly allocated data */
2797         new = i;
2798
2799         /* Shift one or two slots as needed */
2800         if (!copy) {
2801                 sge = sk_msg_elem_cpy(msg, i);
2802
2803                 sk_msg_iter_var_next(i);
2804                 sg_unmark_end(&sge);
2805                 sk_msg_iter_next(msg, end);
2806
2807                 nsge = sk_msg_elem_cpy(msg, i);
2808                 if (rsge.length) {
2809                         sk_msg_iter_var_next(i);
2810                         nnsge = sk_msg_elem_cpy(msg, i);
2811                 }
2812
2813                 while (i != msg->sg.end) {
2814                         msg->sg.data[i] = sge;
2815                         sge = nsge;
2816                         sk_msg_iter_var_next(i);
2817                         if (rsge.length) {
2818                                 nsge = nnsge;
2819                                 nnsge = sk_msg_elem_cpy(msg, i);
2820                         } else {
2821                                 nsge = sk_msg_elem_cpy(msg, i);
2822                         }
2823                 }
2824         }
2825
2826         /* Place newly allocated data buffer */
2827         sk_mem_charge(msg->sk, len);
2828         msg->sg.size += len;
2829         __clear_bit(new, &msg->sg.copy);
2830         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2831         if (rsge.length) {
2832                 get_page(sg_page(&rsge));
2833                 sk_msg_iter_var_next(new);
2834                 msg->sg.data[new] = rsge;
2835         }
2836
2837         sk_msg_compute_data_pointers(msg);
2838         return 0;
2839 }
2840
2841 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2842         .func           = bpf_msg_push_data,
2843         .gpl_only       = false,
2844         .ret_type       = RET_INTEGER,
2845         .arg1_type      = ARG_PTR_TO_CTX,
2846         .arg2_type      = ARG_ANYTHING,
2847         .arg3_type      = ARG_ANYTHING,
2848         .arg4_type      = ARG_ANYTHING,
2849 };
2850
2851 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2852 {
2853         int prev;
2854
2855         do {
2856                 prev = i;
2857                 sk_msg_iter_var_next(i);
2858                 msg->sg.data[prev] = msg->sg.data[i];
2859         } while (i != msg->sg.end);
2860
2861         sk_msg_iter_prev(msg, end);
2862 }
2863
2864 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2865 {
2866         struct scatterlist tmp, sge;
2867
2868         sk_msg_iter_next(msg, end);
2869         sge = sk_msg_elem_cpy(msg, i);
2870         sk_msg_iter_var_next(i);
2871         tmp = sk_msg_elem_cpy(msg, i);
2872
2873         while (i != msg->sg.end) {
2874                 msg->sg.data[i] = sge;
2875                 sk_msg_iter_var_next(i);
2876                 sge = tmp;
2877                 tmp = sk_msg_elem_cpy(msg, i);
2878         }
2879 }
2880
2881 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2882            u32, len, u64, flags)
2883 {
2884         u32 i = 0, l = 0, space, offset = 0;
2885         u64 last = start + len;
2886         int pop;
2887
2888         if (unlikely(flags))
2889                 return -EINVAL;
2890
2891         /* First find the starting scatterlist element */
2892         i = msg->sg.start;
2893         do {
2894                 offset += l;
2895                 l = sk_msg_elem(msg, i)->length;
2896
2897                 if (start < offset + l)
2898                         break;
2899                 sk_msg_iter_var_next(i);
2900         } while (i != msg->sg.end);
2901
2902         /* Bounds checks: start and pop must be inside message */
2903         if (start >= offset + l || last >= msg->sg.size)
2904                 return -EINVAL;
2905
2906         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2907
2908         pop = len;
2909         /* --------------| offset
2910          * -| start      |-------- len -------|
2911          *
2912          *  |----- a ----|-------- pop -------|----- b ----|
2913          *  |______________________________________________| length
2914          *
2915          *
2916          * a:   region at front of scatter element to save
2917          * b:   region at back of scatter element to save when length > A + pop
2918          * pop: region to pop from element, same as input 'pop' here will be
2919          *      decremented below per iteration.
2920          *
2921          * Two top-level cases to handle when start != offset, first B is non
2922          * zero and second B is zero corresponding to when a pop includes more
2923          * than one element.
2924          *
2925          * Then if B is non-zero AND there is no space allocate space and
2926          * compact A, B regions into page. If there is space shift ring to
2927          * the rigth free'ing the next element in ring to place B, leaving
2928          * A untouched except to reduce length.
2929          */
2930         if (start != offset) {
2931                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2932                 int a = start;
2933                 int b = sge->length - pop - a;
2934
2935                 sk_msg_iter_var_next(i);
2936
2937                 if (pop < sge->length - a) {
2938                         if (space) {
2939                                 sge->length = a;
2940                                 sk_msg_shift_right(msg, i);
2941                                 nsge = sk_msg_elem(msg, i);
2942                                 get_page(sg_page(sge));
2943                                 sg_set_page(nsge,
2944                                             sg_page(sge),
2945                                             b, sge->offset + pop + a);
2946                         } else {
2947                                 struct page *page, *orig;
2948                                 u8 *to, *from;
2949
2950                                 page = alloc_pages(__GFP_NOWARN |
2951                                                    __GFP_COMP   | GFP_ATOMIC,
2952                                                    get_order(a + b));
2953                                 if (unlikely(!page))
2954                                         return -ENOMEM;
2955
2956                                 sge->length = a;
2957                                 orig = sg_page(sge);
2958                                 from = sg_virt(sge);
2959                                 to = page_address(page);
2960                                 memcpy(to, from, a);
2961                                 memcpy(to + a, from + a + pop, b);
2962                                 sg_set_page(sge, page, a + b, 0);
2963                                 put_page(orig);
2964                         }
2965                         pop = 0;
2966                 } else if (pop >= sge->length - a) {
2967                         pop -= (sge->length - a);
2968                         sge->length = a;
2969                 }
2970         }
2971
2972         /* From above the current layout _must_ be as follows,
2973          *
2974          * -| offset
2975          * -| start
2976          *
2977          *  |---- pop ---|---------------- b ------------|
2978          *  |____________________________________________| length
2979          *
2980          * Offset and start of the current msg elem are equal because in the
2981          * previous case we handled offset != start and either consumed the
2982          * entire element and advanced to the next element OR pop == 0.
2983          *
2984          * Two cases to handle here are first pop is less than the length
2985          * leaving some remainder b above. Simply adjust the element's layout
2986          * in this case. Or pop >= length of the element so that b = 0. In this
2987          * case advance to next element decrementing pop.
2988          */
2989         while (pop) {
2990                 struct scatterlist *sge = sk_msg_elem(msg, i);
2991
2992                 if (pop < sge->length) {
2993                         sge->length -= pop;
2994                         sge->offset += pop;
2995                         pop = 0;
2996                 } else {
2997                         pop -= sge->length;
2998                         sk_msg_shift_left(msg, i);
2999                 }
3000                 sk_msg_iter_var_next(i);
3001         }
3002
3003         sk_mem_uncharge(msg->sk, len - pop);
3004         msg->sg.size -= (len - pop);
3005         sk_msg_compute_data_pointers(msg);
3006         return 0;
3007 }
3008
3009 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
3010         .func           = bpf_msg_pop_data,
3011         .gpl_only       = false,
3012         .ret_type       = RET_INTEGER,
3013         .arg1_type      = ARG_PTR_TO_CTX,
3014         .arg2_type      = ARG_ANYTHING,
3015         .arg3_type      = ARG_ANYTHING,
3016         .arg4_type      = ARG_ANYTHING,
3017 };
3018
3019 #ifdef CONFIG_CGROUP_NET_CLASSID
3020 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3021 {
3022         return __task_get_classid(current);
3023 }
3024
3025 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3026         .func           = bpf_get_cgroup_classid_curr,
3027         .gpl_only       = false,
3028         .ret_type       = RET_INTEGER,
3029 };
3030
3031 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3032 {
3033         struct sock *sk = skb_to_full_sk(skb);
3034
3035         if (!sk || !sk_fullsock(sk))
3036                 return 0;
3037
3038         return sock_cgroup_classid(&sk->sk_cgrp_data);
3039 }
3040
3041 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3042         .func           = bpf_skb_cgroup_classid,
3043         .gpl_only       = false,
3044         .ret_type       = RET_INTEGER,
3045         .arg1_type      = ARG_PTR_TO_CTX,
3046 };
3047 #endif
3048
3049 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3050 {
3051         return task_get_classid(skb);
3052 }
3053
3054 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3055         .func           = bpf_get_cgroup_classid,
3056         .gpl_only       = false,
3057         .ret_type       = RET_INTEGER,
3058         .arg1_type      = ARG_PTR_TO_CTX,
3059 };
3060
3061 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3062 {
3063         return dst_tclassid(skb);
3064 }
3065
3066 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3067         .func           = bpf_get_route_realm,
3068         .gpl_only       = false,
3069         .ret_type       = RET_INTEGER,
3070         .arg1_type      = ARG_PTR_TO_CTX,
3071 };
3072
3073 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3074 {
3075         /* If skb_clear_hash() was called due to mangling, we can
3076          * trigger SW recalculation here. Later access to hash
3077          * can then use the inline skb->hash via context directly
3078          * instead of calling this helper again.
3079          */
3080         return skb_get_hash(skb);
3081 }
3082
3083 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3084         .func           = bpf_get_hash_recalc,
3085         .gpl_only       = false,
3086         .ret_type       = RET_INTEGER,
3087         .arg1_type      = ARG_PTR_TO_CTX,
3088 };
3089
3090 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3091 {
3092         /* After all direct packet write, this can be used once for
3093          * triggering a lazy recalc on next skb_get_hash() invocation.
3094          */
3095         skb_clear_hash(skb);
3096         return 0;
3097 }
3098
3099 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3100         .func           = bpf_set_hash_invalid,
3101         .gpl_only       = false,
3102         .ret_type       = RET_INTEGER,
3103         .arg1_type      = ARG_PTR_TO_CTX,
3104 };
3105
3106 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3107 {
3108         /* Set user specified hash as L4(+), so that it gets returned
3109          * on skb_get_hash() call unless BPF prog later on triggers a
3110          * skb_clear_hash().
3111          */
3112         __skb_set_sw_hash(skb, hash, true);
3113         return 0;
3114 }
3115
3116 static const struct bpf_func_proto bpf_set_hash_proto = {
3117         .func           = bpf_set_hash,
3118         .gpl_only       = false,
3119         .ret_type       = RET_INTEGER,
3120         .arg1_type      = ARG_PTR_TO_CTX,
3121         .arg2_type      = ARG_ANYTHING,
3122 };
3123
3124 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3125            u16, vlan_tci)
3126 {
3127         int ret;
3128
3129         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3130                      vlan_proto != htons(ETH_P_8021AD)))
3131                 vlan_proto = htons(ETH_P_8021Q);
3132
3133         bpf_push_mac_rcsum(skb);
3134         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3135         bpf_pull_mac_rcsum(skb);
3136
3137         bpf_compute_data_pointers(skb);
3138         return ret;
3139 }
3140
3141 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3142         .func           = bpf_skb_vlan_push,
3143         .gpl_only       = false,
3144         .ret_type       = RET_INTEGER,
3145         .arg1_type      = ARG_PTR_TO_CTX,
3146         .arg2_type      = ARG_ANYTHING,
3147         .arg3_type      = ARG_ANYTHING,
3148 };
3149
3150 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3151 {
3152         int ret;
3153
3154         bpf_push_mac_rcsum(skb);
3155         ret = skb_vlan_pop(skb);
3156         bpf_pull_mac_rcsum(skb);
3157
3158         bpf_compute_data_pointers(skb);
3159         return ret;
3160 }
3161
3162 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3163         .func           = bpf_skb_vlan_pop,
3164         .gpl_only       = false,
3165         .ret_type       = RET_INTEGER,
3166         .arg1_type      = ARG_PTR_TO_CTX,
3167 };
3168
3169 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3170 {
3171         /* Caller already did skb_cow() with len as headroom,
3172          * so no need to do it here.
3173          */
3174         skb_push(skb, len);
3175         memmove(skb->data, skb->data + len, off);
3176         memset(skb->data + off, 0, len);
3177
3178         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3179          * needed here as it does not change the skb->csum
3180          * result for checksum complete when summing over
3181          * zeroed blocks.
3182          */
3183         return 0;
3184 }
3185
3186 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3187 {
3188         /* skb_ensure_writable() is not needed here, as we're
3189          * already working on an uncloned skb.
3190          */
3191         if (unlikely(!pskb_may_pull(skb, off + len)))
3192                 return -ENOMEM;
3193
3194         skb_postpull_rcsum(skb, skb->data + off, len);
3195         memmove(skb->data + len, skb->data, off);
3196         __skb_pull(skb, len);
3197
3198         return 0;
3199 }
3200
3201 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3202 {
3203         bool trans_same = skb->transport_header == skb->network_header;
3204         int ret;
3205
3206         /* There's no need for __skb_push()/__skb_pull() pair to
3207          * get to the start of the mac header as we're guaranteed
3208          * to always start from here under eBPF.
3209          */
3210         ret = bpf_skb_generic_push(skb, off, len);
3211         if (likely(!ret)) {
3212                 skb->mac_header -= len;
3213                 skb->network_header -= len;
3214                 if (trans_same)
3215                         skb->transport_header = skb->network_header;
3216         }
3217
3218         return ret;
3219 }
3220
3221 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3222 {
3223         bool trans_same = skb->transport_header == skb->network_header;
3224         int ret;
3225
3226         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3227         ret = bpf_skb_generic_pop(skb, off, len);
3228         if (likely(!ret)) {
3229                 skb->mac_header += len;
3230                 skb->network_header += len;
3231                 if (trans_same)
3232                         skb->transport_header = skb->network_header;
3233         }
3234
3235         return ret;
3236 }
3237
3238 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3239 {
3240         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3241         u32 off = skb_mac_header_len(skb);
3242         int ret;
3243
3244         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3245                 return -ENOTSUPP;
3246
3247         ret = skb_cow(skb, len_diff);
3248         if (unlikely(ret < 0))
3249                 return ret;
3250
3251         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3252         if (unlikely(ret < 0))
3253                 return ret;
3254
3255         if (skb_is_gso(skb)) {
3256                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3257
3258                 /* SKB_GSO_TCPV4 needs to be changed into
3259                  * SKB_GSO_TCPV6.
3260                  */
3261                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3262                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3263                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3264                 }
3265
3266                 /* Due to IPv6 header, MSS needs to be downgraded. */
3267                 skb_decrease_gso_size(shinfo, len_diff);
3268                 /* Header must be checked, and gso_segs recomputed. */
3269                 shinfo->gso_type |= SKB_GSO_DODGY;
3270                 shinfo->gso_segs = 0;
3271         }
3272
3273         skb->protocol = htons(ETH_P_IPV6);
3274         skb_clear_hash(skb);
3275
3276         return 0;
3277 }
3278
3279 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3280 {
3281         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3282         u32 off = skb_mac_header_len(skb);
3283         int ret;
3284
3285         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb))
3286                 return -ENOTSUPP;
3287
3288         ret = skb_unclone(skb, GFP_ATOMIC);
3289         if (unlikely(ret < 0))
3290                 return ret;
3291
3292         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3293         if (unlikely(ret < 0))
3294                 return ret;
3295
3296         if (skb_is_gso(skb)) {
3297                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3298
3299                 /* SKB_GSO_TCPV6 needs to be changed into
3300                  * SKB_GSO_TCPV4.
3301                  */
3302                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3303                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3304                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3305                 }
3306
3307                 /* Due to IPv4 header, MSS can be upgraded. */
3308                 skb_increase_gso_size(shinfo, len_diff);
3309                 /* Header must be checked, and gso_segs recomputed. */
3310                 shinfo->gso_type |= SKB_GSO_DODGY;
3311                 shinfo->gso_segs = 0;
3312         }
3313
3314         skb->protocol = htons(ETH_P_IP);
3315         skb_clear_hash(skb);
3316
3317         return 0;
3318 }
3319
3320 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3321 {
3322         __be16 from_proto = skb->protocol;
3323
3324         if (from_proto == htons(ETH_P_IP) &&
3325               to_proto == htons(ETH_P_IPV6))
3326                 return bpf_skb_proto_4_to_6(skb);
3327
3328         if (from_proto == htons(ETH_P_IPV6) &&
3329               to_proto == htons(ETH_P_IP))
3330                 return bpf_skb_proto_6_to_4(skb);
3331
3332         return -ENOTSUPP;
3333 }
3334
3335 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3336            u64, flags)
3337 {
3338         int ret;
3339
3340         if (unlikely(flags))
3341                 return -EINVAL;
3342
3343         /* General idea is that this helper does the basic groundwork
3344          * needed for changing the protocol, and eBPF program fills the
3345          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3346          * and other helpers, rather than passing a raw buffer here.
3347          *
3348          * The rationale is to keep this minimal and without a need to
3349          * deal with raw packet data. F.e. even if we would pass buffers
3350          * here, the program still needs to call the bpf_lX_csum_replace()
3351          * helpers anyway. Plus, this way we keep also separation of
3352          * concerns, since f.e. bpf_skb_store_bytes() should only take
3353          * care of stores.
3354          *
3355          * Currently, additional options and extension header space are
3356          * not supported, but flags register is reserved so we can adapt
3357          * that. For offloads, we mark packet as dodgy, so that headers
3358          * need to be verified first.
3359          */
3360         ret = bpf_skb_proto_xlat(skb, proto);
3361         bpf_compute_data_pointers(skb);
3362         return ret;
3363 }
3364
3365 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3366         .func           = bpf_skb_change_proto,
3367         .gpl_only       = false,
3368         .ret_type       = RET_INTEGER,
3369         .arg1_type      = ARG_PTR_TO_CTX,
3370         .arg2_type      = ARG_ANYTHING,
3371         .arg3_type      = ARG_ANYTHING,
3372 };
3373
3374 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3375 {
3376         /* We only allow a restricted subset to be changed for now. */
3377         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3378                      !skb_pkt_type_ok(pkt_type)))
3379                 return -EINVAL;
3380
3381         skb->pkt_type = pkt_type;
3382         return 0;
3383 }
3384
3385 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3386         .func           = bpf_skb_change_type,
3387         .gpl_only       = false,
3388         .ret_type       = RET_INTEGER,
3389         .arg1_type      = ARG_PTR_TO_CTX,
3390         .arg2_type      = ARG_ANYTHING,
3391 };
3392
3393 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3394 {
3395         switch (skb->protocol) {
3396         case htons(ETH_P_IP):
3397                 return sizeof(struct iphdr);
3398         case htons(ETH_P_IPV6):
3399                 return sizeof(struct ipv6hdr);
3400         default:
3401                 return ~0U;
3402         }
3403 }
3404
3405 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3406                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3407
3408 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3409                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3410                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3411                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3412                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3413                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3414                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3415
3416 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3417                             u64 flags)
3418 {
3419         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3420         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3421         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3422         unsigned int gso_type = SKB_GSO_DODGY;
3423         int ret;
3424
3425         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3426                 /* udp gso_size delineates datagrams, only allow if fixed */
3427                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3428                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3429                         return -ENOTSUPP;
3430         }
3431
3432         ret = skb_cow_head(skb, len_diff);
3433         if (unlikely(ret < 0))
3434                 return ret;
3435
3436         if (encap) {
3437                 if (skb->protocol != htons(ETH_P_IP) &&
3438                     skb->protocol != htons(ETH_P_IPV6))
3439                         return -ENOTSUPP;
3440
3441                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3442                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3443                         return -EINVAL;
3444
3445                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3446                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3447                         return -EINVAL;
3448
3449                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3450                     inner_mac_len < ETH_HLEN)
3451                         return -EINVAL;
3452
3453                 if (skb->encapsulation)
3454                         return -EALREADY;
3455
3456                 mac_len = skb->network_header - skb->mac_header;
3457                 inner_net = skb->network_header;
3458                 if (inner_mac_len > len_diff)
3459                         return -EINVAL;
3460                 inner_trans = skb->transport_header;
3461         }
3462
3463         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3464         if (unlikely(ret < 0))
3465                 return ret;
3466
3467         if (encap) {
3468                 skb->inner_mac_header = inner_net - inner_mac_len;
3469                 skb->inner_network_header = inner_net;
3470                 skb->inner_transport_header = inner_trans;
3471
3472                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3473                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3474                 else
3475                         skb_set_inner_protocol(skb, skb->protocol);
3476
3477                 skb->encapsulation = 1;
3478                 skb_set_network_header(skb, mac_len);
3479
3480                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3481                         gso_type |= SKB_GSO_UDP_TUNNEL;
3482                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3483                         gso_type |= SKB_GSO_GRE;
3484                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3485                         gso_type |= SKB_GSO_IPXIP6;
3486                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3487                         gso_type |= SKB_GSO_IPXIP4;
3488
3489                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3490                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3491                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3492                                         sizeof(struct ipv6hdr) :
3493                                         sizeof(struct iphdr);
3494
3495                         skb_set_transport_header(skb, mac_len + nh_len);
3496                 }
3497
3498                 /* Match skb->protocol to new outer l3 protocol */
3499                 if (skb->protocol == htons(ETH_P_IP) &&
3500                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3501                         skb->protocol = htons(ETH_P_IPV6);
3502                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3503                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3504                         skb->protocol = htons(ETH_P_IP);
3505         }
3506
3507         if (skb_is_gso(skb)) {
3508                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3509
3510                 /* Due to header grow, MSS needs to be downgraded. */
3511                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3512                         skb_decrease_gso_size(shinfo, len_diff);
3513
3514                 /* Header must be checked, and gso_segs recomputed. */
3515                 shinfo->gso_type |= gso_type;
3516                 shinfo->gso_segs = 0;
3517         }
3518
3519         return 0;
3520 }
3521
3522 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3523                               u64 flags)
3524 {
3525         int ret;
3526
3527         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3528                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3529                 return -EINVAL;
3530
3531         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3532                 /* udp gso_size delineates datagrams, only allow if fixed */
3533                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3534                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3535                         return -ENOTSUPP;
3536         }
3537
3538         ret = skb_unclone(skb, GFP_ATOMIC);
3539         if (unlikely(ret < 0))
3540                 return ret;
3541
3542         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3543         if (unlikely(ret < 0))
3544                 return ret;
3545
3546         if (skb_is_gso(skb)) {
3547                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3548
3549                 /* Due to header shrink, MSS can be upgraded. */
3550                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3551                         skb_increase_gso_size(shinfo, len_diff);
3552
3553                 /* Header must be checked, and gso_segs recomputed. */
3554                 shinfo->gso_type |= SKB_GSO_DODGY;
3555                 shinfo->gso_segs = 0;
3556         }
3557
3558         return 0;
3559 }
3560
3561 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3562
3563 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3564            u32, mode, u64, flags)
3565 {
3566         u32 len_diff_abs = abs(len_diff);
3567         bool shrink = len_diff < 0;
3568         int ret = 0;
3569
3570         if (unlikely(flags || mode))
3571                 return -EINVAL;
3572         if (unlikely(len_diff_abs > 0xfffU))
3573                 return -EFAULT;
3574
3575         if (!shrink) {
3576                 ret = skb_cow(skb, len_diff);
3577                 if (unlikely(ret < 0))
3578                         return ret;
3579                 __skb_push(skb, len_diff_abs);
3580                 memset(skb->data, 0, len_diff_abs);
3581         } else {
3582                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3583                         return -ENOMEM;
3584                 __skb_pull(skb, len_diff_abs);
3585         }
3586         if (tls_sw_has_ctx_rx(skb->sk)) {
3587                 struct strp_msg *rxm = strp_msg(skb);
3588
3589                 rxm->full_len += len_diff;
3590         }
3591         return ret;
3592 }
3593
3594 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3595         .func           = sk_skb_adjust_room,
3596         .gpl_only       = false,
3597         .ret_type       = RET_INTEGER,
3598         .arg1_type      = ARG_PTR_TO_CTX,
3599         .arg2_type      = ARG_ANYTHING,
3600         .arg3_type      = ARG_ANYTHING,
3601         .arg4_type      = ARG_ANYTHING,
3602 };
3603
3604 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3605            u32, mode, u64, flags)
3606 {
3607         u32 len_cur, len_diff_abs = abs(len_diff);
3608         u32 len_min = bpf_skb_net_base_len(skb);
3609         u32 len_max = BPF_SKB_MAX_LEN;
3610         __be16 proto = skb->protocol;
3611         bool shrink = len_diff < 0;
3612         u32 off;
3613         int ret;
3614
3615         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3616                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3617                 return -EINVAL;
3618         if (unlikely(len_diff_abs > 0xfffU))
3619                 return -EFAULT;
3620         if (unlikely(proto != htons(ETH_P_IP) &&
3621                      proto != htons(ETH_P_IPV6)))
3622                 return -ENOTSUPP;
3623
3624         off = skb_mac_header_len(skb);
3625         switch (mode) {
3626         case BPF_ADJ_ROOM_NET:
3627                 off += bpf_skb_net_base_len(skb);
3628                 break;
3629         case BPF_ADJ_ROOM_MAC:
3630                 break;
3631         default:
3632                 return -ENOTSUPP;
3633         }
3634
3635         len_cur = skb->len - skb_network_offset(skb);
3636         if ((shrink && (len_diff_abs >= len_cur ||
3637                         len_cur - len_diff_abs < len_min)) ||
3638             (!shrink && (skb->len + len_diff_abs > len_max &&
3639                          !skb_is_gso(skb))))
3640                 return -ENOTSUPP;
3641
3642         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3643                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3644         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3645                 __skb_reset_checksum_unnecessary(skb);
3646
3647         bpf_compute_data_pointers(skb);
3648         return ret;
3649 }
3650
3651 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3652         .func           = bpf_skb_adjust_room,
3653         .gpl_only       = false,
3654         .ret_type       = RET_INTEGER,
3655         .arg1_type      = ARG_PTR_TO_CTX,
3656         .arg2_type      = ARG_ANYTHING,
3657         .arg3_type      = ARG_ANYTHING,
3658         .arg4_type      = ARG_ANYTHING,
3659 };
3660
3661 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3662 {
3663         u32 min_len = skb_network_offset(skb);
3664
3665         if (skb_transport_header_was_set(skb))
3666                 min_len = skb_transport_offset(skb);
3667         if (skb->ip_summed == CHECKSUM_PARTIAL)
3668                 min_len = skb_checksum_start_offset(skb) +
3669                           skb->csum_offset + sizeof(__sum16);
3670         return min_len;
3671 }
3672
3673 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3674 {
3675         unsigned int old_len = skb->len;
3676         int ret;
3677
3678         ret = __skb_grow_rcsum(skb, new_len);
3679         if (!ret)
3680                 memset(skb->data + old_len, 0, new_len - old_len);
3681         return ret;
3682 }
3683
3684 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3685 {
3686         return __skb_trim_rcsum(skb, new_len);
3687 }
3688
3689 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3690                                         u64 flags)
3691 {
3692         u32 max_len = BPF_SKB_MAX_LEN;
3693         u32 min_len = __bpf_skb_min_len(skb);
3694         int ret;
3695
3696         if (unlikely(flags || new_len > max_len || new_len < min_len))
3697                 return -EINVAL;
3698         if (skb->encapsulation)
3699                 return -ENOTSUPP;
3700
3701         /* The basic idea of this helper is that it's performing the
3702          * needed work to either grow or trim an skb, and eBPF program
3703          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3704          * bpf_lX_csum_replace() and others rather than passing a raw
3705          * buffer here. This one is a slow path helper and intended
3706          * for replies with control messages.
3707          *
3708          * Like in bpf_skb_change_proto(), we want to keep this rather
3709          * minimal and without protocol specifics so that we are able
3710          * to separate concerns as in bpf_skb_store_bytes() should only
3711          * be the one responsible for writing buffers.
3712          *
3713          * It's really expected to be a slow path operation here for
3714          * control message replies, so we're implicitly linearizing,
3715          * uncloning and drop offloads from the skb by this.
3716          */
3717         ret = __bpf_try_make_writable(skb, skb->len);
3718         if (!ret) {
3719                 if (new_len > skb->len)
3720                         ret = bpf_skb_grow_rcsum(skb, new_len);
3721                 else if (new_len < skb->len)
3722                         ret = bpf_skb_trim_rcsum(skb, new_len);
3723                 if (!ret && skb_is_gso(skb))
3724                         skb_gso_reset(skb);
3725         }
3726         return ret;
3727 }
3728
3729 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3730            u64, flags)
3731 {
3732         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3733
3734         bpf_compute_data_pointers(skb);
3735         return ret;
3736 }
3737
3738 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3739         .func           = bpf_skb_change_tail,
3740         .gpl_only       = false,
3741         .ret_type       = RET_INTEGER,
3742         .arg1_type      = ARG_PTR_TO_CTX,
3743         .arg2_type      = ARG_ANYTHING,
3744         .arg3_type      = ARG_ANYTHING,
3745 };
3746
3747 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3748            u64, flags)
3749 {
3750         return __bpf_skb_change_tail(skb, new_len, flags);
3751 }
3752
3753 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3754         .func           = sk_skb_change_tail,
3755         .gpl_only       = false,
3756         .ret_type       = RET_INTEGER,
3757         .arg1_type      = ARG_PTR_TO_CTX,
3758         .arg2_type      = ARG_ANYTHING,
3759         .arg3_type      = ARG_ANYTHING,
3760 };
3761
3762 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3763                                         u64 flags)
3764 {
3765         u32 max_len = BPF_SKB_MAX_LEN;
3766         u32 new_len = skb->len + head_room;
3767         int ret;
3768
3769         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3770                      new_len < skb->len))
3771                 return -EINVAL;
3772
3773         ret = skb_cow(skb, head_room);
3774         if (likely(!ret)) {
3775                 /* Idea for this helper is that we currently only
3776                  * allow to expand on mac header. This means that
3777                  * skb->protocol network header, etc, stay as is.
3778                  * Compared to bpf_skb_change_tail(), we're more
3779                  * flexible due to not needing to linearize or
3780                  * reset GSO. Intention for this helper is to be
3781                  * used by an L3 skb that needs to push mac header
3782                  * for redirection into L2 device.
3783                  */
3784                 __skb_push(skb, head_room);
3785                 memset(skb->data, 0, head_room);
3786                 skb_reset_mac_header(skb);
3787                 skb_reset_mac_len(skb);
3788         }
3789
3790         return ret;
3791 }
3792
3793 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3794            u64, flags)
3795 {
3796         int ret = __bpf_skb_change_head(skb, head_room, flags);
3797
3798         bpf_compute_data_pointers(skb);
3799         return ret;
3800 }
3801
3802 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3803         .func           = bpf_skb_change_head,
3804         .gpl_only       = false,
3805         .ret_type       = RET_INTEGER,
3806         .arg1_type      = ARG_PTR_TO_CTX,
3807         .arg2_type      = ARG_ANYTHING,
3808         .arg3_type      = ARG_ANYTHING,
3809 };
3810
3811 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3812            u64, flags)
3813 {
3814         return __bpf_skb_change_head(skb, head_room, flags);
3815 }
3816
3817 static const struct bpf_func_proto sk_skb_change_head_proto = {
3818         .func           = sk_skb_change_head,
3819         .gpl_only       = false,
3820         .ret_type       = RET_INTEGER,
3821         .arg1_type      = ARG_PTR_TO_CTX,
3822         .arg2_type      = ARG_ANYTHING,
3823         .arg3_type      = ARG_ANYTHING,
3824 };
3825 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3826 {
3827         return xdp_data_meta_unsupported(xdp) ? 0 :
3828                xdp->data - xdp->data_meta;
3829 }
3830
3831 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3832 {
3833         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3834         unsigned long metalen = xdp_get_metalen(xdp);
3835         void *data_start = xdp_frame_end + metalen;
3836         void *data = xdp->data + offset;
3837
3838         if (unlikely(data < data_start ||
3839                      data > xdp->data_end - ETH_HLEN))
3840                 return -EINVAL;
3841
3842         if (metalen)
3843                 memmove(xdp->data_meta + offset,
3844                         xdp->data_meta, metalen);
3845         xdp->data_meta += offset;
3846         xdp->data = data;
3847
3848         return 0;
3849 }
3850
3851 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3852         .func           = bpf_xdp_adjust_head,
3853         .gpl_only       = false,
3854         .ret_type       = RET_INTEGER,
3855         .arg1_type      = ARG_PTR_TO_CTX,
3856         .arg2_type      = ARG_ANYTHING,
3857 };
3858
3859 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3860 {
3861         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3862         void *data_end = xdp->data_end + offset;
3863
3864         /* Notice that xdp_data_hard_end have reserved some tailroom */
3865         if (unlikely(data_end > data_hard_end))
3866                 return -EINVAL;
3867
3868         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3869         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3870                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3871                 return -EINVAL;
3872         }
3873
3874         if (unlikely(data_end < xdp->data + ETH_HLEN))
3875                 return -EINVAL;
3876
3877         /* Clear memory area on grow, can contain uninit kernel memory */
3878         if (offset > 0)
3879                 memset(xdp->data_end, 0, offset);
3880
3881         xdp->data_end = data_end;
3882
3883         return 0;
3884 }
3885
3886 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3887         .func           = bpf_xdp_adjust_tail,
3888         .gpl_only       = false,
3889         .ret_type       = RET_INTEGER,
3890         .arg1_type      = ARG_PTR_TO_CTX,
3891         .arg2_type      = ARG_ANYTHING,
3892 };
3893
3894 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3895 {
3896         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3897         void *meta = xdp->data_meta + offset;
3898         unsigned long metalen = xdp->data - meta;
3899
3900         if (xdp_data_meta_unsupported(xdp))
3901                 return -ENOTSUPP;
3902         if (unlikely(meta < xdp_frame_end ||
3903                      meta > xdp->data))
3904                 return -EINVAL;
3905         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3906                      (metalen > 32)))
3907                 return -EACCES;
3908
3909         xdp->data_meta = meta;
3910
3911         return 0;
3912 }
3913
3914 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3915         .func           = bpf_xdp_adjust_meta,
3916         .gpl_only       = false,
3917         .ret_type       = RET_INTEGER,
3918         .arg1_type      = ARG_PTR_TO_CTX,
3919         .arg2_type      = ARG_ANYTHING,
3920 };
3921
3922 void xdp_do_flush(void)
3923 {
3924         __dev_flush();
3925         __cpu_map_flush();
3926         __xsk_map_flush();
3927 }
3928 EXPORT_SYMBOL_GPL(xdp_do_flush);
3929
3930 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3931                     struct bpf_prog *xdp_prog)
3932 {
3933         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3934         enum bpf_map_type map_type = ri->map_type;
3935         void *fwd = ri->tgt_value;
3936         u32 map_id = ri->map_id;
3937         int err;
3938
3939         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3940         ri->map_type = BPF_MAP_TYPE_UNSPEC;
3941
3942         switch (map_type) {
3943         case BPF_MAP_TYPE_DEVMAP:
3944                 fallthrough;
3945         case BPF_MAP_TYPE_DEVMAP_HASH:
3946                 err = dev_map_enqueue(fwd, xdp, dev);
3947                 break;
3948         case BPF_MAP_TYPE_CPUMAP:
3949                 err = cpu_map_enqueue(fwd, xdp, dev);
3950                 break;
3951         case BPF_MAP_TYPE_XSKMAP:
3952                 err = __xsk_map_redirect(fwd, xdp);
3953                 break;
3954         case BPF_MAP_TYPE_UNSPEC:
3955                 if (map_id == INT_MAX) {
3956                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3957                         if (unlikely(!fwd)) {
3958                                 err = -EINVAL;
3959                                 break;
3960                         }
3961                         err = dev_xdp_enqueue(fwd, xdp, dev);
3962                         break;
3963                 }
3964                 fallthrough;
3965         default:
3966                 err = -EBADRQC;
3967         }
3968
3969         if (unlikely(err))
3970                 goto err;
3971
3972         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
3973         return 0;
3974 err:
3975         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
3976         return err;
3977 }
3978 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3979
3980 static int xdp_do_generic_redirect_map(struct net_device *dev,
3981                                        struct sk_buff *skb,
3982                                        struct xdp_buff *xdp,
3983                                        struct bpf_prog *xdp_prog,
3984                                        void *fwd,
3985                                        enum bpf_map_type map_type, u32 map_id)
3986 {
3987         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3988         int err;
3989
3990         switch (map_type) {
3991         case BPF_MAP_TYPE_DEVMAP:
3992                 fallthrough;
3993         case BPF_MAP_TYPE_DEVMAP_HASH:
3994                 err = dev_map_generic_redirect(fwd, skb, xdp_prog);
3995                 if (unlikely(err))
3996                         goto err;
3997                 break;
3998         case BPF_MAP_TYPE_XSKMAP:
3999                 err = xsk_generic_rcv(fwd, xdp);
4000                 if (err)
4001                         goto err;
4002                 consume_skb(skb);
4003                 break;
4004         default:
4005                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4006                 err = -EBADRQC;
4007                 goto err;
4008         }
4009
4010         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4011         return 0;
4012 err:
4013         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4014         return err;
4015 }
4016
4017 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4018                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4019 {
4020         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4021         enum bpf_map_type map_type = ri->map_type;
4022         void *fwd = ri->tgt_value;
4023         u32 map_id = ri->map_id;
4024         int err;
4025
4026         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4027         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4028
4029         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4030                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4031                 if (unlikely(!fwd)) {
4032                         err = -EINVAL;
4033                         goto err;
4034                 }
4035
4036                 err = xdp_ok_fwd_dev(fwd, skb->len);
4037                 if (unlikely(err))
4038                         goto err;
4039
4040                 skb->dev = fwd;
4041                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4042                 generic_xdp_tx(skb, xdp_prog);
4043                 return 0;
4044         }
4045
4046         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4047 err:
4048         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4049         return err;
4050 }
4051
4052 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4053 {
4054         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4055
4056         if (unlikely(flags))
4057                 return XDP_ABORTED;
4058
4059         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4060          * by map_idr) is used for ifindex based XDP redirect.
4061          */
4062         ri->tgt_index = ifindex;
4063         ri->map_id = INT_MAX;
4064         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4065
4066         return XDP_REDIRECT;
4067 }
4068
4069 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4070         .func           = bpf_xdp_redirect,
4071         .gpl_only       = false,
4072         .ret_type       = RET_INTEGER,
4073         .arg1_type      = ARG_ANYTHING,
4074         .arg2_type      = ARG_ANYTHING,
4075 };
4076
4077 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4078            u64, flags)
4079 {
4080         return map->ops->map_redirect(map, ifindex, flags);
4081 }
4082
4083 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4084         .func           = bpf_xdp_redirect_map,
4085         .gpl_only       = false,
4086         .ret_type       = RET_INTEGER,
4087         .arg1_type      = ARG_CONST_MAP_PTR,
4088         .arg2_type      = ARG_ANYTHING,
4089         .arg3_type      = ARG_ANYTHING,
4090 };
4091
4092 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4093                                   unsigned long off, unsigned long len)
4094 {
4095         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4096
4097         if (unlikely(!ptr))
4098                 return len;
4099         if (ptr != dst_buff)
4100                 memcpy(dst_buff, ptr, len);
4101
4102         return 0;
4103 }
4104
4105 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4106            u64, flags, void *, meta, u64, meta_size)
4107 {
4108         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4109
4110         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4111                 return -EINVAL;
4112         if (unlikely(!skb || skb_size > skb->len))
4113                 return -EFAULT;
4114
4115         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4116                                 bpf_skb_copy);
4117 }
4118
4119 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4120         .func           = bpf_skb_event_output,
4121         .gpl_only       = true,
4122         .ret_type       = RET_INTEGER,
4123         .arg1_type      = ARG_PTR_TO_CTX,
4124         .arg2_type      = ARG_CONST_MAP_PTR,
4125         .arg3_type      = ARG_ANYTHING,
4126         .arg4_type      = ARG_PTR_TO_MEM,
4127         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4128 };
4129
4130 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4131
4132 const struct bpf_func_proto bpf_skb_output_proto = {
4133         .func           = bpf_skb_event_output,
4134         .gpl_only       = true,
4135         .ret_type       = RET_INTEGER,
4136         .arg1_type      = ARG_PTR_TO_BTF_ID,
4137         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4138         .arg2_type      = ARG_CONST_MAP_PTR,
4139         .arg3_type      = ARG_ANYTHING,
4140         .arg4_type      = ARG_PTR_TO_MEM,
4141         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4142 };
4143
4144 static unsigned short bpf_tunnel_key_af(u64 flags)
4145 {
4146         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4147 }
4148
4149 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4150            u32, size, u64, flags)
4151 {
4152         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4153         u8 compat[sizeof(struct bpf_tunnel_key)];
4154         void *to_orig = to;
4155         int err;
4156
4157         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4158                 err = -EINVAL;
4159                 goto err_clear;
4160         }
4161         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4162                 err = -EPROTO;
4163                 goto err_clear;
4164         }
4165         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4166                 err = -EINVAL;
4167                 switch (size) {
4168                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4169                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4170                         goto set_compat;
4171                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4172                         /* Fixup deprecated structure layouts here, so we have
4173                          * a common path later on.
4174                          */
4175                         if (ip_tunnel_info_af(info) != AF_INET)
4176                                 goto err_clear;
4177 set_compat:
4178                         to = (struct bpf_tunnel_key *)compat;
4179                         break;
4180                 default:
4181                         goto err_clear;
4182                 }
4183         }
4184
4185         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4186         to->tunnel_tos = info->key.tos;
4187         to->tunnel_ttl = info->key.ttl;
4188         to->tunnel_ext = 0;
4189
4190         if (flags & BPF_F_TUNINFO_IPV6) {
4191                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4192                        sizeof(to->remote_ipv6));
4193                 to->tunnel_label = be32_to_cpu(info->key.label);
4194         } else {
4195                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4196                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4197                 to->tunnel_label = 0;
4198         }
4199
4200         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4201                 memcpy(to_orig, to, size);
4202
4203         return 0;
4204 err_clear:
4205         memset(to_orig, 0, size);
4206         return err;
4207 }
4208
4209 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4210         .func           = bpf_skb_get_tunnel_key,
4211         .gpl_only       = false,
4212         .ret_type       = RET_INTEGER,
4213         .arg1_type      = ARG_PTR_TO_CTX,
4214         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4215         .arg3_type      = ARG_CONST_SIZE,
4216         .arg4_type      = ARG_ANYTHING,
4217 };
4218
4219 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4220 {
4221         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4222         int err;
4223
4224         if (unlikely(!info ||
4225                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4226                 err = -ENOENT;
4227                 goto err_clear;
4228         }
4229         if (unlikely(size < info->options_len)) {
4230                 err = -ENOMEM;
4231                 goto err_clear;
4232         }
4233
4234         ip_tunnel_info_opts_get(to, info);
4235         if (size > info->options_len)
4236                 memset(to + info->options_len, 0, size - info->options_len);
4237
4238         return info->options_len;
4239 err_clear:
4240         memset(to, 0, size);
4241         return err;
4242 }
4243
4244 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4245         .func           = bpf_skb_get_tunnel_opt,
4246         .gpl_only       = false,
4247         .ret_type       = RET_INTEGER,
4248         .arg1_type      = ARG_PTR_TO_CTX,
4249         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4250         .arg3_type      = ARG_CONST_SIZE,
4251 };
4252
4253 static struct metadata_dst __percpu *md_dst;
4254
4255 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4256            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4257 {
4258         struct metadata_dst *md = this_cpu_ptr(md_dst);
4259         u8 compat[sizeof(struct bpf_tunnel_key)];
4260         struct ip_tunnel_info *info;
4261
4262         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4263                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4264                 return -EINVAL;
4265         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4266                 switch (size) {
4267                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4268                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4269                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4270                         /* Fixup deprecated structure layouts here, so we have
4271                          * a common path later on.
4272                          */
4273                         memcpy(compat, from, size);
4274                         memset(compat + size, 0, sizeof(compat) - size);
4275                         from = (const struct bpf_tunnel_key *) compat;
4276                         break;
4277                 default:
4278                         return -EINVAL;
4279                 }
4280         }
4281         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4282                      from->tunnel_ext))
4283                 return -EINVAL;
4284
4285         skb_dst_drop(skb);
4286         dst_hold((struct dst_entry *) md);
4287         skb_dst_set(skb, (struct dst_entry *) md);
4288
4289         info = &md->u.tun_info;
4290         memset(info, 0, sizeof(*info));
4291         info->mode = IP_TUNNEL_INFO_TX;
4292
4293         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4294         if (flags & BPF_F_DONT_FRAGMENT)
4295                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4296         if (flags & BPF_F_ZERO_CSUM_TX)
4297                 info->key.tun_flags &= ~TUNNEL_CSUM;
4298         if (flags & BPF_F_SEQ_NUMBER)
4299                 info->key.tun_flags |= TUNNEL_SEQ;
4300
4301         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4302         info->key.tos = from->tunnel_tos;
4303         info->key.ttl = from->tunnel_ttl;
4304
4305         if (flags & BPF_F_TUNINFO_IPV6) {
4306                 info->mode |= IP_TUNNEL_INFO_IPV6;
4307                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4308                        sizeof(from->remote_ipv6));
4309                 info->key.label = cpu_to_be32(from->tunnel_label) &
4310                                   IPV6_FLOWLABEL_MASK;
4311         } else {
4312                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4313         }
4314
4315         return 0;
4316 }
4317
4318 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4319         .func           = bpf_skb_set_tunnel_key,
4320         .gpl_only       = false,
4321         .ret_type       = RET_INTEGER,
4322         .arg1_type      = ARG_PTR_TO_CTX,
4323         .arg2_type      = ARG_PTR_TO_MEM,
4324         .arg3_type      = ARG_CONST_SIZE,
4325         .arg4_type      = ARG_ANYTHING,
4326 };
4327
4328 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4329            const u8 *, from, u32, size)
4330 {
4331         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4332         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4333
4334         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4335                 return -EINVAL;
4336         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4337                 return -ENOMEM;
4338
4339         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4340
4341         return 0;
4342 }
4343
4344 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4345         .func           = bpf_skb_set_tunnel_opt,
4346         .gpl_only       = false,
4347         .ret_type       = RET_INTEGER,
4348         .arg1_type      = ARG_PTR_TO_CTX,
4349         .arg2_type      = ARG_PTR_TO_MEM,
4350         .arg3_type      = ARG_CONST_SIZE,
4351 };
4352
4353 static const struct bpf_func_proto *
4354 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4355 {
4356         if (!md_dst) {
4357                 struct metadata_dst __percpu *tmp;
4358
4359                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4360                                                 METADATA_IP_TUNNEL,
4361                                                 GFP_KERNEL);
4362                 if (!tmp)
4363                         return NULL;
4364                 if (cmpxchg(&md_dst, NULL, tmp))
4365                         metadata_dst_free_percpu(tmp);
4366         }
4367
4368         switch (which) {
4369         case BPF_FUNC_skb_set_tunnel_key:
4370                 return &bpf_skb_set_tunnel_key_proto;
4371         case BPF_FUNC_skb_set_tunnel_opt:
4372                 return &bpf_skb_set_tunnel_opt_proto;
4373         default:
4374                 return NULL;
4375         }
4376 }
4377
4378 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4379            u32, idx)
4380 {
4381         struct bpf_array *array = container_of(map, struct bpf_array, map);
4382         struct cgroup *cgrp;
4383         struct sock *sk;
4384
4385         sk = skb_to_full_sk(skb);
4386         if (!sk || !sk_fullsock(sk))
4387                 return -ENOENT;
4388         if (unlikely(idx >= array->map.max_entries))
4389                 return -E2BIG;
4390
4391         cgrp = READ_ONCE(array->ptrs[idx]);
4392         if (unlikely(!cgrp))
4393                 return -EAGAIN;
4394
4395         return sk_under_cgroup_hierarchy(sk, cgrp);
4396 }
4397
4398 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4399         .func           = bpf_skb_under_cgroup,
4400         .gpl_only       = false,
4401         .ret_type       = RET_INTEGER,
4402         .arg1_type      = ARG_PTR_TO_CTX,
4403         .arg2_type      = ARG_CONST_MAP_PTR,
4404         .arg3_type      = ARG_ANYTHING,
4405 };
4406
4407 #ifdef CONFIG_SOCK_CGROUP_DATA
4408 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4409 {
4410         struct cgroup *cgrp;
4411
4412         sk = sk_to_full_sk(sk);
4413         if (!sk || !sk_fullsock(sk))
4414                 return 0;
4415
4416         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4417         return cgroup_id(cgrp);
4418 }
4419
4420 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4421 {
4422         return __bpf_sk_cgroup_id(skb->sk);
4423 }
4424
4425 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4426         .func           = bpf_skb_cgroup_id,
4427         .gpl_only       = false,
4428         .ret_type       = RET_INTEGER,
4429         .arg1_type      = ARG_PTR_TO_CTX,
4430 };
4431
4432 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4433                                               int ancestor_level)
4434 {
4435         struct cgroup *ancestor;
4436         struct cgroup *cgrp;
4437
4438         sk = sk_to_full_sk(sk);
4439         if (!sk || !sk_fullsock(sk))
4440                 return 0;
4441
4442         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4443         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4444         if (!ancestor)
4445                 return 0;
4446
4447         return cgroup_id(ancestor);
4448 }
4449
4450 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4451            ancestor_level)
4452 {
4453         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4454 }
4455
4456 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4457         .func           = bpf_skb_ancestor_cgroup_id,
4458         .gpl_only       = false,
4459         .ret_type       = RET_INTEGER,
4460         .arg1_type      = ARG_PTR_TO_CTX,
4461         .arg2_type      = ARG_ANYTHING,
4462 };
4463
4464 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4465 {
4466         return __bpf_sk_cgroup_id(sk);
4467 }
4468
4469 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4470         .func           = bpf_sk_cgroup_id,
4471         .gpl_only       = false,
4472         .ret_type       = RET_INTEGER,
4473         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4474 };
4475
4476 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4477 {
4478         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4479 }
4480
4481 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4482         .func           = bpf_sk_ancestor_cgroup_id,
4483         .gpl_only       = false,
4484         .ret_type       = RET_INTEGER,
4485         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4486         .arg2_type      = ARG_ANYTHING,
4487 };
4488 #endif
4489
4490 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4491                                   unsigned long off, unsigned long len)
4492 {
4493         memcpy(dst_buff, src_buff + off, len);
4494         return 0;
4495 }
4496
4497 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4498            u64, flags, void *, meta, u64, meta_size)
4499 {
4500         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4501
4502         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4503                 return -EINVAL;
4504         if (unlikely(!xdp ||
4505                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4506                 return -EFAULT;
4507
4508         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4509                                 xdp_size, bpf_xdp_copy);
4510 }
4511
4512 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4513         .func           = bpf_xdp_event_output,
4514         .gpl_only       = true,
4515         .ret_type       = RET_INTEGER,
4516         .arg1_type      = ARG_PTR_TO_CTX,
4517         .arg2_type      = ARG_CONST_MAP_PTR,
4518         .arg3_type      = ARG_ANYTHING,
4519         .arg4_type      = ARG_PTR_TO_MEM,
4520         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4521 };
4522
4523 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4524
4525 const struct bpf_func_proto bpf_xdp_output_proto = {
4526         .func           = bpf_xdp_event_output,
4527         .gpl_only       = true,
4528         .ret_type       = RET_INTEGER,
4529         .arg1_type      = ARG_PTR_TO_BTF_ID,
4530         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4531         .arg2_type      = ARG_CONST_MAP_PTR,
4532         .arg3_type      = ARG_ANYTHING,
4533         .arg4_type      = ARG_PTR_TO_MEM,
4534         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4535 };
4536
4537 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4538 {
4539         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4540 }
4541
4542 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4543         .func           = bpf_get_socket_cookie,
4544         .gpl_only       = false,
4545         .ret_type       = RET_INTEGER,
4546         .arg1_type      = ARG_PTR_TO_CTX,
4547 };
4548
4549 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4550 {
4551         return __sock_gen_cookie(ctx->sk);
4552 }
4553
4554 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4555         .func           = bpf_get_socket_cookie_sock_addr,
4556         .gpl_only       = false,
4557         .ret_type       = RET_INTEGER,
4558         .arg1_type      = ARG_PTR_TO_CTX,
4559 };
4560
4561 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4562 {
4563         return __sock_gen_cookie(ctx);
4564 }
4565
4566 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4567         .func           = bpf_get_socket_cookie_sock,
4568         .gpl_only       = false,
4569         .ret_type       = RET_INTEGER,
4570         .arg1_type      = ARG_PTR_TO_CTX,
4571 };
4572
4573 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4574 {
4575         return sk ? sock_gen_cookie(sk) : 0;
4576 }
4577
4578 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4579         .func           = bpf_get_socket_ptr_cookie,
4580         .gpl_only       = false,
4581         .ret_type       = RET_INTEGER,
4582         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4583 };
4584
4585 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4586 {
4587         return __sock_gen_cookie(ctx->sk);
4588 }
4589
4590 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4591         .func           = bpf_get_socket_cookie_sock_ops,
4592         .gpl_only       = false,
4593         .ret_type       = RET_INTEGER,
4594         .arg1_type      = ARG_PTR_TO_CTX,
4595 };
4596
4597 static u64 __bpf_get_netns_cookie(struct sock *sk)
4598 {
4599         const struct net *net = sk ? sock_net(sk) : &init_net;
4600
4601         return net->net_cookie;
4602 }
4603
4604 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4605 {
4606         return __bpf_get_netns_cookie(ctx);
4607 }
4608
4609 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4610         .func           = bpf_get_netns_cookie_sock,
4611         .gpl_only       = false,
4612         .ret_type       = RET_INTEGER,
4613         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4614 };
4615
4616 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4617 {
4618         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4619 }
4620
4621 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4622         .func           = bpf_get_netns_cookie_sock_addr,
4623         .gpl_only       = false,
4624         .ret_type       = RET_INTEGER,
4625         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4626 };
4627
4628 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4629 {
4630         struct sock *sk = sk_to_full_sk(skb->sk);
4631         kuid_t kuid;
4632
4633         if (!sk || !sk_fullsock(sk))
4634                 return overflowuid;
4635         kuid = sock_net_uid(sock_net(sk), sk);
4636         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4637 }
4638
4639 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4640         .func           = bpf_get_socket_uid,
4641         .gpl_only       = false,
4642         .ret_type       = RET_INTEGER,
4643         .arg1_type      = ARG_PTR_TO_CTX,
4644 };
4645
4646 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4647                            char *optval, int optlen)
4648 {
4649         char devname[IFNAMSIZ];
4650         int val, valbool;
4651         struct net *net;
4652         int ifindex;
4653         int ret = 0;
4654
4655         if (!sk_fullsock(sk))
4656                 return -EINVAL;
4657
4658         sock_owned_by_me(sk);
4659
4660         if (level == SOL_SOCKET) {
4661                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4662                         return -EINVAL;
4663                 val = *((int *)optval);
4664                 valbool = val ? 1 : 0;
4665
4666                 /* Only some socketops are supported */
4667                 switch (optname) {
4668                 case SO_RCVBUF:
4669                         val = min_t(u32, val, sysctl_rmem_max);
4670                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4671                         WRITE_ONCE(sk->sk_rcvbuf,
4672                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4673                         break;
4674                 case SO_SNDBUF:
4675                         val = min_t(u32, val, sysctl_wmem_max);
4676                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4677                         WRITE_ONCE(sk->sk_sndbuf,
4678                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4679                         break;
4680                 case SO_MAX_PACING_RATE: /* 32bit version */
4681                         if (val != ~0U)
4682                                 cmpxchg(&sk->sk_pacing_status,
4683                                         SK_PACING_NONE,
4684                                         SK_PACING_NEEDED);
4685                         sk->sk_max_pacing_rate = (val == ~0U) ?
4686                                                  ~0UL : (unsigned int)val;
4687                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4688                                                  sk->sk_max_pacing_rate);
4689                         break;
4690                 case SO_PRIORITY:
4691                         sk->sk_priority = val;
4692                         break;
4693                 case SO_RCVLOWAT:
4694                         if (val < 0)
4695                                 val = INT_MAX;
4696                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4697                         break;
4698                 case SO_MARK:
4699                         if (sk->sk_mark != val) {
4700                                 sk->sk_mark = val;
4701                                 sk_dst_reset(sk);
4702                         }
4703                         break;
4704                 case SO_BINDTODEVICE:
4705                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4706                         strncpy(devname, optval, optlen);
4707                         devname[optlen] = 0;
4708
4709                         ifindex = 0;
4710                         if (devname[0] != '\0') {
4711                                 struct net_device *dev;
4712
4713                                 ret = -ENODEV;
4714
4715                                 net = sock_net(sk);
4716                                 dev = dev_get_by_name(net, devname);
4717                                 if (!dev)
4718                                         break;
4719                                 ifindex = dev->ifindex;
4720                                 dev_put(dev);
4721                         }
4722                         fallthrough;
4723                 case SO_BINDTOIFINDEX:
4724                         if (optname == SO_BINDTOIFINDEX)
4725                                 ifindex = val;
4726                         ret = sock_bindtoindex(sk, ifindex, false);
4727                         break;
4728                 case SO_KEEPALIVE:
4729                         if (sk->sk_prot->keepalive)
4730                                 sk->sk_prot->keepalive(sk, valbool);
4731                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4732                         break;
4733                 case SO_REUSEPORT:
4734                         sk->sk_reuseport = valbool;
4735                         break;
4736                 default:
4737                         ret = -EINVAL;
4738                 }
4739 #ifdef CONFIG_INET
4740         } else if (level == SOL_IP) {
4741                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4742                         return -EINVAL;
4743
4744                 val = *((int *)optval);
4745                 /* Only some options are supported */
4746                 switch (optname) {
4747                 case IP_TOS:
4748                         if (val < -1 || val > 0xff) {
4749                                 ret = -EINVAL;
4750                         } else {
4751                                 struct inet_sock *inet = inet_sk(sk);
4752
4753                                 if (val == -1)
4754                                         val = 0;
4755                                 inet->tos = val;
4756                         }
4757                         break;
4758                 default:
4759                         ret = -EINVAL;
4760                 }
4761 #if IS_ENABLED(CONFIG_IPV6)
4762         } else if (level == SOL_IPV6) {
4763                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4764                         return -EINVAL;
4765
4766                 val = *((int *)optval);
4767                 /* Only some options are supported */
4768                 switch (optname) {
4769                 case IPV6_TCLASS:
4770                         if (val < -1 || val > 0xff) {
4771                                 ret = -EINVAL;
4772                         } else {
4773                                 struct ipv6_pinfo *np = inet6_sk(sk);
4774
4775                                 if (val == -1)
4776                                         val = 0;
4777                                 np->tclass = val;
4778                         }
4779                         break;
4780                 default:
4781                         ret = -EINVAL;
4782                 }
4783 #endif
4784         } else if (level == SOL_TCP &&
4785                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4786                 if (optname == TCP_CONGESTION) {
4787                         char name[TCP_CA_NAME_MAX];
4788
4789                         strncpy(name, optval, min_t(long, optlen,
4790                                                     TCP_CA_NAME_MAX-1));
4791                         name[TCP_CA_NAME_MAX-1] = 0;
4792                         ret = tcp_set_congestion_control(sk, name, false, true);
4793                 } else {
4794                         struct inet_connection_sock *icsk = inet_csk(sk);
4795                         struct tcp_sock *tp = tcp_sk(sk);
4796                         unsigned long timeout;
4797
4798                         if (optlen != sizeof(int))
4799                                 return -EINVAL;
4800
4801                         val = *((int *)optval);
4802                         /* Only some options are supported */
4803                         switch (optname) {
4804                         case TCP_BPF_IW:
4805                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4806                                         ret = -EINVAL;
4807                                 else
4808                                         tp->snd_cwnd = val;
4809                                 break;
4810                         case TCP_BPF_SNDCWND_CLAMP:
4811                                 if (val <= 0) {
4812                                         ret = -EINVAL;
4813                                 } else {
4814                                         tp->snd_cwnd_clamp = val;
4815                                         tp->snd_ssthresh = val;
4816                                 }
4817                                 break;
4818                         case TCP_BPF_DELACK_MAX:
4819                                 timeout = usecs_to_jiffies(val);
4820                                 if (timeout > TCP_DELACK_MAX ||
4821                                     timeout < TCP_TIMEOUT_MIN)
4822                                         return -EINVAL;
4823                                 inet_csk(sk)->icsk_delack_max = timeout;
4824                                 break;
4825                         case TCP_BPF_RTO_MIN:
4826                                 timeout = usecs_to_jiffies(val);
4827                                 if (timeout > TCP_RTO_MIN ||
4828                                     timeout < TCP_TIMEOUT_MIN)
4829                                         return -EINVAL;
4830                                 inet_csk(sk)->icsk_rto_min = timeout;
4831                                 break;
4832                         case TCP_SAVE_SYN:
4833                                 if (val < 0 || val > 1)
4834                                         ret = -EINVAL;
4835                                 else
4836                                         tp->save_syn = val;
4837                                 break;
4838                         case TCP_KEEPIDLE:
4839                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4840                                 break;
4841                         case TCP_KEEPINTVL:
4842                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4843                                         ret = -EINVAL;
4844                                 else
4845                                         tp->keepalive_intvl = val * HZ;
4846                                 break;
4847                         case TCP_KEEPCNT:
4848                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4849                                         ret = -EINVAL;
4850                                 else
4851                                         tp->keepalive_probes = val;
4852                                 break;
4853                         case TCP_SYNCNT:
4854                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4855                                         ret = -EINVAL;
4856                                 else
4857                                         icsk->icsk_syn_retries = val;
4858                                 break;
4859                         case TCP_USER_TIMEOUT:
4860                                 if (val < 0)
4861                                         ret = -EINVAL;
4862                                 else
4863                                         icsk->icsk_user_timeout = val;
4864                                 break;
4865                         case TCP_NOTSENT_LOWAT:
4866                                 tp->notsent_lowat = val;
4867                                 sk->sk_write_space(sk);
4868                                 break;
4869                         case TCP_WINDOW_CLAMP:
4870                                 ret = tcp_set_window_clamp(sk, val);
4871                                 break;
4872                         default:
4873                                 ret = -EINVAL;
4874                         }
4875                 }
4876 #endif
4877         } else {
4878                 ret = -EINVAL;
4879         }
4880         return ret;
4881 }
4882
4883 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4884                            char *optval, int optlen)
4885 {
4886         if (!sk_fullsock(sk))
4887                 goto err_clear;
4888
4889         sock_owned_by_me(sk);
4890
4891         if (level == SOL_SOCKET) {
4892                 if (optlen != sizeof(int))
4893                         goto err_clear;
4894
4895                 switch (optname) {
4896                 case SO_MARK:
4897                         *((int *)optval) = sk->sk_mark;
4898                         break;
4899                 case SO_PRIORITY:
4900                         *((int *)optval) = sk->sk_priority;
4901                         break;
4902                 case SO_BINDTOIFINDEX:
4903                         *((int *)optval) = sk->sk_bound_dev_if;
4904                         break;
4905                 case SO_REUSEPORT:
4906                         *((int *)optval) = sk->sk_reuseport;
4907                         break;
4908                 default:
4909                         goto err_clear;
4910                 }
4911 #ifdef CONFIG_INET
4912         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4913                 struct inet_connection_sock *icsk;
4914                 struct tcp_sock *tp;
4915
4916                 switch (optname) {
4917                 case TCP_CONGESTION:
4918                         icsk = inet_csk(sk);
4919
4920                         if (!icsk->icsk_ca_ops || optlen <= 1)
4921                                 goto err_clear;
4922                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4923                         optval[optlen - 1] = 0;
4924                         break;
4925                 case TCP_SAVED_SYN:
4926                         tp = tcp_sk(sk);
4927
4928                         if (optlen <= 0 || !tp->saved_syn ||
4929                             optlen > tcp_saved_syn_len(tp->saved_syn))
4930                                 goto err_clear;
4931                         memcpy(optval, tp->saved_syn->data, optlen);
4932                         break;
4933                 default:
4934                         goto err_clear;
4935                 }
4936         } else if (level == SOL_IP) {
4937                 struct inet_sock *inet = inet_sk(sk);
4938
4939                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4940                         goto err_clear;
4941
4942                 /* Only some options are supported */
4943                 switch (optname) {
4944                 case IP_TOS:
4945                         *((int *)optval) = (int)inet->tos;
4946                         break;
4947                 default:
4948                         goto err_clear;
4949                 }
4950 #if IS_ENABLED(CONFIG_IPV6)
4951         } else if (level == SOL_IPV6) {
4952                 struct ipv6_pinfo *np = inet6_sk(sk);
4953
4954                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4955                         goto err_clear;
4956
4957                 /* Only some options are supported */
4958                 switch (optname) {
4959                 case IPV6_TCLASS:
4960                         *((int *)optval) = (int)np->tclass;
4961                         break;
4962                 default:
4963                         goto err_clear;
4964                 }
4965 #endif
4966 #endif
4967         } else {
4968                 goto err_clear;
4969         }
4970         return 0;
4971 err_clear:
4972         memset(optval, 0, optlen);
4973         return -EINVAL;
4974 }
4975
4976 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
4977            int, level, int, optname, char *, optval, int, optlen)
4978 {
4979         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
4980 }
4981
4982 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
4983         .func           = bpf_sock_addr_setsockopt,
4984         .gpl_only       = false,
4985         .ret_type       = RET_INTEGER,
4986         .arg1_type      = ARG_PTR_TO_CTX,
4987         .arg2_type      = ARG_ANYTHING,
4988         .arg3_type      = ARG_ANYTHING,
4989         .arg4_type      = ARG_PTR_TO_MEM,
4990         .arg5_type      = ARG_CONST_SIZE,
4991 };
4992
4993 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
4994            int, level, int, optname, char *, optval, int, optlen)
4995 {
4996         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
4997 }
4998
4999 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5000         .func           = bpf_sock_addr_getsockopt,
5001         .gpl_only       = false,
5002         .ret_type       = RET_INTEGER,
5003         .arg1_type      = ARG_PTR_TO_CTX,
5004         .arg2_type      = ARG_ANYTHING,
5005         .arg3_type      = ARG_ANYTHING,
5006         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5007         .arg5_type      = ARG_CONST_SIZE,
5008 };
5009
5010 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5011            int, level, int, optname, char *, optval, int, optlen)
5012 {
5013         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5014 }
5015
5016 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5017         .func           = bpf_sock_ops_setsockopt,
5018         .gpl_only       = false,
5019         .ret_type       = RET_INTEGER,
5020         .arg1_type      = ARG_PTR_TO_CTX,
5021         .arg2_type      = ARG_ANYTHING,
5022         .arg3_type      = ARG_ANYTHING,
5023         .arg4_type      = ARG_PTR_TO_MEM,
5024         .arg5_type      = ARG_CONST_SIZE,
5025 };
5026
5027 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5028                                 int optname, const u8 **start)
5029 {
5030         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5031         const u8 *hdr_start;
5032         int ret;
5033
5034         if (syn_skb) {
5035                 /* sk is a request_sock here */
5036
5037                 if (optname == TCP_BPF_SYN) {
5038                         hdr_start = syn_skb->data;
5039                         ret = tcp_hdrlen(syn_skb);
5040                 } else if (optname == TCP_BPF_SYN_IP) {
5041                         hdr_start = skb_network_header(syn_skb);
5042                         ret = skb_network_header_len(syn_skb) +
5043                                 tcp_hdrlen(syn_skb);
5044                 } else {
5045                         /* optname == TCP_BPF_SYN_MAC */
5046                         hdr_start = skb_mac_header(syn_skb);
5047                         ret = skb_mac_header_len(syn_skb) +
5048                                 skb_network_header_len(syn_skb) +
5049                                 tcp_hdrlen(syn_skb);
5050                 }
5051         } else {
5052                 struct sock *sk = bpf_sock->sk;
5053                 struct saved_syn *saved_syn;
5054
5055                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5056                         /* synack retransmit. bpf_sock->syn_skb will
5057                          * not be available.  It has to resort to
5058                          * saved_syn (if it is saved).
5059                          */
5060                         saved_syn = inet_reqsk(sk)->saved_syn;
5061                 else
5062                         saved_syn = tcp_sk(sk)->saved_syn;
5063
5064                 if (!saved_syn)
5065                         return -ENOENT;
5066
5067                 if (optname == TCP_BPF_SYN) {
5068                         hdr_start = saved_syn->data +
5069                                 saved_syn->mac_hdrlen +
5070                                 saved_syn->network_hdrlen;
5071                         ret = saved_syn->tcp_hdrlen;
5072                 } else if (optname == TCP_BPF_SYN_IP) {
5073                         hdr_start = saved_syn->data +
5074                                 saved_syn->mac_hdrlen;
5075                         ret = saved_syn->network_hdrlen +
5076                                 saved_syn->tcp_hdrlen;
5077                 } else {
5078                         /* optname == TCP_BPF_SYN_MAC */
5079
5080                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5081                         if (!saved_syn->mac_hdrlen)
5082                                 return -ENOENT;
5083
5084                         hdr_start = saved_syn->data;
5085                         ret = saved_syn->mac_hdrlen +
5086                                 saved_syn->network_hdrlen +
5087                                 saved_syn->tcp_hdrlen;
5088                 }
5089         }
5090
5091         *start = hdr_start;
5092         return ret;
5093 }
5094
5095 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5096            int, level, int, optname, char *, optval, int, optlen)
5097 {
5098         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5099             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5100                 int ret, copy_len = 0;
5101                 const u8 *start;
5102
5103                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5104                 if (ret > 0) {
5105                         copy_len = ret;
5106                         if (optlen < copy_len) {
5107                                 copy_len = optlen;
5108                                 ret = -ENOSPC;
5109                         }
5110
5111                         memcpy(optval, start, copy_len);
5112                 }
5113
5114                 /* Zero out unused buffer at the end */
5115                 memset(optval + copy_len, 0, optlen - copy_len);
5116
5117                 return ret;
5118         }
5119
5120         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5121 }
5122
5123 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5124         .func           = bpf_sock_ops_getsockopt,
5125         .gpl_only       = false,
5126         .ret_type       = RET_INTEGER,
5127         .arg1_type      = ARG_PTR_TO_CTX,
5128         .arg2_type      = ARG_ANYTHING,
5129         .arg3_type      = ARG_ANYTHING,
5130         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5131         .arg5_type      = ARG_CONST_SIZE,
5132 };
5133
5134 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5135            int, argval)
5136 {
5137         struct sock *sk = bpf_sock->sk;
5138         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5139
5140         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5141                 return -EINVAL;
5142
5143         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5144
5145         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5146 }
5147
5148 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5149         .func           = bpf_sock_ops_cb_flags_set,
5150         .gpl_only       = false,
5151         .ret_type       = RET_INTEGER,
5152         .arg1_type      = ARG_PTR_TO_CTX,
5153         .arg2_type      = ARG_ANYTHING,
5154 };
5155
5156 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5157 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5158
5159 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5160            int, addr_len)
5161 {
5162 #ifdef CONFIG_INET
5163         struct sock *sk = ctx->sk;
5164         u32 flags = BIND_FROM_BPF;
5165         int err;
5166
5167         err = -EINVAL;
5168         if (addr_len < offsetofend(struct sockaddr, sa_family))
5169                 return err;
5170         if (addr->sa_family == AF_INET) {
5171                 if (addr_len < sizeof(struct sockaddr_in))
5172                         return err;
5173                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5174                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5175                 return __inet_bind(sk, addr, addr_len, flags);
5176 #if IS_ENABLED(CONFIG_IPV6)
5177         } else if (addr->sa_family == AF_INET6) {
5178                 if (addr_len < SIN6_LEN_RFC2133)
5179                         return err;
5180                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5181                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5182                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5183                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5184                  */
5185                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5186 #endif /* CONFIG_IPV6 */
5187         }
5188 #endif /* CONFIG_INET */
5189
5190         return -EAFNOSUPPORT;
5191 }
5192
5193 static const struct bpf_func_proto bpf_bind_proto = {
5194         .func           = bpf_bind,
5195         .gpl_only       = false,
5196         .ret_type       = RET_INTEGER,
5197         .arg1_type      = ARG_PTR_TO_CTX,
5198         .arg2_type      = ARG_PTR_TO_MEM,
5199         .arg3_type      = ARG_CONST_SIZE,
5200 };
5201
5202 #ifdef CONFIG_XFRM
5203 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5204            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5205 {
5206         const struct sec_path *sp = skb_sec_path(skb);
5207         const struct xfrm_state *x;
5208
5209         if (!sp || unlikely(index >= sp->len || flags))
5210                 goto err_clear;
5211
5212         x = sp->xvec[index];
5213
5214         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5215                 goto err_clear;
5216
5217         to->reqid = x->props.reqid;
5218         to->spi = x->id.spi;
5219         to->family = x->props.family;
5220         to->ext = 0;
5221
5222         if (to->family == AF_INET6) {
5223                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5224                        sizeof(to->remote_ipv6));
5225         } else {
5226                 to->remote_ipv4 = x->props.saddr.a4;
5227                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5228         }
5229
5230         return 0;
5231 err_clear:
5232         memset(to, 0, size);
5233         return -EINVAL;
5234 }
5235
5236 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5237         .func           = bpf_skb_get_xfrm_state,
5238         .gpl_only       = false,
5239         .ret_type       = RET_INTEGER,
5240         .arg1_type      = ARG_PTR_TO_CTX,
5241         .arg2_type      = ARG_ANYTHING,
5242         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5243         .arg4_type      = ARG_CONST_SIZE,
5244         .arg5_type      = ARG_ANYTHING,
5245 };
5246 #endif
5247
5248 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5249 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5250                                   const struct neighbour *neigh,
5251                                   const struct net_device *dev, u32 mtu)
5252 {
5253         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5254         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5255         params->h_vlan_TCI = 0;
5256         params->h_vlan_proto = 0;
5257         if (mtu)
5258                 params->mtu_result = mtu; /* union with tot_len */
5259
5260         return 0;
5261 }
5262 #endif
5263
5264 #if IS_ENABLED(CONFIG_INET)
5265 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5266                                u32 flags, bool check_mtu)
5267 {
5268         struct fib_nh_common *nhc;
5269         struct in_device *in_dev;
5270         struct neighbour *neigh;
5271         struct net_device *dev;
5272         struct fib_result res;
5273         struct flowi4 fl4;
5274         u32 mtu = 0;
5275         int err;
5276
5277         dev = dev_get_by_index_rcu(net, params->ifindex);
5278         if (unlikely(!dev))
5279                 return -ENODEV;
5280
5281         /* verify forwarding is enabled on this interface */
5282         in_dev = __in_dev_get_rcu(dev);
5283         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5284                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5285
5286         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5287                 fl4.flowi4_iif = 1;
5288                 fl4.flowi4_oif = params->ifindex;
5289         } else {
5290                 fl4.flowi4_iif = params->ifindex;
5291                 fl4.flowi4_oif = 0;
5292         }
5293         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5294         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5295         fl4.flowi4_flags = 0;
5296
5297         fl4.flowi4_proto = params->l4_protocol;
5298         fl4.daddr = params->ipv4_dst;
5299         fl4.saddr = params->ipv4_src;
5300         fl4.fl4_sport = params->sport;
5301         fl4.fl4_dport = params->dport;
5302         fl4.flowi4_multipath_hash = 0;
5303
5304         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5305                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5306                 struct fib_table *tb;
5307
5308                 tb = fib_get_table(net, tbid);
5309                 if (unlikely(!tb))
5310                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5311
5312                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5313         } else {
5314                 fl4.flowi4_mark = 0;
5315                 fl4.flowi4_secid = 0;
5316                 fl4.flowi4_tun_key.tun_id = 0;
5317                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5318
5319                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5320         }
5321
5322         if (err) {
5323                 /* map fib lookup errors to RTN_ type */
5324                 if (err == -EINVAL)
5325                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5326                 if (err == -EHOSTUNREACH)
5327                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5328                 if (err == -EACCES)
5329                         return BPF_FIB_LKUP_RET_PROHIBIT;
5330
5331                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5332         }
5333
5334         if (res.type != RTN_UNICAST)
5335                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5336
5337         if (fib_info_num_path(res.fi) > 1)
5338                 fib_select_path(net, &res, &fl4, NULL);
5339
5340         if (check_mtu) {
5341                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5342                 if (params->tot_len > mtu) {
5343                         params->mtu_result = mtu; /* union with tot_len */
5344                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5345                 }
5346         }
5347
5348         nhc = res.nhc;
5349
5350         /* do not handle lwt encaps right now */
5351         if (nhc->nhc_lwtstate)
5352                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5353
5354         dev = nhc->nhc_dev;
5355
5356         params->rt_metric = res.fi->fib_priority;
5357         params->ifindex = dev->ifindex;
5358
5359         /* xdp and cls_bpf programs are run in RCU-bh so
5360          * rcu_read_lock_bh is not needed here
5361          */
5362         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5363                 if (nhc->nhc_gw_family)
5364                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5365
5366                 neigh = __ipv4_neigh_lookup_noref(dev,
5367                                                  (__force u32)params->ipv4_dst);
5368         } else {
5369                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5370
5371                 params->family = AF_INET6;
5372                 *dst = nhc->nhc_gw.ipv6;
5373                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5374         }
5375
5376         if (!neigh)
5377                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5378
5379         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5380 }
5381 #endif
5382
5383 #if IS_ENABLED(CONFIG_IPV6)
5384 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5385                                u32 flags, bool check_mtu)
5386 {
5387         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5388         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5389         struct fib6_result res = {};
5390         struct neighbour *neigh;
5391         struct net_device *dev;
5392         struct inet6_dev *idev;
5393         struct flowi6 fl6;
5394         int strict = 0;
5395         int oif, err;
5396         u32 mtu = 0;
5397
5398         /* link local addresses are never forwarded */
5399         if (rt6_need_strict(dst) || rt6_need_strict(src))
5400                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5401
5402         dev = dev_get_by_index_rcu(net, params->ifindex);
5403         if (unlikely(!dev))
5404                 return -ENODEV;
5405
5406         idev = __in6_dev_get_safely(dev);
5407         if (unlikely(!idev || !idev->cnf.forwarding))
5408                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5409
5410         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5411                 fl6.flowi6_iif = 1;
5412                 oif = fl6.flowi6_oif = params->ifindex;
5413         } else {
5414                 oif = fl6.flowi6_iif = params->ifindex;
5415                 fl6.flowi6_oif = 0;
5416                 strict = RT6_LOOKUP_F_HAS_SADDR;
5417         }
5418         fl6.flowlabel = params->flowinfo;
5419         fl6.flowi6_scope = 0;
5420         fl6.flowi6_flags = 0;
5421         fl6.mp_hash = 0;
5422
5423         fl6.flowi6_proto = params->l4_protocol;
5424         fl6.daddr = *dst;
5425         fl6.saddr = *src;
5426         fl6.fl6_sport = params->sport;
5427         fl6.fl6_dport = params->dport;
5428
5429         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5430                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5431                 struct fib6_table *tb;
5432
5433                 tb = ipv6_stub->fib6_get_table(net, tbid);
5434                 if (unlikely(!tb))
5435                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5436
5437                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5438                                                    strict);
5439         } else {
5440                 fl6.flowi6_mark = 0;
5441                 fl6.flowi6_secid = 0;
5442                 fl6.flowi6_tun_key.tun_id = 0;
5443                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5444
5445                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5446         }
5447
5448         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5449                      res.f6i == net->ipv6.fib6_null_entry))
5450                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5451
5452         switch (res.fib6_type) {
5453         /* only unicast is forwarded */
5454         case RTN_UNICAST:
5455                 break;
5456         case RTN_BLACKHOLE:
5457                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5458         case RTN_UNREACHABLE:
5459                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5460         case RTN_PROHIBIT:
5461                 return BPF_FIB_LKUP_RET_PROHIBIT;
5462         default:
5463                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5464         }
5465
5466         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5467                                     fl6.flowi6_oif != 0, NULL, strict);
5468
5469         if (check_mtu) {
5470                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5471                 if (params->tot_len > mtu) {
5472                         params->mtu_result = mtu; /* union with tot_len */
5473                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5474                 }
5475         }
5476
5477         if (res.nh->fib_nh_lws)
5478                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5479
5480         if (res.nh->fib_nh_gw_family)
5481                 *dst = res.nh->fib_nh_gw6;
5482
5483         dev = res.nh->fib_nh_dev;
5484         params->rt_metric = res.f6i->fib6_metric;
5485         params->ifindex = dev->ifindex;
5486
5487         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5488          * not needed here.
5489          */
5490         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5491         if (!neigh)
5492                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5493
5494         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5495 }
5496 #endif
5497
5498 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5499            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5500 {
5501         if (plen < sizeof(*params))
5502                 return -EINVAL;
5503
5504         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5505                 return -EINVAL;
5506
5507         switch (params->family) {
5508 #if IS_ENABLED(CONFIG_INET)
5509         case AF_INET:
5510                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5511                                            flags, true);
5512 #endif
5513 #if IS_ENABLED(CONFIG_IPV6)
5514         case AF_INET6:
5515                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5516                                            flags, true);
5517 #endif
5518         }
5519         return -EAFNOSUPPORT;
5520 }
5521
5522 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5523         .func           = bpf_xdp_fib_lookup,
5524         .gpl_only       = true,
5525         .ret_type       = RET_INTEGER,
5526         .arg1_type      = ARG_PTR_TO_CTX,
5527         .arg2_type      = ARG_PTR_TO_MEM,
5528         .arg3_type      = ARG_CONST_SIZE,
5529         .arg4_type      = ARG_ANYTHING,
5530 };
5531
5532 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5533            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5534 {
5535         struct net *net = dev_net(skb->dev);
5536         int rc = -EAFNOSUPPORT;
5537         bool check_mtu = false;
5538
5539         if (plen < sizeof(*params))
5540                 return -EINVAL;
5541
5542         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5543                 return -EINVAL;
5544
5545         if (params->tot_len)
5546                 check_mtu = true;
5547
5548         switch (params->family) {
5549 #if IS_ENABLED(CONFIG_INET)
5550         case AF_INET:
5551                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5552                 break;
5553 #endif
5554 #if IS_ENABLED(CONFIG_IPV6)
5555         case AF_INET6:
5556                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5557                 break;
5558 #endif
5559         }
5560
5561         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5562                 struct net_device *dev;
5563
5564                 /* When tot_len isn't provided by user, check skb
5565                  * against MTU of FIB lookup resulting net_device
5566                  */
5567                 dev = dev_get_by_index_rcu(net, params->ifindex);
5568                 if (!is_skb_forwardable(dev, skb))
5569                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5570
5571                 params->mtu_result = dev->mtu; /* union with tot_len */
5572         }
5573
5574         return rc;
5575 }
5576
5577 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5578         .func           = bpf_skb_fib_lookup,
5579         .gpl_only       = true,
5580         .ret_type       = RET_INTEGER,
5581         .arg1_type      = ARG_PTR_TO_CTX,
5582         .arg2_type      = ARG_PTR_TO_MEM,
5583         .arg3_type      = ARG_CONST_SIZE,
5584         .arg4_type      = ARG_ANYTHING,
5585 };
5586
5587 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5588                                             u32 ifindex)
5589 {
5590         struct net *netns = dev_net(dev_curr);
5591
5592         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5593         if (ifindex == 0)
5594                 return dev_curr;
5595
5596         return dev_get_by_index_rcu(netns, ifindex);
5597 }
5598
5599 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5600            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5601 {
5602         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5603         struct net_device *dev = skb->dev;
5604         int skb_len, dev_len;
5605         int mtu;
5606
5607         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5608                 return -EINVAL;
5609
5610         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5611                 return -EINVAL;
5612
5613         dev = __dev_via_ifindex(dev, ifindex);
5614         if (unlikely(!dev))
5615                 return -ENODEV;
5616
5617         mtu = READ_ONCE(dev->mtu);
5618
5619         dev_len = mtu + dev->hard_header_len;
5620
5621         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5622         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5623
5624         skb_len += len_diff; /* minus result pass check */
5625         if (skb_len <= dev_len) {
5626                 ret = BPF_MTU_CHK_RET_SUCCESS;
5627                 goto out;
5628         }
5629         /* At this point, skb->len exceed MTU, but as it include length of all
5630          * segments, it can still be below MTU.  The SKB can possibly get
5631          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5632          * must choose if segs are to be MTU checked.
5633          */
5634         if (skb_is_gso(skb)) {
5635                 ret = BPF_MTU_CHK_RET_SUCCESS;
5636
5637                 if (flags & BPF_MTU_CHK_SEGS &&
5638                     !skb_gso_validate_network_len(skb, mtu))
5639                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5640         }
5641 out:
5642         /* BPF verifier guarantees valid pointer */
5643         *mtu_len = mtu;
5644
5645         return ret;
5646 }
5647
5648 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5649            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5650 {
5651         struct net_device *dev = xdp->rxq->dev;
5652         int xdp_len = xdp->data_end - xdp->data;
5653         int ret = BPF_MTU_CHK_RET_SUCCESS;
5654         int mtu, dev_len;
5655
5656         /* XDP variant doesn't support multi-buffer segment check (yet) */
5657         if (unlikely(flags))
5658                 return -EINVAL;
5659
5660         dev = __dev_via_ifindex(dev, ifindex);
5661         if (unlikely(!dev))
5662                 return -ENODEV;
5663
5664         mtu = READ_ONCE(dev->mtu);
5665
5666         /* Add L2-header as dev MTU is L3 size */
5667         dev_len = mtu + dev->hard_header_len;
5668
5669         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5670         if (*mtu_len)
5671                 xdp_len = *mtu_len + dev->hard_header_len;
5672
5673         xdp_len += len_diff; /* minus result pass check */
5674         if (xdp_len > dev_len)
5675                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5676
5677         /* BPF verifier guarantees valid pointer */
5678         *mtu_len = mtu;
5679
5680         return ret;
5681 }
5682
5683 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5684         .func           = bpf_skb_check_mtu,
5685         .gpl_only       = true,
5686         .ret_type       = RET_INTEGER,
5687         .arg1_type      = ARG_PTR_TO_CTX,
5688         .arg2_type      = ARG_ANYTHING,
5689         .arg3_type      = ARG_PTR_TO_INT,
5690         .arg4_type      = ARG_ANYTHING,
5691         .arg5_type      = ARG_ANYTHING,
5692 };
5693
5694 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5695         .func           = bpf_xdp_check_mtu,
5696         .gpl_only       = true,
5697         .ret_type       = RET_INTEGER,
5698         .arg1_type      = ARG_PTR_TO_CTX,
5699         .arg2_type      = ARG_ANYTHING,
5700         .arg3_type      = ARG_PTR_TO_INT,
5701         .arg4_type      = ARG_ANYTHING,
5702         .arg5_type      = ARG_ANYTHING,
5703 };
5704
5705 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5706 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5707 {
5708         int err;
5709         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5710
5711         if (!seg6_validate_srh(srh, len, false))
5712                 return -EINVAL;
5713
5714         switch (type) {
5715         case BPF_LWT_ENCAP_SEG6_INLINE:
5716                 if (skb->protocol != htons(ETH_P_IPV6))
5717                         return -EBADMSG;
5718
5719                 err = seg6_do_srh_inline(skb, srh);
5720                 break;
5721         case BPF_LWT_ENCAP_SEG6:
5722                 skb_reset_inner_headers(skb);
5723                 skb->encapsulation = 1;
5724                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5725                 break;
5726         default:
5727                 return -EINVAL;
5728         }
5729
5730         bpf_compute_data_pointers(skb);
5731         if (err)
5732                 return err;
5733
5734         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5735         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5736
5737         return seg6_lookup_nexthop(skb, NULL, 0);
5738 }
5739 #endif /* CONFIG_IPV6_SEG6_BPF */
5740
5741 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5742 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5743                              bool ingress)
5744 {
5745         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5746 }
5747 #endif
5748
5749 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5750            u32, len)
5751 {
5752         switch (type) {
5753 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5754         case BPF_LWT_ENCAP_SEG6:
5755         case BPF_LWT_ENCAP_SEG6_INLINE:
5756                 return bpf_push_seg6_encap(skb, type, hdr, len);
5757 #endif
5758 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5759         case BPF_LWT_ENCAP_IP:
5760                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5761 #endif
5762         default:
5763                 return -EINVAL;
5764         }
5765 }
5766
5767 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5768            void *, hdr, u32, len)
5769 {
5770         switch (type) {
5771 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5772         case BPF_LWT_ENCAP_IP:
5773                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5774 #endif
5775         default:
5776                 return -EINVAL;
5777         }
5778 }
5779
5780 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5781         .func           = bpf_lwt_in_push_encap,
5782         .gpl_only       = false,
5783         .ret_type       = RET_INTEGER,
5784         .arg1_type      = ARG_PTR_TO_CTX,
5785         .arg2_type      = ARG_ANYTHING,
5786         .arg3_type      = ARG_PTR_TO_MEM,
5787         .arg4_type      = ARG_CONST_SIZE
5788 };
5789
5790 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5791         .func           = bpf_lwt_xmit_push_encap,
5792         .gpl_only       = false,
5793         .ret_type       = RET_INTEGER,
5794         .arg1_type      = ARG_PTR_TO_CTX,
5795         .arg2_type      = ARG_ANYTHING,
5796         .arg3_type      = ARG_PTR_TO_MEM,
5797         .arg4_type      = ARG_CONST_SIZE
5798 };
5799
5800 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5801 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5802            const void *, from, u32, len)
5803 {
5804         struct seg6_bpf_srh_state *srh_state =
5805                 this_cpu_ptr(&seg6_bpf_srh_states);
5806         struct ipv6_sr_hdr *srh = srh_state->srh;
5807         void *srh_tlvs, *srh_end, *ptr;
5808         int srhoff = 0;
5809
5810         if (srh == NULL)
5811                 return -EINVAL;
5812
5813         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5814         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5815
5816         ptr = skb->data + offset;
5817         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5818                 srh_state->valid = false;
5819         else if (ptr < (void *)&srh->flags ||
5820                  ptr + len > (void *)&srh->segments)
5821                 return -EFAULT;
5822
5823         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5824                 return -EFAULT;
5825         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5826                 return -EINVAL;
5827         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5828
5829         memcpy(skb->data + offset, from, len);
5830         return 0;
5831 }
5832
5833 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5834         .func           = bpf_lwt_seg6_store_bytes,
5835         .gpl_only       = false,
5836         .ret_type       = RET_INTEGER,
5837         .arg1_type      = ARG_PTR_TO_CTX,
5838         .arg2_type      = ARG_ANYTHING,
5839         .arg3_type      = ARG_PTR_TO_MEM,
5840         .arg4_type      = ARG_CONST_SIZE
5841 };
5842
5843 static void bpf_update_srh_state(struct sk_buff *skb)
5844 {
5845         struct seg6_bpf_srh_state *srh_state =
5846                 this_cpu_ptr(&seg6_bpf_srh_states);
5847         int srhoff = 0;
5848
5849         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5850                 srh_state->srh = NULL;
5851         } else {
5852                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5853                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5854                 srh_state->valid = true;
5855         }
5856 }
5857
5858 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5859            u32, action, void *, param, u32, param_len)
5860 {
5861         struct seg6_bpf_srh_state *srh_state =
5862                 this_cpu_ptr(&seg6_bpf_srh_states);
5863         int hdroff = 0;
5864         int err;
5865
5866         switch (action) {
5867         case SEG6_LOCAL_ACTION_END_X:
5868                 if (!seg6_bpf_has_valid_srh(skb))
5869                         return -EBADMSG;
5870                 if (param_len != sizeof(struct in6_addr))
5871                         return -EINVAL;
5872                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5873         case SEG6_LOCAL_ACTION_END_T:
5874                 if (!seg6_bpf_has_valid_srh(skb))
5875                         return -EBADMSG;
5876                 if (param_len != sizeof(int))
5877                         return -EINVAL;
5878                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5879         case SEG6_LOCAL_ACTION_END_DT6:
5880                 if (!seg6_bpf_has_valid_srh(skb))
5881                         return -EBADMSG;
5882                 if (param_len != sizeof(int))
5883                         return -EINVAL;
5884
5885                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5886                         return -EBADMSG;
5887                 if (!pskb_pull(skb, hdroff))
5888                         return -EBADMSG;
5889
5890                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5891                 skb_reset_network_header(skb);
5892                 skb_reset_transport_header(skb);
5893                 skb->encapsulation = 0;
5894
5895                 bpf_compute_data_pointers(skb);
5896                 bpf_update_srh_state(skb);
5897                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5898         case SEG6_LOCAL_ACTION_END_B6:
5899                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5900                         return -EBADMSG;
5901                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5902                                           param, param_len);
5903                 if (!err)
5904                         bpf_update_srh_state(skb);
5905
5906                 return err;
5907         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5908                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5909                         return -EBADMSG;
5910                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5911                                           param, param_len);
5912                 if (!err)
5913                         bpf_update_srh_state(skb);
5914
5915                 return err;
5916         default:
5917                 return -EINVAL;
5918         }
5919 }
5920
5921 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5922         .func           = bpf_lwt_seg6_action,
5923         .gpl_only       = false,
5924         .ret_type       = RET_INTEGER,
5925         .arg1_type      = ARG_PTR_TO_CTX,
5926         .arg2_type      = ARG_ANYTHING,
5927         .arg3_type      = ARG_PTR_TO_MEM,
5928         .arg4_type      = ARG_CONST_SIZE
5929 };
5930
5931 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5932            s32, len)
5933 {
5934         struct seg6_bpf_srh_state *srh_state =
5935                 this_cpu_ptr(&seg6_bpf_srh_states);
5936         struct ipv6_sr_hdr *srh = srh_state->srh;
5937         void *srh_end, *srh_tlvs, *ptr;
5938         struct ipv6hdr *hdr;
5939         int srhoff = 0;
5940         int ret;
5941
5942         if (unlikely(srh == NULL))
5943                 return -EINVAL;
5944
5945         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5946                         ((srh->first_segment + 1) << 4));
5947         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5948                         srh_state->hdrlen);
5949         ptr = skb->data + offset;
5950
5951         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5952                 return -EFAULT;
5953         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5954                 return -EFAULT;
5955
5956         if (len > 0) {
5957                 ret = skb_cow_head(skb, len);
5958                 if (unlikely(ret < 0))
5959                         return ret;
5960
5961                 ret = bpf_skb_net_hdr_push(skb, offset, len);
5962         } else {
5963                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
5964         }
5965
5966         bpf_compute_data_pointers(skb);
5967         if (unlikely(ret < 0))
5968                 return ret;
5969
5970         hdr = (struct ipv6hdr *)skb->data;
5971         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5972
5973         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5974                 return -EINVAL;
5975         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5976         srh_state->hdrlen += len;
5977         srh_state->valid = false;
5978         return 0;
5979 }
5980
5981 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
5982         .func           = bpf_lwt_seg6_adjust_srh,
5983         .gpl_only       = false,
5984         .ret_type       = RET_INTEGER,
5985         .arg1_type      = ARG_PTR_TO_CTX,
5986         .arg2_type      = ARG_ANYTHING,
5987         .arg3_type      = ARG_ANYTHING,
5988 };
5989 #endif /* CONFIG_IPV6_SEG6_BPF */
5990
5991 #ifdef CONFIG_INET
5992 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
5993                               int dif, int sdif, u8 family, u8 proto)
5994 {
5995         bool refcounted = false;
5996         struct sock *sk = NULL;
5997
5998         if (family == AF_INET) {
5999                 __be32 src4 = tuple->ipv4.saddr;
6000                 __be32 dst4 = tuple->ipv4.daddr;
6001
6002                 if (proto == IPPROTO_TCP)
6003                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6004                                            src4, tuple->ipv4.sport,
6005                                            dst4, tuple->ipv4.dport,
6006                                            dif, sdif, &refcounted);
6007                 else
6008                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6009                                                dst4, tuple->ipv4.dport,
6010                                                dif, sdif, &udp_table, NULL);
6011 #if IS_ENABLED(CONFIG_IPV6)
6012         } else {
6013                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6014                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6015
6016                 if (proto == IPPROTO_TCP)
6017                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6018                                             src6, tuple->ipv6.sport,
6019                                             dst6, ntohs(tuple->ipv6.dport),
6020                                             dif, sdif, &refcounted);
6021                 else if (likely(ipv6_bpf_stub))
6022                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6023                                                             src6, tuple->ipv6.sport,
6024                                                             dst6, tuple->ipv6.dport,
6025                                                             dif, sdif,
6026                                                             &udp_table, NULL);
6027 #endif
6028         }
6029
6030         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6031                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6032                 sk = NULL;
6033         }
6034         return sk;
6035 }
6036
6037 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6038  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6039  * Returns the socket as an 'unsigned long' to simplify the casting in the
6040  * callers to satisfy BPF_CALL declarations.
6041  */
6042 static struct sock *
6043 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6044                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6045                  u64 flags)
6046 {
6047         struct sock *sk = NULL;
6048         u8 family = AF_UNSPEC;
6049         struct net *net;
6050         int sdif;
6051
6052         if (len == sizeof(tuple->ipv4))
6053                 family = AF_INET;
6054         else if (len == sizeof(tuple->ipv6))
6055                 family = AF_INET6;
6056         else
6057                 return NULL;
6058
6059         if (unlikely(family == AF_UNSPEC || flags ||
6060                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6061                 goto out;
6062
6063         if (family == AF_INET)
6064                 sdif = inet_sdif(skb);
6065         else
6066                 sdif = inet6_sdif(skb);
6067
6068         if ((s32)netns_id < 0) {
6069                 net = caller_net;
6070                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6071         } else {
6072                 net = get_net_ns_by_id(caller_net, netns_id);
6073                 if (unlikely(!net))
6074                         goto out;
6075                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6076                 put_net(net);
6077         }
6078
6079 out:
6080         return sk;
6081 }
6082
6083 static struct sock *
6084 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6085                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6086                 u64 flags)
6087 {
6088         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6089                                            ifindex, proto, netns_id, flags);
6090
6091         if (sk) {
6092                 sk = sk_to_full_sk(sk);
6093                 if (!sk_fullsock(sk)) {
6094                         sock_gen_put(sk);
6095                         return NULL;
6096                 }
6097         }
6098
6099         return sk;
6100 }
6101
6102 static struct sock *
6103 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6104                u8 proto, u64 netns_id, u64 flags)
6105 {
6106         struct net *caller_net;
6107         int ifindex;
6108
6109         if (skb->dev) {
6110                 caller_net = dev_net(skb->dev);
6111                 ifindex = skb->dev->ifindex;
6112         } else {
6113                 caller_net = sock_net(skb->sk);
6114                 ifindex = 0;
6115         }
6116
6117         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6118                                 netns_id, flags);
6119 }
6120
6121 static struct sock *
6122 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6123               u8 proto, u64 netns_id, u64 flags)
6124 {
6125         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6126                                          flags);
6127
6128         if (sk) {
6129                 sk = sk_to_full_sk(sk);
6130                 if (!sk_fullsock(sk)) {
6131                         sock_gen_put(sk);
6132                         return NULL;
6133                 }
6134         }
6135
6136         return sk;
6137 }
6138
6139 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6140            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6141 {
6142         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6143                                              netns_id, flags);
6144 }
6145
6146 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6147         .func           = bpf_skc_lookup_tcp,
6148         .gpl_only       = false,
6149         .pkt_access     = true,
6150         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6151         .arg1_type      = ARG_PTR_TO_CTX,
6152         .arg2_type      = ARG_PTR_TO_MEM,
6153         .arg3_type      = ARG_CONST_SIZE,
6154         .arg4_type      = ARG_ANYTHING,
6155         .arg5_type      = ARG_ANYTHING,
6156 };
6157
6158 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6159            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6160 {
6161         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6162                                             netns_id, flags);
6163 }
6164
6165 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6166         .func           = bpf_sk_lookup_tcp,
6167         .gpl_only       = false,
6168         .pkt_access     = true,
6169         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6170         .arg1_type      = ARG_PTR_TO_CTX,
6171         .arg2_type      = ARG_PTR_TO_MEM,
6172         .arg3_type      = ARG_CONST_SIZE,
6173         .arg4_type      = ARG_ANYTHING,
6174         .arg5_type      = ARG_ANYTHING,
6175 };
6176
6177 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6178            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6179 {
6180         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6181                                             netns_id, flags);
6182 }
6183
6184 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6185         .func           = bpf_sk_lookup_udp,
6186         .gpl_only       = false,
6187         .pkt_access     = true,
6188         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6189         .arg1_type      = ARG_PTR_TO_CTX,
6190         .arg2_type      = ARG_PTR_TO_MEM,
6191         .arg3_type      = ARG_CONST_SIZE,
6192         .arg4_type      = ARG_ANYTHING,
6193         .arg5_type      = ARG_ANYTHING,
6194 };
6195
6196 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6197 {
6198         if (sk && sk_is_refcounted(sk))
6199                 sock_gen_put(sk);
6200         return 0;
6201 }
6202
6203 static const struct bpf_func_proto bpf_sk_release_proto = {
6204         .func           = bpf_sk_release,
6205         .gpl_only       = false,
6206         .ret_type       = RET_INTEGER,
6207         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6208 };
6209
6210 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6211            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6212 {
6213         struct net *caller_net = dev_net(ctx->rxq->dev);
6214         int ifindex = ctx->rxq->dev->ifindex;
6215
6216         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6217                                               ifindex, IPPROTO_UDP, netns_id,
6218                                               flags);
6219 }
6220
6221 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6222         .func           = bpf_xdp_sk_lookup_udp,
6223         .gpl_only       = false,
6224         .pkt_access     = true,
6225         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6226         .arg1_type      = ARG_PTR_TO_CTX,
6227         .arg2_type      = ARG_PTR_TO_MEM,
6228         .arg3_type      = ARG_CONST_SIZE,
6229         .arg4_type      = ARG_ANYTHING,
6230         .arg5_type      = ARG_ANYTHING,
6231 };
6232
6233 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6234            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6235 {
6236         struct net *caller_net = dev_net(ctx->rxq->dev);
6237         int ifindex = ctx->rxq->dev->ifindex;
6238
6239         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6240                                                ifindex, IPPROTO_TCP, netns_id,
6241                                                flags);
6242 }
6243
6244 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6245         .func           = bpf_xdp_skc_lookup_tcp,
6246         .gpl_only       = false,
6247         .pkt_access     = true,
6248         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6249         .arg1_type      = ARG_PTR_TO_CTX,
6250         .arg2_type      = ARG_PTR_TO_MEM,
6251         .arg3_type      = ARG_CONST_SIZE,
6252         .arg4_type      = ARG_ANYTHING,
6253         .arg5_type      = ARG_ANYTHING,
6254 };
6255
6256 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6257            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6258 {
6259         struct net *caller_net = dev_net(ctx->rxq->dev);
6260         int ifindex = ctx->rxq->dev->ifindex;
6261
6262         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6263                                               ifindex, IPPROTO_TCP, netns_id,
6264                                               flags);
6265 }
6266
6267 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6268         .func           = bpf_xdp_sk_lookup_tcp,
6269         .gpl_only       = false,
6270         .pkt_access     = true,
6271         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6272         .arg1_type      = ARG_PTR_TO_CTX,
6273         .arg2_type      = ARG_PTR_TO_MEM,
6274         .arg3_type      = ARG_CONST_SIZE,
6275         .arg4_type      = ARG_ANYTHING,
6276         .arg5_type      = ARG_ANYTHING,
6277 };
6278
6279 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6280            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6281 {
6282         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6283                                                sock_net(ctx->sk), 0,
6284                                                IPPROTO_TCP, netns_id, flags);
6285 }
6286
6287 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6288         .func           = bpf_sock_addr_skc_lookup_tcp,
6289         .gpl_only       = false,
6290         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6291         .arg1_type      = ARG_PTR_TO_CTX,
6292         .arg2_type      = ARG_PTR_TO_MEM,
6293         .arg3_type      = ARG_CONST_SIZE,
6294         .arg4_type      = ARG_ANYTHING,
6295         .arg5_type      = ARG_ANYTHING,
6296 };
6297
6298 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6299            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6300 {
6301         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6302                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6303                                               netns_id, flags);
6304 }
6305
6306 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6307         .func           = bpf_sock_addr_sk_lookup_tcp,
6308         .gpl_only       = false,
6309         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6310         .arg1_type      = ARG_PTR_TO_CTX,
6311         .arg2_type      = ARG_PTR_TO_MEM,
6312         .arg3_type      = ARG_CONST_SIZE,
6313         .arg4_type      = ARG_ANYTHING,
6314         .arg5_type      = ARG_ANYTHING,
6315 };
6316
6317 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6318            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6319 {
6320         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6321                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6322                                               netns_id, flags);
6323 }
6324
6325 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6326         .func           = bpf_sock_addr_sk_lookup_udp,
6327         .gpl_only       = false,
6328         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6329         .arg1_type      = ARG_PTR_TO_CTX,
6330         .arg2_type      = ARG_PTR_TO_MEM,
6331         .arg3_type      = ARG_CONST_SIZE,
6332         .arg4_type      = ARG_ANYTHING,
6333         .arg5_type      = ARG_ANYTHING,
6334 };
6335
6336 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6337                                   struct bpf_insn_access_aux *info)
6338 {
6339         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6340                                           icsk_retransmits))
6341                 return false;
6342
6343         if (off % size != 0)
6344                 return false;
6345
6346         switch (off) {
6347         case offsetof(struct bpf_tcp_sock, bytes_received):
6348         case offsetof(struct bpf_tcp_sock, bytes_acked):
6349                 return size == sizeof(__u64);
6350         default:
6351                 return size == sizeof(__u32);
6352         }
6353 }
6354
6355 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6356                                     const struct bpf_insn *si,
6357                                     struct bpf_insn *insn_buf,
6358                                     struct bpf_prog *prog, u32 *target_size)
6359 {
6360         struct bpf_insn *insn = insn_buf;
6361
6362 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6363         do {                                                            \
6364                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6365                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6366                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6367                                       si->dst_reg, si->src_reg,         \
6368                                       offsetof(struct tcp_sock, FIELD)); \
6369         } while (0)
6370
6371 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6372         do {                                                            \
6373                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6374                                           FIELD) >                      \
6375                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6376                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6377                                         struct inet_connection_sock,    \
6378                                         FIELD),                         \
6379                                       si->dst_reg, si->src_reg,         \
6380                                       offsetof(                         \
6381                                         struct inet_connection_sock,    \
6382                                         FIELD));                        \
6383         } while (0)
6384
6385         if (insn > insn_buf)
6386                 return insn - insn_buf;
6387
6388         switch (si->off) {
6389         case offsetof(struct bpf_tcp_sock, rtt_min):
6390                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6391                              sizeof(struct minmax));
6392                 BUILD_BUG_ON(sizeof(struct minmax) <
6393                              sizeof(struct minmax_sample));
6394
6395                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6396                                       offsetof(struct tcp_sock, rtt_min) +
6397                                       offsetof(struct minmax_sample, v));
6398                 break;
6399         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6400                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6401                 break;
6402         case offsetof(struct bpf_tcp_sock, srtt_us):
6403                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6404                 break;
6405         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6406                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6407                 break;
6408         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6409                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6410                 break;
6411         case offsetof(struct bpf_tcp_sock, snd_nxt):
6412                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6413                 break;
6414         case offsetof(struct bpf_tcp_sock, snd_una):
6415                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6416                 break;
6417         case offsetof(struct bpf_tcp_sock, mss_cache):
6418                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6419                 break;
6420         case offsetof(struct bpf_tcp_sock, ecn_flags):
6421                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6422                 break;
6423         case offsetof(struct bpf_tcp_sock, rate_delivered):
6424                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6425                 break;
6426         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6427                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6428                 break;
6429         case offsetof(struct bpf_tcp_sock, packets_out):
6430                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6431                 break;
6432         case offsetof(struct bpf_tcp_sock, retrans_out):
6433                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6434                 break;
6435         case offsetof(struct bpf_tcp_sock, total_retrans):
6436                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6437                 break;
6438         case offsetof(struct bpf_tcp_sock, segs_in):
6439                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6440                 break;
6441         case offsetof(struct bpf_tcp_sock, data_segs_in):
6442                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6443                 break;
6444         case offsetof(struct bpf_tcp_sock, segs_out):
6445                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6446                 break;
6447         case offsetof(struct bpf_tcp_sock, data_segs_out):
6448                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6449                 break;
6450         case offsetof(struct bpf_tcp_sock, lost_out):
6451                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6452                 break;
6453         case offsetof(struct bpf_tcp_sock, sacked_out):
6454                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6455                 break;
6456         case offsetof(struct bpf_tcp_sock, bytes_received):
6457                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6458                 break;
6459         case offsetof(struct bpf_tcp_sock, bytes_acked):
6460                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6461                 break;
6462         case offsetof(struct bpf_tcp_sock, dsack_dups):
6463                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6464                 break;
6465         case offsetof(struct bpf_tcp_sock, delivered):
6466                 BPF_TCP_SOCK_GET_COMMON(delivered);
6467                 break;
6468         case offsetof(struct bpf_tcp_sock, delivered_ce):
6469                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6470                 break;
6471         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6472                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6473                 break;
6474         }
6475
6476         return insn - insn_buf;
6477 }
6478
6479 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6480 {
6481         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6482                 return (unsigned long)sk;
6483
6484         return (unsigned long)NULL;
6485 }
6486
6487 const struct bpf_func_proto bpf_tcp_sock_proto = {
6488         .func           = bpf_tcp_sock,
6489         .gpl_only       = false,
6490         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6491         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6492 };
6493
6494 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6495 {
6496         sk = sk_to_full_sk(sk);
6497
6498         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6499                 return (unsigned long)sk;
6500
6501         return (unsigned long)NULL;
6502 }
6503
6504 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6505         .func           = bpf_get_listener_sock,
6506         .gpl_only       = false,
6507         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6508         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6509 };
6510
6511 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6512 {
6513         unsigned int iphdr_len;
6514
6515         switch (skb_protocol(skb, true)) {
6516         case cpu_to_be16(ETH_P_IP):
6517                 iphdr_len = sizeof(struct iphdr);
6518                 break;
6519         case cpu_to_be16(ETH_P_IPV6):
6520                 iphdr_len = sizeof(struct ipv6hdr);
6521                 break;
6522         default:
6523                 return 0;
6524         }
6525
6526         if (skb_headlen(skb) < iphdr_len)
6527                 return 0;
6528
6529         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6530                 return 0;
6531
6532         return INET_ECN_set_ce(skb);
6533 }
6534
6535 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6536                                   struct bpf_insn_access_aux *info)
6537 {
6538         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6539                 return false;
6540
6541         if (off % size != 0)
6542                 return false;
6543
6544         switch (off) {
6545         default:
6546                 return size == sizeof(__u32);
6547         }
6548 }
6549
6550 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6551                                     const struct bpf_insn *si,
6552                                     struct bpf_insn *insn_buf,
6553                                     struct bpf_prog *prog, u32 *target_size)
6554 {
6555         struct bpf_insn *insn = insn_buf;
6556
6557 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6558         do {                                                            \
6559                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6560                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6561                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6562                                       si->dst_reg, si->src_reg,         \
6563                                       offsetof(struct xdp_sock, FIELD)); \
6564         } while (0)
6565
6566         switch (si->off) {
6567         case offsetof(struct bpf_xdp_sock, queue_id):
6568                 BPF_XDP_SOCK_GET(queue_id);
6569                 break;
6570         }
6571
6572         return insn - insn_buf;
6573 }
6574
6575 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6576         .func           = bpf_skb_ecn_set_ce,
6577         .gpl_only       = false,
6578         .ret_type       = RET_INTEGER,
6579         .arg1_type      = ARG_PTR_TO_CTX,
6580 };
6581
6582 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6583            struct tcphdr *, th, u32, th_len)
6584 {
6585 #ifdef CONFIG_SYN_COOKIES
6586         u32 cookie;
6587         int ret;
6588
6589         if (unlikely(!sk || th_len < sizeof(*th)))
6590                 return -EINVAL;
6591
6592         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6593         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6594                 return -EINVAL;
6595
6596         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6597                 return -EINVAL;
6598
6599         if (!th->ack || th->rst || th->syn)
6600                 return -ENOENT;
6601
6602         if (tcp_synq_no_recent_overflow(sk))
6603                 return -ENOENT;
6604
6605         cookie = ntohl(th->ack_seq) - 1;
6606
6607         switch (sk->sk_family) {
6608         case AF_INET:
6609                 if (unlikely(iph_len < sizeof(struct iphdr)))
6610                         return -EINVAL;
6611
6612                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6613                 break;
6614
6615 #if IS_BUILTIN(CONFIG_IPV6)
6616         case AF_INET6:
6617                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6618                         return -EINVAL;
6619
6620                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6621                 break;
6622 #endif /* CONFIG_IPV6 */
6623
6624         default:
6625                 return -EPROTONOSUPPORT;
6626         }
6627
6628         if (ret > 0)
6629                 return 0;
6630
6631         return -ENOENT;
6632 #else
6633         return -ENOTSUPP;
6634 #endif
6635 }
6636
6637 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6638         .func           = bpf_tcp_check_syncookie,
6639         .gpl_only       = true,
6640         .pkt_access     = true,
6641         .ret_type       = RET_INTEGER,
6642         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6643         .arg2_type      = ARG_PTR_TO_MEM,
6644         .arg3_type      = ARG_CONST_SIZE,
6645         .arg4_type      = ARG_PTR_TO_MEM,
6646         .arg5_type      = ARG_CONST_SIZE,
6647 };
6648
6649 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6650            struct tcphdr *, th, u32, th_len)
6651 {
6652 #ifdef CONFIG_SYN_COOKIES
6653         u32 cookie;
6654         u16 mss;
6655
6656         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6657                 return -EINVAL;
6658
6659         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6660                 return -EINVAL;
6661
6662         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6663                 return -ENOENT;
6664
6665         if (!th->syn || th->ack || th->fin || th->rst)
6666                 return -EINVAL;
6667
6668         if (unlikely(iph_len < sizeof(struct iphdr)))
6669                 return -EINVAL;
6670
6671         /* Both struct iphdr and struct ipv6hdr have the version field at the
6672          * same offset so we can cast to the shorter header (struct iphdr).
6673          */
6674         switch (((struct iphdr *)iph)->version) {
6675         case 4:
6676                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6677                         return -EINVAL;
6678
6679                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6680                 break;
6681
6682 #if IS_BUILTIN(CONFIG_IPV6)
6683         case 6:
6684                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6685                         return -EINVAL;
6686
6687                 if (sk->sk_family != AF_INET6)
6688                         return -EINVAL;
6689
6690                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6691                 break;
6692 #endif /* CONFIG_IPV6 */
6693
6694         default:
6695                 return -EPROTONOSUPPORT;
6696         }
6697         if (mss == 0)
6698                 return -ENOENT;
6699
6700         return cookie | ((u64)mss << 32);
6701 #else
6702         return -EOPNOTSUPP;
6703 #endif /* CONFIG_SYN_COOKIES */
6704 }
6705
6706 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6707         .func           = bpf_tcp_gen_syncookie,
6708         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6709         .pkt_access     = true,
6710         .ret_type       = RET_INTEGER,
6711         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6712         .arg2_type      = ARG_PTR_TO_MEM,
6713         .arg3_type      = ARG_CONST_SIZE,
6714         .arg4_type      = ARG_PTR_TO_MEM,
6715         .arg5_type      = ARG_CONST_SIZE,
6716 };
6717
6718 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6719 {
6720         if (!sk || flags != 0)
6721                 return -EINVAL;
6722         if (!skb_at_tc_ingress(skb))
6723                 return -EOPNOTSUPP;
6724         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6725                 return -ENETUNREACH;
6726         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6727                 return -ESOCKTNOSUPPORT;
6728         if (sk_is_refcounted(sk) &&
6729             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6730                 return -ENOENT;
6731
6732         skb_orphan(skb);
6733         skb->sk = sk;
6734         skb->destructor = sock_pfree;
6735
6736         return 0;
6737 }
6738
6739 static const struct bpf_func_proto bpf_sk_assign_proto = {
6740         .func           = bpf_sk_assign,
6741         .gpl_only       = false,
6742         .ret_type       = RET_INTEGER,
6743         .arg1_type      = ARG_PTR_TO_CTX,
6744         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6745         .arg3_type      = ARG_ANYTHING,
6746 };
6747
6748 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6749                                     u8 search_kind, const u8 *magic,
6750                                     u8 magic_len, bool *eol)
6751 {
6752         u8 kind, kind_len;
6753
6754         *eol = false;
6755
6756         while (op < opend) {
6757                 kind = op[0];
6758
6759                 if (kind == TCPOPT_EOL) {
6760                         *eol = true;
6761                         return ERR_PTR(-ENOMSG);
6762                 } else if (kind == TCPOPT_NOP) {
6763                         op++;
6764                         continue;
6765                 }
6766
6767                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6768                         /* Something is wrong in the received header.
6769                          * Follow the TCP stack's tcp_parse_options()
6770                          * and just bail here.
6771                          */
6772                         return ERR_PTR(-EFAULT);
6773
6774                 kind_len = op[1];
6775                 if (search_kind == kind) {
6776                         if (!magic_len)
6777                                 return op;
6778
6779                         if (magic_len > kind_len - 2)
6780                                 return ERR_PTR(-ENOMSG);
6781
6782                         if (!memcmp(&op[2], magic, magic_len))
6783                                 return op;
6784                 }
6785
6786                 op += kind_len;
6787         }
6788
6789         return ERR_PTR(-ENOMSG);
6790 }
6791
6792 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6793            void *, search_res, u32, len, u64, flags)
6794 {
6795         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6796         const u8 *op, *opend, *magic, *search = search_res;
6797         u8 search_kind, search_len, copy_len, magic_len;
6798         int ret;
6799
6800         /* 2 byte is the minimal option len except TCPOPT_NOP and
6801          * TCPOPT_EOL which are useless for the bpf prog to learn
6802          * and this helper disallow loading them also.
6803          */
6804         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6805                 return -EINVAL;
6806
6807         search_kind = search[0];
6808         search_len = search[1];
6809
6810         if (search_len > len || search_kind == TCPOPT_NOP ||
6811             search_kind == TCPOPT_EOL)
6812                 return -EINVAL;
6813
6814         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6815                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6816                 if (search_len != 4 && search_len != 6)
6817                         return -EINVAL;
6818                 magic = &search[2];
6819                 magic_len = search_len - 2;
6820         } else {
6821                 if (search_len)
6822                         return -EINVAL;
6823                 magic = NULL;
6824                 magic_len = 0;
6825         }
6826
6827         if (load_syn) {
6828                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6829                 if (ret < 0)
6830                         return ret;
6831
6832                 opend = op + ret;
6833                 op += sizeof(struct tcphdr);
6834         } else {
6835                 if (!bpf_sock->skb ||
6836                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6837                         /* This bpf_sock->op cannot call this helper */
6838                         return -EPERM;
6839
6840                 opend = bpf_sock->skb_data_end;
6841                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6842         }
6843
6844         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6845                                 &eol);
6846         if (IS_ERR(op))
6847                 return PTR_ERR(op);
6848
6849         copy_len = op[1];
6850         ret = copy_len;
6851         if (copy_len > len) {
6852                 ret = -ENOSPC;
6853                 copy_len = len;
6854         }
6855
6856         memcpy(search_res, op, copy_len);
6857         return ret;
6858 }
6859
6860 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6861         .func           = bpf_sock_ops_load_hdr_opt,
6862         .gpl_only       = false,
6863         .ret_type       = RET_INTEGER,
6864         .arg1_type      = ARG_PTR_TO_CTX,
6865         .arg2_type      = ARG_PTR_TO_MEM,
6866         .arg3_type      = ARG_CONST_SIZE,
6867         .arg4_type      = ARG_ANYTHING,
6868 };
6869
6870 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6871            const void *, from, u32, len, u64, flags)
6872 {
6873         u8 new_kind, new_kind_len, magic_len = 0, *opend;
6874         const u8 *op, *new_op, *magic = NULL;
6875         struct sk_buff *skb;
6876         bool eol;
6877
6878         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6879                 return -EPERM;
6880
6881         if (len < 2 || flags)
6882                 return -EINVAL;
6883
6884         new_op = from;
6885         new_kind = new_op[0];
6886         new_kind_len = new_op[1];
6887
6888         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6889             new_kind == TCPOPT_EOL)
6890                 return -EINVAL;
6891
6892         if (new_kind_len > bpf_sock->remaining_opt_len)
6893                 return -ENOSPC;
6894
6895         /* 253 is another experimental kind */
6896         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
6897                 if (new_kind_len < 4)
6898                         return -EINVAL;
6899                 /* Match for the 2 byte magic also.
6900                  * RFC 6994: the magic could be 2 or 4 bytes.
6901                  * Hence, matching by 2 byte only is on the
6902                  * conservative side but it is the right
6903                  * thing to do for the 'search-for-duplication'
6904                  * purpose.
6905                  */
6906                 magic = &new_op[2];
6907                 magic_len = 2;
6908         }
6909
6910         /* Check for duplication */
6911         skb = bpf_sock->skb;
6912         op = skb->data + sizeof(struct tcphdr);
6913         opend = bpf_sock->skb_data_end;
6914
6915         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6916                                 &eol);
6917         if (!IS_ERR(op))
6918                 return -EEXIST;
6919
6920         if (PTR_ERR(op) != -ENOMSG)
6921                 return PTR_ERR(op);
6922
6923         if (eol)
6924                 /* The option has been ended.  Treat it as no more
6925                  * header option can be written.
6926                  */
6927                 return -ENOSPC;
6928
6929         /* No duplication found.  Store the header option. */
6930         memcpy(opend, from, new_kind_len);
6931
6932         bpf_sock->remaining_opt_len -= new_kind_len;
6933         bpf_sock->skb_data_end += new_kind_len;
6934
6935         return 0;
6936 }
6937
6938 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6939         .func           = bpf_sock_ops_store_hdr_opt,
6940         .gpl_only       = false,
6941         .ret_type       = RET_INTEGER,
6942         .arg1_type      = ARG_PTR_TO_CTX,
6943         .arg2_type      = ARG_PTR_TO_MEM,
6944         .arg3_type      = ARG_CONST_SIZE,
6945         .arg4_type      = ARG_ANYTHING,
6946 };
6947
6948 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6949            u32, len, u64, flags)
6950 {
6951         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6952                 return -EPERM;
6953
6954         if (flags || len < 2)
6955                 return -EINVAL;
6956
6957         if (len > bpf_sock->remaining_opt_len)
6958                 return -ENOSPC;
6959
6960         bpf_sock->remaining_opt_len -= len;
6961
6962         return 0;
6963 }
6964
6965 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
6966         .func           = bpf_sock_ops_reserve_hdr_opt,
6967         .gpl_only       = false,
6968         .ret_type       = RET_INTEGER,
6969         .arg1_type      = ARG_PTR_TO_CTX,
6970         .arg2_type      = ARG_ANYTHING,
6971         .arg3_type      = ARG_ANYTHING,
6972 };
6973
6974 #endif /* CONFIG_INET */
6975
6976 bool bpf_helper_changes_pkt_data(void *func)
6977 {
6978         if (func == bpf_skb_vlan_push ||
6979             func == bpf_skb_vlan_pop ||
6980             func == bpf_skb_store_bytes ||
6981             func == bpf_skb_change_proto ||
6982             func == bpf_skb_change_head ||
6983             func == sk_skb_change_head ||
6984             func == bpf_skb_change_tail ||
6985             func == sk_skb_change_tail ||
6986             func == bpf_skb_adjust_room ||
6987             func == sk_skb_adjust_room ||
6988             func == bpf_skb_pull_data ||
6989             func == sk_skb_pull_data ||
6990             func == bpf_clone_redirect ||
6991             func == bpf_l3_csum_replace ||
6992             func == bpf_l4_csum_replace ||
6993             func == bpf_xdp_adjust_head ||
6994             func == bpf_xdp_adjust_meta ||
6995             func == bpf_msg_pull_data ||
6996             func == bpf_msg_push_data ||
6997             func == bpf_msg_pop_data ||
6998             func == bpf_xdp_adjust_tail ||
6999 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7000             func == bpf_lwt_seg6_store_bytes ||
7001             func == bpf_lwt_seg6_adjust_srh ||
7002             func == bpf_lwt_seg6_action ||
7003 #endif
7004 #ifdef CONFIG_INET
7005             func == bpf_sock_ops_store_hdr_opt ||
7006 #endif
7007             func == bpf_lwt_in_push_encap ||
7008             func == bpf_lwt_xmit_push_encap)
7009                 return true;
7010
7011         return false;
7012 }
7013
7014 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7015 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7016
7017 static const struct bpf_func_proto *
7018 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7019 {
7020         switch (func_id) {
7021         /* inet and inet6 sockets are created in a process
7022          * context so there is always a valid uid/gid
7023          */
7024         case BPF_FUNC_get_current_uid_gid:
7025                 return &bpf_get_current_uid_gid_proto;
7026         case BPF_FUNC_get_local_storage:
7027                 return &bpf_get_local_storage_proto;
7028         case BPF_FUNC_get_socket_cookie:
7029                 return &bpf_get_socket_cookie_sock_proto;
7030         case BPF_FUNC_get_netns_cookie:
7031                 return &bpf_get_netns_cookie_sock_proto;
7032         case BPF_FUNC_perf_event_output:
7033                 return &bpf_event_output_data_proto;
7034         case BPF_FUNC_get_current_pid_tgid:
7035                 return &bpf_get_current_pid_tgid_proto;
7036         case BPF_FUNC_get_current_comm:
7037                 return &bpf_get_current_comm_proto;
7038 #ifdef CONFIG_CGROUPS
7039         case BPF_FUNC_get_current_cgroup_id:
7040                 return &bpf_get_current_cgroup_id_proto;
7041         case BPF_FUNC_get_current_ancestor_cgroup_id:
7042                 return &bpf_get_current_ancestor_cgroup_id_proto;
7043 #endif
7044 #ifdef CONFIG_CGROUP_NET_CLASSID
7045         case BPF_FUNC_get_cgroup_classid:
7046                 return &bpf_get_cgroup_classid_curr_proto;
7047 #endif
7048         case BPF_FUNC_sk_storage_get:
7049                 return &bpf_sk_storage_get_cg_sock_proto;
7050         default:
7051                 return bpf_base_func_proto(func_id);
7052         }
7053 }
7054
7055 static const struct bpf_func_proto *
7056 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7057 {
7058         switch (func_id) {
7059         /* inet and inet6 sockets are created in a process
7060          * context so there is always a valid uid/gid
7061          */
7062         case BPF_FUNC_get_current_uid_gid:
7063                 return &bpf_get_current_uid_gid_proto;
7064         case BPF_FUNC_bind:
7065                 switch (prog->expected_attach_type) {
7066                 case BPF_CGROUP_INET4_CONNECT:
7067                 case BPF_CGROUP_INET6_CONNECT:
7068                         return &bpf_bind_proto;
7069                 default:
7070                         return NULL;
7071                 }
7072         case BPF_FUNC_get_socket_cookie:
7073                 return &bpf_get_socket_cookie_sock_addr_proto;
7074         case BPF_FUNC_get_netns_cookie:
7075                 return &bpf_get_netns_cookie_sock_addr_proto;
7076         case BPF_FUNC_get_local_storage:
7077                 return &bpf_get_local_storage_proto;
7078         case BPF_FUNC_perf_event_output:
7079                 return &bpf_event_output_data_proto;
7080         case BPF_FUNC_get_current_pid_tgid:
7081                 return &bpf_get_current_pid_tgid_proto;
7082         case BPF_FUNC_get_current_comm:
7083                 return &bpf_get_current_comm_proto;
7084 #ifdef CONFIG_CGROUPS
7085         case BPF_FUNC_get_current_cgroup_id:
7086                 return &bpf_get_current_cgroup_id_proto;
7087         case BPF_FUNC_get_current_ancestor_cgroup_id:
7088                 return &bpf_get_current_ancestor_cgroup_id_proto;
7089 #endif
7090 #ifdef CONFIG_CGROUP_NET_CLASSID
7091         case BPF_FUNC_get_cgroup_classid:
7092                 return &bpf_get_cgroup_classid_curr_proto;
7093 #endif
7094 #ifdef CONFIG_INET
7095         case BPF_FUNC_sk_lookup_tcp:
7096                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7097         case BPF_FUNC_sk_lookup_udp:
7098                 return &bpf_sock_addr_sk_lookup_udp_proto;
7099         case BPF_FUNC_sk_release:
7100                 return &bpf_sk_release_proto;
7101         case BPF_FUNC_skc_lookup_tcp:
7102                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7103 #endif /* CONFIG_INET */
7104         case BPF_FUNC_sk_storage_get:
7105                 return &bpf_sk_storage_get_proto;
7106         case BPF_FUNC_sk_storage_delete:
7107                 return &bpf_sk_storage_delete_proto;
7108         case BPF_FUNC_setsockopt:
7109                 switch (prog->expected_attach_type) {
7110                 case BPF_CGROUP_INET4_BIND:
7111                 case BPF_CGROUP_INET6_BIND:
7112                 case BPF_CGROUP_INET4_CONNECT:
7113                 case BPF_CGROUP_INET6_CONNECT:
7114                 case BPF_CGROUP_UDP4_RECVMSG:
7115                 case BPF_CGROUP_UDP6_RECVMSG:
7116                 case BPF_CGROUP_UDP4_SENDMSG:
7117                 case BPF_CGROUP_UDP6_SENDMSG:
7118                 case BPF_CGROUP_INET4_GETPEERNAME:
7119                 case BPF_CGROUP_INET6_GETPEERNAME:
7120                 case BPF_CGROUP_INET4_GETSOCKNAME:
7121                 case BPF_CGROUP_INET6_GETSOCKNAME:
7122                         return &bpf_sock_addr_setsockopt_proto;
7123                 default:
7124                         return NULL;
7125                 }
7126         case BPF_FUNC_getsockopt:
7127                 switch (prog->expected_attach_type) {
7128                 case BPF_CGROUP_INET4_BIND:
7129                 case BPF_CGROUP_INET6_BIND:
7130                 case BPF_CGROUP_INET4_CONNECT:
7131                 case BPF_CGROUP_INET6_CONNECT:
7132                 case BPF_CGROUP_UDP4_RECVMSG:
7133                 case BPF_CGROUP_UDP6_RECVMSG:
7134                 case BPF_CGROUP_UDP4_SENDMSG:
7135                 case BPF_CGROUP_UDP6_SENDMSG:
7136                 case BPF_CGROUP_INET4_GETPEERNAME:
7137                 case BPF_CGROUP_INET6_GETPEERNAME:
7138                 case BPF_CGROUP_INET4_GETSOCKNAME:
7139                 case BPF_CGROUP_INET6_GETSOCKNAME:
7140                         return &bpf_sock_addr_getsockopt_proto;
7141                 default:
7142                         return NULL;
7143                 }
7144         default:
7145                 return bpf_sk_base_func_proto(func_id);
7146         }
7147 }
7148
7149 static const struct bpf_func_proto *
7150 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7151 {
7152         switch (func_id) {
7153         case BPF_FUNC_skb_load_bytes:
7154                 return &bpf_skb_load_bytes_proto;
7155         case BPF_FUNC_skb_load_bytes_relative:
7156                 return &bpf_skb_load_bytes_relative_proto;
7157         case BPF_FUNC_get_socket_cookie:
7158                 return &bpf_get_socket_cookie_proto;
7159         case BPF_FUNC_get_socket_uid:
7160                 return &bpf_get_socket_uid_proto;
7161         case BPF_FUNC_perf_event_output:
7162                 return &bpf_skb_event_output_proto;
7163         default:
7164                 return bpf_sk_base_func_proto(func_id);
7165         }
7166 }
7167
7168 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7169 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7170
7171 static const struct bpf_func_proto *
7172 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7173 {
7174         switch (func_id) {
7175         case BPF_FUNC_get_local_storage:
7176                 return &bpf_get_local_storage_proto;
7177         case BPF_FUNC_sk_fullsock:
7178                 return &bpf_sk_fullsock_proto;
7179         case BPF_FUNC_sk_storage_get:
7180                 return &bpf_sk_storage_get_proto;
7181         case BPF_FUNC_sk_storage_delete:
7182                 return &bpf_sk_storage_delete_proto;
7183         case BPF_FUNC_perf_event_output:
7184                 return &bpf_skb_event_output_proto;
7185 #ifdef CONFIG_SOCK_CGROUP_DATA
7186         case BPF_FUNC_skb_cgroup_id:
7187                 return &bpf_skb_cgroup_id_proto;
7188         case BPF_FUNC_skb_ancestor_cgroup_id:
7189                 return &bpf_skb_ancestor_cgroup_id_proto;
7190         case BPF_FUNC_sk_cgroup_id:
7191                 return &bpf_sk_cgroup_id_proto;
7192         case BPF_FUNC_sk_ancestor_cgroup_id:
7193                 return &bpf_sk_ancestor_cgroup_id_proto;
7194 #endif
7195 #ifdef CONFIG_INET
7196         case BPF_FUNC_sk_lookup_tcp:
7197                 return &bpf_sk_lookup_tcp_proto;
7198         case BPF_FUNC_sk_lookup_udp:
7199                 return &bpf_sk_lookup_udp_proto;
7200         case BPF_FUNC_sk_release:
7201                 return &bpf_sk_release_proto;
7202         case BPF_FUNC_skc_lookup_tcp:
7203                 return &bpf_skc_lookup_tcp_proto;
7204         case BPF_FUNC_tcp_sock:
7205                 return &bpf_tcp_sock_proto;
7206         case BPF_FUNC_get_listener_sock:
7207                 return &bpf_get_listener_sock_proto;
7208         case BPF_FUNC_skb_ecn_set_ce:
7209                 return &bpf_skb_ecn_set_ce_proto;
7210 #endif
7211         default:
7212                 return sk_filter_func_proto(func_id, prog);
7213         }
7214 }
7215
7216 static const struct bpf_func_proto *
7217 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7218 {
7219         switch (func_id) {
7220         case BPF_FUNC_skb_store_bytes:
7221                 return &bpf_skb_store_bytes_proto;
7222         case BPF_FUNC_skb_load_bytes:
7223                 return &bpf_skb_load_bytes_proto;
7224         case BPF_FUNC_skb_load_bytes_relative:
7225                 return &bpf_skb_load_bytes_relative_proto;
7226         case BPF_FUNC_skb_pull_data:
7227                 return &bpf_skb_pull_data_proto;
7228         case BPF_FUNC_csum_diff:
7229                 return &bpf_csum_diff_proto;
7230         case BPF_FUNC_csum_update:
7231                 return &bpf_csum_update_proto;
7232         case BPF_FUNC_csum_level:
7233                 return &bpf_csum_level_proto;
7234         case BPF_FUNC_l3_csum_replace:
7235                 return &bpf_l3_csum_replace_proto;
7236         case BPF_FUNC_l4_csum_replace:
7237                 return &bpf_l4_csum_replace_proto;
7238         case BPF_FUNC_clone_redirect:
7239                 return &bpf_clone_redirect_proto;
7240         case BPF_FUNC_get_cgroup_classid:
7241                 return &bpf_get_cgroup_classid_proto;
7242         case BPF_FUNC_skb_vlan_push:
7243                 return &bpf_skb_vlan_push_proto;
7244         case BPF_FUNC_skb_vlan_pop:
7245                 return &bpf_skb_vlan_pop_proto;
7246         case BPF_FUNC_skb_change_proto:
7247                 return &bpf_skb_change_proto_proto;
7248         case BPF_FUNC_skb_change_type:
7249                 return &bpf_skb_change_type_proto;
7250         case BPF_FUNC_skb_adjust_room:
7251                 return &bpf_skb_adjust_room_proto;
7252         case BPF_FUNC_skb_change_tail:
7253                 return &bpf_skb_change_tail_proto;
7254         case BPF_FUNC_skb_change_head:
7255                 return &bpf_skb_change_head_proto;
7256         case BPF_FUNC_skb_get_tunnel_key:
7257                 return &bpf_skb_get_tunnel_key_proto;
7258         case BPF_FUNC_skb_set_tunnel_key:
7259                 return bpf_get_skb_set_tunnel_proto(func_id);
7260         case BPF_FUNC_skb_get_tunnel_opt:
7261                 return &bpf_skb_get_tunnel_opt_proto;
7262         case BPF_FUNC_skb_set_tunnel_opt:
7263                 return bpf_get_skb_set_tunnel_proto(func_id);
7264         case BPF_FUNC_redirect:
7265                 return &bpf_redirect_proto;
7266         case BPF_FUNC_redirect_neigh:
7267                 return &bpf_redirect_neigh_proto;
7268         case BPF_FUNC_redirect_peer:
7269                 return &bpf_redirect_peer_proto;
7270         case BPF_FUNC_get_route_realm:
7271                 return &bpf_get_route_realm_proto;
7272         case BPF_FUNC_get_hash_recalc:
7273                 return &bpf_get_hash_recalc_proto;
7274         case BPF_FUNC_set_hash_invalid:
7275                 return &bpf_set_hash_invalid_proto;
7276         case BPF_FUNC_set_hash:
7277                 return &bpf_set_hash_proto;
7278         case BPF_FUNC_perf_event_output:
7279                 return &bpf_skb_event_output_proto;
7280         case BPF_FUNC_get_smp_processor_id:
7281                 return &bpf_get_smp_processor_id_proto;
7282         case BPF_FUNC_skb_under_cgroup:
7283                 return &bpf_skb_under_cgroup_proto;
7284         case BPF_FUNC_get_socket_cookie:
7285                 return &bpf_get_socket_cookie_proto;
7286         case BPF_FUNC_get_socket_uid:
7287                 return &bpf_get_socket_uid_proto;
7288         case BPF_FUNC_fib_lookup:
7289                 return &bpf_skb_fib_lookup_proto;
7290         case BPF_FUNC_check_mtu:
7291                 return &bpf_skb_check_mtu_proto;
7292         case BPF_FUNC_sk_fullsock:
7293                 return &bpf_sk_fullsock_proto;
7294         case BPF_FUNC_sk_storage_get:
7295                 return &bpf_sk_storage_get_proto;
7296         case BPF_FUNC_sk_storage_delete:
7297                 return &bpf_sk_storage_delete_proto;
7298 #ifdef CONFIG_XFRM
7299         case BPF_FUNC_skb_get_xfrm_state:
7300                 return &bpf_skb_get_xfrm_state_proto;
7301 #endif
7302 #ifdef CONFIG_CGROUP_NET_CLASSID
7303         case BPF_FUNC_skb_cgroup_classid:
7304                 return &bpf_skb_cgroup_classid_proto;
7305 #endif
7306 #ifdef CONFIG_SOCK_CGROUP_DATA
7307         case BPF_FUNC_skb_cgroup_id:
7308                 return &bpf_skb_cgroup_id_proto;
7309         case BPF_FUNC_skb_ancestor_cgroup_id:
7310                 return &bpf_skb_ancestor_cgroup_id_proto;
7311 #endif
7312 #ifdef CONFIG_INET
7313         case BPF_FUNC_sk_lookup_tcp:
7314                 return &bpf_sk_lookup_tcp_proto;
7315         case BPF_FUNC_sk_lookup_udp:
7316                 return &bpf_sk_lookup_udp_proto;
7317         case BPF_FUNC_sk_release:
7318                 return &bpf_sk_release_proto;
7319         case BPF_FUNC_tcp_sock:
7320                 return &bpf_tcp_sock_proto;
7321         case BPF_FUNC_get_listener_sock:
7322                 return &bpf_get_listener_sock_proto;
7323         case BPF_FUNC_skc_lookup_tcp:
7324                 return &bpf_skc_lookup_tcp_proto;
7325         case BPF_FUNC_tcp_check_syncookie:
7326                 return &bpf_tcp_check_syncookie_proto;
7327         case BPF_FUNC_skb_ecn_set_ce:
7328                 return &bpf_skb_ecn_set_ce_proto;
7329         case BPF_FUNC_tcp_gen_syncookie:
7330                 return &bpf_tcp_gen_syncookie_proto;
7331         case BPF_FUNC_sk_assign:
7332                 return &bpf_sk_assign_proto;
7333 #endif
7334         default:
7335                 return bpf_sk_base_func_proto(func_id);
7336         }
7337 }
7338
7339 static const struct bpf_func_proto *
7340 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7341 {
7342         switch (func_id) {
7343         case BPF_FUNC_perf_event_output:
7344                 return &bpf_xdp_event_output_proto;
7345         case BPF_FUNC_get_smp_processor_id:
7346                 return &bpf_get_smp_processor_id_proto;
7347         case BPF_FUNC_csum_diff:
7348                 return &bpf_csum_diff_proto;
7349         case BPF_FUNC_xdp_adjust_head:
7350                 return &bpf_xdp_adjust_head_proto;
7351         case BPF_FUNC_xdp_adjust_meta:
7352                 return &bpf_xdp_adjust_meta_proto;
7353         case BPF_FUNC_redirect:
7354                 return &bpf_xdp_redirect_proto;
7355         case BPF_FUNC_redirect_map:
7356                 return &bpf_xdp_redirect_map_proto;
7357         case BPF_FUNC_xdp_adjust_tail:
7358                 return &bpf_xdp_adjust_tail_proto;
7359         case BPF_FUNC_fib_lookup:
7360                 return &bpf_xdp_fib_lookup_proto;
7361         case BPF_FUNC_check_mtu:
7362                 return &bpf_xdp_check_mtu_proto;
7363 #ifdef CONFIG_INET
7364         case BPF_FUNC_sk_lookup_udp:
7365                 return &bpf_xdp_sk_lookup_udp_proto;
7366         case BPF_FUNC_sk_lookup_tcp:
7367                 return &bpf_xdp_sk_lookup_tcp_proto;
7368         case BPF_FUNC_sk_release:
7369                 return &bpf_sk_release_proto;
7370         case BPF_FUNC_skc_lookup_tcp:
7371                 return &bpf_xdp_skc_lookup_tcp_proto;
7372         case BPF_FUNC_tcp_check_syncookie:
7373                 return &bpf_tcp_check_syncookie_proto;
7374         case BPF_FUNC_tcp_gen_syncookie:
7375                 return &bpf_tcp_gen_syncookie_proto;
7376 #endif
7377         default:
7378                 return bpf_sk_base_func_proto(func_id);
7379         }
7380 }
7381
7382 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7383 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7384
7385 static const struct bpf_func_proto *
7386 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7387 {
7388         switch (func_id) {
7389         case BPF_FUNC_setsockopt:
7390                 return &bpf_sock_ops_setsockopt_proto;
7391         case BPF_FUNC_getsockopt:
7392                 return &bpf_sock_ops_getsockopt_proto;
7393         case BPF_FUNC_sock_ops_cb_flags_set:
7394                 return &bpf_sock_ops_cb_flags_set_proto;
7395         case BPF_FUNC_sock_map_update:
7396                 return &bpf_sock_map_update_proto;
7397         case BPF_FUNC_sock_hash_update:
7398                 return &bpf_sock_hash_update_proto;
7399         case BPF_FUNC_get_socket_cookie:
7400                 return &bpf_get_socket_cookie_sock_ops_proto;
7401         case BPF_FUNC_get_local_storage:
7402                 return &bpf_get_local_storage_proto;
7403         case BPF_FUNC_perf_event_output:
7404                 return &bpf_event_output_data_proto;
7405         case BPF_FUNC_sk_storage_get:
7406                 return &bpf_sk_storage_get_proto;
7407         case BPF_FUNC_sk_storage_delete:
7408                 return &bpf_sk_storage_delete_proto;
7409 #ifdef CONFIG_INET
7410         case BPF_FUNC_load_hdr_opt:
7411                 return &bpf_sock_ops_load_hdr_opt_proto;
7412         case BPF_FUNC_store_hdr_opt:
7413                 return &bpf_sock_ops_store_hdr_opt_proto;
7414         case BPF_FUNC_reserve_hdr_opt:
7415                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7416         case BPF_FUNC_tcp_sock:
7417                 return &bpf_tcp_sock_proto;
7418 #endif /* CONFIG_INET */
7419         default:
7420                 return bpf_sk_base_func_proto(func_id);
7421         }
7422 }
7423
7424 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7425 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7426
7427 static const struct bpf_func_proto *
7428 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7429 {
7430         switch (func_id) {
7431         case BPF_FUNC_msg_redirect_map:
7432                 return &bpf_msg_redirect_map_proto;
7433         case BPF_FUNC_msg_redirect_hash:
7434                 return &bpf_msg_redirect_hash_proto;
7435         case BPF_FUNC_msg_apply_bytes:
7436                 return &bpf_msg_apply_bytes_proto;
7437         case BPF_FUNC_msg_cork_bytes:
7438                 return &bpf_msg_cork_bytes_proto;
7439         case BPF_FUNC_msg_pull_data:
7440                 return &bpf_msg_pull_data_proto;
7441         case BPF_FUNC_msg_push_data:
7442                 return &bpf_msg_push_data_proto;
7443         case BPF_FUNC_msg_pop_data:
7444                 return &bpf_msg_pop_data_proto;
7445         case BPF_FUNC_perf_event_output:
7446                 return &bpf_event_output_data_proto;
7447         case BPF_FUNC_get_current_uid_gid:
7448                 return &bpf_get_current_uid_gid_proto;
7449         case BPF_FUNC_get_current_pid_tgid:
7450                 return &bpf_get_current_pid_tgid_proto;
7451         case BPF_FUNC_sk_storage_get:
7452                 return &bpf_sk_storage_get_proto;
7453         case BPF_FUNC_sk_storage_delete:
7454                 return &bpf_sk_storage_delete_proto;
7455 #ifdef CONFIG_CGROUPS
7456         case BPF_FUNC_get_current_cgroup_id:
7457                 return &bpf_get_current_cgroup_id_proto;
7458         case BPF_FUNC_get_current_ancestor_cgroup_id:
7459                 return &bpf_get_current_ancestor_cgroup_id_proto;
7460 #endif
7461 #ifdef CONFIG_CGROUP_NET_CLASSID
7462         case BPF_FUNC_get_cgroup_classid:
7463                 return &bpf_get_cgroup_classid_curr_proto;
7464 #endif
7465         default:
7466                 return bpf_sk_base_func_proto(func_id);
7467         }
7468 }
7469
7470 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7471 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7472
7473 static const struct bpf_func_proto *
7474 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7475 {
7476         switch (func_id) {
7477         case BPF_FUNC_skb_store_bytes:
7478                 return &bpf_skb_store_bytes_proto;
7479         case BPF_FUNC_skb_load_bytes:
7480                 return &bpf_skb_load_bytes_proto;
7481         case BPF_FUNC_skb_pull_data:
7482                 return &sk_skb_pull_data_proto;
7483         case BPF_FUNC_skb_change_tail:
7484                 return &sk_skb_change_tail_proto;
7485         case BPF_FUNC_skb_change_head:
7486                 return &sk_skb_change_head_proto;
7487         case BPF_FUNC_skb_adjust_room:
7488                 return &sk_skb_adjust_room_proto;
7489         case BPF_FUNC_get_socket_cookie:
7490                 return &bpf_get_socket_cookie_proto;
7491         case BPF_FUNC_get_socket_uid:
7492                 return &bpf_get_socket_uid_proto;
7493         case BPF_FUNC_sk_redirect_map:
7494                 return &bpf_sk_redirect_map_proto;
7495         case BPF_FUNC_sk_redirect_hash:
7496                 return &bpf_sk_redirect_hash_proto;
7497         case BPF_FUNC_perf_event_output:
7498                 return &bpf_skb_event_output_proto;
7499 #ifdef CONFIG_INET
7500         case BPF_FUNC_sk_lookup_tcp:
7501                 return &bpf_sk_lookup_tcp_proto;
7502         case BPF_FUNC_sk_lookup_udp:
7503                 return &bpf_sk_lookup_udp_proto;
7504         case BPF_FUNC_sk_release:
7505                 return &bpf_sk_release_proto;
7506         case BPF_FUNC_skc_lookup_tcp:
7507                 return &bpf_skc_lookup_tcp_proto;
7508 #endif
7509         default:
7510                 return bpf_sk_base_func_proto(func_id);
7511         }
7512 }
7513
7514 static const struct bpf_func_proto *
7515 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7516 {
7517         switch (func_id) {
7518         case BPF_FUNC_skb_load_bytes:
7519                 return &bpf_flow_dissector_load_bytes_proto;
7520         default:
7521                 return bpf_sk_base_func_proto(func_id);
7522         }
7523 }
7524
7525 static const struct bpf_func_proto *
7526 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7527 {
7528         switch (func_id) {
7529         case BPF_FUNC_skb_load_bytes:
7530                 return &bpf_skb_load_bytes_proto;
7531         case BPF_FUNC_skb_pull_data:
7532                 return &bpf_skb_pull_data_proto;
7533         case BPF_FUNC_csum_diff:
7534                 return &bpf_csum_diff_proto;
7535         case BPF_FUNC_get_cgroup_classid:
7536                 return &bpf_get_cgroup_classid_proto;
7537         case BPF_FUNC_get_route_realm:
7538                 return &bpf_get_route_realm_proto;
7539         case BPF_FUNC_get_hash_recalc:
7540                 return &bpf_get_hash_recalc_proto;
7541         case BPF_FUNC_perf_event_output:
7542                 return &bpf_skb_event_output_proto;
7543         case BPF_FUNC_get_smp_processor_id:
7544                 return &bpf_get_smp_processor_id_proto;
7545         case BPF_FUNC_skb_under_cgroup:
7546                 return &bpf_skb_under_cgroup_proto;
7547         default:
7548                 return bpf_sk_base_func_proto(func_id);
7549         }
7550 }
7551
7552 static const struct bpf_func_proto *
7553 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7554 {
7555         switch (func_id) {
7556         case BPF_FUNC_lwt_push_encap:
7557                 return &bpf_lwt_in_push_encap_proto;
7558         default:
7559                 return lwt_out_func_proto(func_id, prog);
7560         }
7561 }
7562
7563 static const struct bpf_func_proto *
7564 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7565 {
7566         switch (func_id) {
7567         case BPF_FUNC_skb_get_tunnel_key:
7568                 return &bpf_skb_get_tunnel_key_proto;
7569         case BPF_FUNC_skb_set_tunnel_key:
7570                 return bpf_get_skb_set_tunnel_proto(func_id);
7571         case BPF_FUNC_skb_get_tunnel_opt:
7572                 return &bpf_skb_get_tunnel_opt_proto;
7573         case BPF_FUNC_skb_set_tunnel_opt:
7574                 return bpf_get_skb_set_tunnel_proto(func_id);
7575         case BPF_FUNC_redirect:
7576                 return &bpf_redirect_proto;
7577         case BPF_FUNC_clone_redirect:
7578                 return &bpf_clone_redirect_proto;
7579         case BPF_FUNC_skb_change_tail:
7580                 return &bpf_skb_change_tail_proto;
7581         case BPF_FUNC_skb_change_head:
7582                 return &bpf_skb_change_head_proto;
7583         case BPF_FUNC_skb_store_bytes:
7584                 return &bpf_skb_store_bytes_proto;
7585         case BPF_FUNC_csum_update:
7586                 return &bpf_csum_update_proto;
7587         case BPF_FUNC_csum_level:
7588                 return &bpf_csum_level_proto;
7589         case BPF_FUNC_l3_csum_replace:
7590                 return &bpf_l3_csum_replace_proto;
7591         case BPF_FUNC_l4_csum_replace:
7592                 return &bpf_l4_csum_replace_proto;
7593         case BPF_FUNC_set_hash_invalid:
7594                 return &bpf_set_hash_invalid_proto;
7595         case BPF_FUNC_lwt_push_encap:
7596                 return &bpf_lwt_xmit_push_encap_proto;
7597         default:
7598                 return lwt_out_func_proto(func_id, prog);
7599         }
7600 }
7601
7602 static const struct bpf_func_proto *
7603 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7604 {
7605         switch (func_id) {
7606 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7607         case BPF_FUNC_lwt_seg6_store_bytes:
7608                 return &bpf_lwt_seg6_store_bytes_proto;
7609         case BPF_FUNC_lwt_seg6_action:
7610                 return &bpf_lwt_seg6_action_proto;
7611         case BPF_FUNC_lwt_seg6_adjust_srh:
7612                 return &bpf_lwt_seg6_adjust_srh_proto;
7613 #endif
7614         default:
7615                 return lwt_out_func_proto(func_id, prog);
7616         }
7617 }
7618
7619 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7620                                     const struct bpf_prog *prog,
7621                                     struct bpf_insn_access_aux *info)
7622 {
7623         const int size_default = sizeof(__u32);
7624
7625         if (off < 0 || off >= sizeof(struct __sk_buff))
7626                 return false;
7627
7628         /* The verifier guarantees that size > 0. */
7629         if (off % size != 0)
7630                 return false;
7631
7632         switch (off) {
7633         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7634                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7635                         return false;
7636                 break;
7637         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7638         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7639         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7640         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
7641         case bpf_ctx_range(struct __sk_buff, data):
7642         case bpf_ctx_range(struct __sk_buff, data_meta):
7643         case bpf_ctx_range(struct __sk_buff, data_end):
7644                 if (size != size_default)
7645                         return false;
7646                 break;
7647         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7648                 return false;
7649         case bpf_ctx_range(struct __sk_buff, tstamp):
7650                 if (size != sizeof(__u64))
7651                         return false;
7652                 break;
7653         case offsetof(struct __sk_buff, sk):
7654                 if (type == BPF_WRITE || size != sizeof(__u64))
7655                         return false;
7656                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7657                 break;
7658         default:
7659                 /* Only narrow read access allowed for now. */
7660                 if (type == BPF_WRITE) {
7661                         if (size != size_default)
7662                                 return false;
7663                 } else {
7664                         bpf_ctx_record_field_size(info, size_default);
7665                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7666                                 return false;
7667                 }
7668         }
7669
7670         return true;
7671 }
7672
7673 static bool sk_filter_is_valid_access(int off, int size,
7674                                       enum bpf_access_type type,
7675                                       const struct bpf_prog *prog,
7676                                       struct bpf_insn_access_aux *info)
7677 {
7678         switch (off) {
7679         case bpf_ctx_range(struct __sk_buff, tc_classid):
7680         case bpf_ctx_range(struct __sk_buff, data):
7681         case bpf_ctx_range(struct __sk_buff, data_meta):
7682         case bpf_ctx_range(struct __sk_buff, data_end):
7683         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7684         case bpf_ctx_range(struct __sk_buff, tstamp):
7685         case bpf_ctx_range(struct __sk_buff, wire_len):
7686                 return false;
7687         }
7688
7689         if (type == BPF_WRITE) {
7690                 switch (off) {
7691                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7692                         break;
7693                 default:
7694                         return false;
7695                 }
7696         }
7697
7698         return bpf_skb_is_valid_access(off, size, type, prog, info);
7699 }
7700
7701 static bool cg_skb_is_valid_access(int off, int size,
7702                                    enum bpf_access_type type,
7703                                    const struct bpf_prog *prog,
7704                                    struct bpf_insn_access_aux *info)
7705 {
7706         switch (off) {
7707         case bpf_ctx_range(struct __sk_buff, tc_classid):
7708         case bpf_ctx_range(struct __sk_buff, data_meta):
7709         case bpf_ctx_range(struct __sk_buff, wire_len):
7710                 return false;
7711         case bpf_ctx_range(struct __sk_buff, data):
7712         case bpf_ctx_range(struct __sk_buff, data_end):
7713                 if (!bpf_capable())
7714                         return false;
7715                 break;
7716         }
7717
7718         if (type == BPF_WRITE) {
7719                 switch (off) {
7720                 case bpf_ctx_range(struct __sk_buff, mark):
7721                 case bpf_ctx_range(struct __sk_buff, priority):
7722                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7723                         break;
7724                 case bpf_ctx_range(struct __sk_buff, tstamp):
7725                         if (!bpf_capable())
7726                                 return false;
7727                         break;
7728                 default:
7729                         return false;
7730                 }
7731         }
7732
7733         switch (off) {
7734         case bpf_ctx_range(struct __sk_buff, data):
7735                 info->reg_type = PTR_TO_PACKET;
7736                 break;
7737         case bpf_ctx_range(struct __sk_buff, data_end):
7738                 info->reg_type = PTR_TO_PACKET_END;
7739                 break;
7740         }
7741
7742         return bpf_skb_is_valid_access(off, size, type, prog, info);
7743 }
7744
7745 static bool lwt_is_valid_access(int off, int size,
7746                                 enum bpf_access_type type,
7747                                 const struct bpf_prog *prog,
7748                                 struct bpf_insn_access_aux *info)
7749 {
7750         switch (off) {
7751         case bpf_ctx_range(struct __sk_buff, tc_classid):
7752         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7753         case bpf_ctx_range(struct __sk_buff, data_meta):
7754         case bpf_ctx_range(struct __sk_buff, tstamp):
7755         case bpf_ctx_range(struct __sk_buff, wire_len):
7756                 return false;
7757         }
7758
7759         if (type == BPF_WRITE) {
7760                 switch (off) {
7761                 case bpf_ctx_range(struct __sk_buff, mark):
7762                 case bpf_ctx_range(struct __sk_buff, priority):
7763                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7764                         break;
7765                 default:
7766                         return false;
7767                 }
7768         }
7769
7770         switch (off) {
7771         case bpf_ctx_range(struct __sk_buff, data):
7772                 info->reg_type = PTR_TO_PACKET;
7773                 break;
7774         case bpf_ctx_range(struct __sk_buff, data_end):
7775                 info->reg_type = PTR_TO_PACKET_END;
7776                 break;
7777         }
7778
7779         return bpf_skb_is_valid_access(off, size, type, prog, info);
7780 }
7781
7782 /* Attach type specific accesses */
7783 static bool __sock_filter_check_attach_type(int off,
7784                                             enum bpf_access_type access_type,
7785                                             enum bpf_attach_type attach_type)
7786 {
7787         switch (off) {
7788         case offsetof(struct bpf_sock, bound_dev_if):
7789         case offsetof(struct bpf_sock, mark):
7790         case offsetof(struct bpf_sock, priority):
7791                 switch (attach_type) {
7792                 case BPF_CGROUP_INET_SOCK_CREATE:
7793                 case BPF_CGROUP_INET_SOCK_RELEASE:
7794                         goto full_access;
7795                 default:
7796                         return false;
7797                 }
7798         case bpf_ctx_range(struct bpf_sock, src_ip4):
7799                 switch (attach_type) {
7800                 case BPF_CGROUP_INET4_POST_BIND:
7801                         goto read_only;
7802                 default:
7803                         return false;
7804                 }
7805         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7806                 switch (attach_type) {
7807                 case BPF_CGROUP_INET6_POST_BIND:
7808                         goto read_only;
7809                 default:
7810                         return false;
7811                 }
7812         case bpf_ctx_range(struct bpf_sock, src_port):
7813                 switch (attach_type) {
7814                 case BPF_CGROUP_INET4_POST_BIND:
7815                 case BPF_CGROUP_INET6_POST_BIND:
7816                         goto read_only;
7817                 default:
7818                         return false;
7819                 }
7820         }
7821 read_only:
7822         return access_type == BPF_READ;
7823 full_access:
7824         return true;
7825 }
7826
7827 bool bpf_sock_common_is_valid_access(int off, int size,
7828                                      enum bpf_access_type type,
7829                                      struct bpf_insn_access_aux *info)
7830 {
7831         switch (off) {
7832         case bpf_ctx_range_till(struct bpf_sock, type, priority):
7833                 return false;
7834         default:
7835                 return bpf_sock_is_valid_access(off, size, type, info);
7836         }
7837 }
7838
7839 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7840                               struct bpf_insn_access_aux *info)
7841 {
7842         const int size_default = sizeof(__u32);
7843
7844         if (off < 0 || off >= sizeof(struct bpf_sock))
7845                 return false;
7846         if (off % size != 0)
7847                 return false;
7848
7849         switch (off) {
7850         case offsetof(struct bpf_sock, state):
7851         case offsetof(struct bpf_sock, family):
7852         case offsetof(struct bpf_sock, type):
7853         case offsetof(struct bpf_sock, protocol):
7854         case offsetof(struct bpf_sock, dst_port):
7855         case offsetof(struct bpf_sock, src_port):
7856         case offsetof(struct bpf_sock, rx_queue_mapping):
7857         case bpf_ctx_range(struct bpf_sock, src_ip4):
7858         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7859         case bpf_ctx_range(struct bpf_sock, dst_ip4):
7860         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7861                 bpf_ctx_record_field_size(info, size_default);
7862                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7863         }
7864
7865         return size == size_default;
7866 }
7867
7868 static bool sock_filter_is_valid_access(int off, int size,
7869                                         enum bpf_access_type type,
7870                                         const struct bpf_prog *prog,
7871                                         struct bpf_insn_access_aux *info)
7872 {
7873         if (!bpf_sock_is_valid_access(off, size, type, info))
7874                 return false;
7875         return __sock_filter_check_attach_type(off, type,
7876                                                prog->expected_attach_type);
7877 }
7878
7879 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7880                              const struct bpf_prog *prog)
7881 {
7882         /* Neither direct read nor direct write requires any preliminary
7883          * action.
7884          */
7885         return 0;
7886 }
7887
7888 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7889                                 const struct bpf_prog *prog, int drop_verdict)
7890 {
7891         struct bpf_insn *insn = insn_buf;
7892
7893         if (!direct_write)
7894                 return 0;
7895
7896         /* if (!skb->cloned)
7897          *       goto start;
7898          *
7899          * (Fast-path, otherwise approximation that we might be
7900          *  a clone, do the rest in helper.)
7901          */
7902         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7903         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7904         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7905
7906         /* ret = bpf_skb_pull_data(skb, 0); */
7907         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7908         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7909         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7910                                BPF_FUNC_skb_pull_data);
7911         /* if (!ret)
7912          *      goto restore;
7913          * return TC_ACT_SHOT;
7914          */
7915         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7916         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7917         *insn++ = BPF_EXIT_INSN();
7918
7919         /* restore: */
7920         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7921         /* start: */
7922         *insn++ = prog->insnsi[0];
7923
7924         return insn - insn_buf;
7925 }
7926
7927 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7928                           struct bpf_insn *insn_buf)
7929 {
7930         bool indirect = BPF_MODE(orig->code) == BPF_IND;
7931         struct bpf_insn *insn = insn_buf;
7932
7933         if (!indirect) {
7934                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7935         } else {
7936                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7937                 if (orig->imm)
7938                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7939         }
7940         /* We're guaranteed here that CTX is in R6. */
7941         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7942
7943         switch (BPF_SIZE(orig->code)) {
7944         case BPF_B:
7945                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7946                 break;
7947         case BPF_H:
7948                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7949                 break;
7950         case BPF_W:
7951                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7952                 break;
7953         }
7954
7955         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7956         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7957         *insn++ = BPF_EXIT_INSN();
7958
7959         return insn - insn_buf;
7960 }
7961
7962 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
7963                                const struct bpf_prog *prog)
7964 {
7965         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
7966 }
7967
7968 static bool tc_cls_act_is_valid_access(int off, int size,
7969                                        enum bpf_access_type type,
7970                                        const struct bpf_prog *prog,
7971                                        struct bpf_insn_access_aux *info)
7972 {
7973         if (type == BPF_WRITE) {
7974                 switch (off) {
7975                 case bpf_ctx_range(struct __sk_buff, mark):
7976                 case bpf_ctx_range(struct __sk_buff, tc_index):
7977                 case bpf_ctx_range(struct __sk_buff, priority):
7978                 case bpf_ctx_range(struct __sk_buff, tc_classid):
7979                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7980                 case bpf_ctx_range(struct __sk_buff, tstamp):
7981                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
7982                         break;
7983                 default:
7984                         return false;
7985                 }
7986         }
7987
7988         switch (off) {
7989         case bpf_ctx_range(struct __sk_buff, data):
7990                 info->reg_type = PTR_TO_PACKET;
7991                 break;
7992         case bpf_ctx_range(struct __sk_buff, data_meta):
7993                 info->reg_type = PTR_TO_PACKET_META;
7994                 break;
7995         case bpf_ctx_range(struct __sk_buff, data_end):
7996                 info->reg_type = PTR_TO_PACKET_END;
7997                 break;
7998         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7999                 return false;
8000         }
8001
8002         return bpf_skb_is_valid_access(off, size, type, prog, info);
8003 }
8004
8005 static bool __is_valid_xdp_access(int off, int size)
8006 {
8007         if (off < 0 || off >= sizeof(struct xdp_md))
8008                 return false;
8009         if (off % size != 0)
8010                 return false;
8011         if (size != sizeof(__u32))
8012                 return false;
8013
8014         return true;
8015 }
8016
8017 static bool xdp_is_valid_access(int off, int size,
8018                                 enum bpf_access_type type,
8019                                 const struct bpf_prog *prog,
8020                                 struct bpf_insn_access_aux *info)
8021 {
8022         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8023                 switch (off) {
8024                 case offsetof(struct xdp_md, egress_ifindex):
8025                         return false;
8026                 }
8027         }
8028
8029         if (type == BPF_WRITE) {
8030                 if (bpf_prog_is_dev_bound(prog->aux)) {
8031                         switch (off) {
8032                         case offsetof(struct xdp_md, rx_queue_index):
8033                                 return __is_valid_xdp_access(off, size);
8034                         }
8035                 }
8036                 return false;
8037         }
8038
8039         switch (off) {
8040         case offsetof(struct xdp_md, data):
8041                 info->reg_type = PTR_TO_PACKET;
8042                 break;
8043         case offsetof(struct xdp_md, data_meta):
8044                 info->reg_type = PTR_TO_PACKET_META;
8045                 break;
8046         case offsetof(struct xdp_md, data_end):
8047                 info->reg_type = PTR_TO_PACKET_END;
8048                 break;
8049         }
8050
8051         return __is_valid_xdp_access(off, size);
8052 }
8053
8054 void bpf_warn_invalid_xdp_action(u32 act)
8055 {
8056         const u32 act_max = XDP_REDIRECT;
8057
8058         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8059                   act > act_max ? "Illegal" : "Driver unsupported",
8060                   act);
8061 }
8062 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8063
8064 static bool sock_addr_is_valid_access(int off, int size,
8065                                       enum bpf_access_type type,
8066                                       const struct bpf_prog *prog,
8067                                       struct bpf_insn_access_aux *info)
8068 {
8069         const int size_default = sizeof(__u32);
8070
8071         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8072                 return false;
8073         if (off % size != 0)
8074                 return false;
8075
8076         /* Disallow access to IPv6 fields from IPv4 contex and vise
8077          * versa.
8078          */
8079         switch (off) {
8080         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8081                 switch (prog->expected_attach_type) {
8082                 case BPF_CGROUP_INET4_BIND:
8083                 case BPF_CGROUP_INET4_CONNECT:
8084                 case BPF_CGROUP_INET4_GETPEERNAME:
8085                 case BPF_CGROUP_INET4_GETSOCKNAME:
8086                 case BPF_CGROUP_UDP4_SENDMSG:
8087                 case BPF_CGROUP_UDP4_RECVMSG:
8088                         break;
8089                 default:
8090                         return false;
8091                 }
8092                 break;
8093         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8094                 switch (prog->expected_attach_type) {
8095                 case BPF_CGROUP_INET6_BIND:
8096                 case BPF_CGROUP_INET6_CONNECT:
8097                 case BPF_CGROUP_INET6_GETPEERNAME:
8098                 case BPF_CGROUP_INET6_GETSOCKNAME:
8099                 case BPF_CGROUP_UDP6_SENDMSG:
8100                 case BPF_CGROUP_UDP6_RECVMSG:
8101                         break;
8102                 default:
8103                         return false;
8104                 }
8105                 break;
8106         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8107                 switch (prog->expected_attach_type) {
8108                 case BPF_CGROUP_UDP4_SENDMSG:
8109                         break;
8110                 default:
8111                         return false;
8112                 }
8113                 break;
8114         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8115                                 msg_src_ip6[3]):
8116                 switch (prog->expected_attach_type) {
8117                 case BPF_CGROUP_UDP6_SENDMSG:
8118                         break;
8119                 default:
8120                         return false;
8121                 }
8122                 break;
8123         }
8124
8125         switch (off) {
8126         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8127         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8128         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8129         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8130                                 msg_src_ip6[3]):
8131         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8132                 if (type == BPF_READ) {
8133                         bpf_ctx_record_field_size(info, size_default);
8134
8135                         if (bpf_ctx_wide_access_ok(off, size,
8136                                                    struct bpf_sock_addr,
8137                                                    user_ip6))
8138                                 return true;
8139
8140                         if (bpf_ctx_wide_access_ok(off, size,
8141                                                    struct bpf_sock_addr,
8142                                                    msg_src_ip6))
8143                                 return true;
8144
8145                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8146                                 return false;
8147                 } else {
8148                         if (bpf_ctx_wide_access_ok(off, size,
8149                                                    struct bpf_sock_addr,
8150                                                    user_ip6))
8151                                 return true;
8152
8153                         if (bpf_ctx_wide_access_ok(off, size,
8154                                                    struct bpf_sock_addr,
8155                                                    msg_src_ip6))
8156                                 return true;
8157
8158                         if (size != size_default)
8159                                 return false;
8160                 }
8161                 break;
8162         case offsetof(struct bpf_sock_addr, sk):
8163                 if (type != BPF_READ)
8164                         return false;
8165                 if (size != sizeof(__u64))
8166                         return false;
8167                 info->reg_type = PTR_TO_SOCKET;
8168                 break;
8169         default:
8170                 if (type == BPF_READ) {
8171                         if (size != size_default)
8172                                 return false;
8173                 } else {
8174                         return false;
8175                 }
8176         }
8177
8178         return true;
8179 }
8180
8181 static bool sock_ops_is_valid_access(int off, int size,
8182                                      enum bpf_access_type type,
8183                                      const struct bpf_prog *prog,
8184                                      struct bpf_insn_access_aux *info)
8185 {
8186         const int size_default = sizeof(__u32);
8187
8188         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8189                 return false;
8190
8191         /* The verifier guarantees that size > 0. */
8192         if (off % size != 0)
8193                 return false;
8194
8195         if (type == BPF_WRITE) {
8196                 switch (off) {
8197                 case offsetof(struct bpf_sock_ops, reply):
8198                 case offsetof(struct bpf_sock_ops, sk_txhash):
8199                         if (size != size_default)
8200                                 return false;
8201                         break;
8202                 default:
8203                         return false;
8204                 }
8205         } else {
8206                 switch (off) {
8207                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8208                                         bytes_acked):
8209                         if (size != sizeof(__u64))
8210                                 return false;
8211                         break;
8212                 case offsetof(struct bpf_sock_ops, sk):
8213                         if (size != sizeof(__u64))
8214                                 return false;
8215                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8216                         break;
8217                 case offsetof(struct bpf_sock_ops, skb_data):
8218                         if (size != sizeof(__u64))
8219                                 return false;
8220                         info->reg_type = PTR_TO_PACKET;
8221                         break;
8222                 case offsetof(struct bpf_sock_ops, skb_data_end):
8223                         if (size != sizeof(__u64))
8224                                 return false;
8225                         info->reg_type = PTR_TO_PACKET_END;
8226                         break;
8227                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8228                         bpf_ctx_record_field_size(info, size_default);
8229                         return bpf_ctx_narrow_access_ok(off, size,
8230                                                         size_default);
8231                 default:
8232                         if (size != size_default)
8233                                 return false;
8234                         break;
8235                 }
8236         }
8237
8238         return true;
8239 }
8240
8241 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8242                            const struct bpf_prog *prog)
8243 {
8244         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8245 }
8246
8247 static bool sk_skb_is_valid_access(int off, int size,
8248                                    enum bpf_access_type type,
8249                                    const struct bpf_prog *prog,
8250                                    struct bpf_insn_access_aux *info)
8251 {
8252         switch (off) {
8253         case bpf_ctx_range(struct __sk_buff, tc_classid):
8254         case bpf_ctx_range(struct __sk_buff, data_meta):
8255         case bpf_ctx_range(struct __sk_buff, tstamp):
8256         case bpf_ctx_range(struct __sk_buff, wire_len):
8257                 return false;
8258         }
8259
8260         if (type == BPF_WRITE) {
8261                 switch (off) {
8262                 case bpf_ctx_range(struct __sk_buff, tc_index):
8263                 case bpf_ctx_range(struct __sk_buff, priority):
8264                         break;
8265                 default:
8266                         return false;
8267                 }
8268         }
8269
8270         switch (off) {
8271         case bpf_ctx_range(struct __sk_buff, mark):
8272                 return false;
8273         case bpf_ctx_range(struct __sk_buff, data):
8274                 info->reg_type = PTR_TO_PACKET;
8275                 break;
8276         case bpf_ctx_range(struct __sk_buff, data_end):
8277                 info->reg_type = PTR_TO_PACKET_END;
8278                 break;
8279         }
8280
8281         return bpf_skb_is_valid_access(off, size, type, prog, info);
8282 }
8283
8284 static bool sk_msg_is_valid_access(int off, int size,
8285                                    enum bpf_access_type type,
8286                                    const struct bpf_prog *prog,
8287                                    struct bpf_insn_access_aux *info)
8288 {
8289         if (type == BPF_WRITE)
8290                 return false;
8291
8292         if (off % size != 0)
8293                 return false;
8294
8295         switch (off) {
8296         case offsetof(struct sk_msg_md, data):
8297                 info->reg_type = PTR_TO_PACKET;
8298                 if (size != sizeof(__u64))
8299                         return false;
8300                 break;
8301         case offsetof(struct sk_msg_md, data_end):
8302                 info->reg_type = PTR_TO_PACKET_END;
8303                 if (size != sizeof(__u64))
8304                         return false;
8305                 break;
8306         case offsetof(struct sk_msg_md, sk):
8307                 if (size != sizeof(__u64))
8308                         return false;
8309                 info->reg_type = PTR_TO_SOCKET;
8310                 break;
8311         case bpf_ctx_range(struct sk_msg_md, family):
8312         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8313         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8314         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8315         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8316         case bpf_ctx_range(struct sk_msg_md, remote_port):
8317         case bpf_ctx_range(struct sk_msg_md, local_port):
8318         case bpf_ctx_range(struct sk_msg_md, size):
8319                 if (size != sizeof(__u32))
8320                         return false;
8321                 break;
8322         default:
8323                 return false;
8324         }
8325         return true;
8326 }
8327
8328 static bool flow_dissector_is_valid_access(int off, int size,
8329                                            enum bpf_access_type type,
8330                                            const struct bpf_prog *prog,
8331                                            struct bpf_insn_access_aux *info)
8332 {
8333         const int size_default = sizeof(__u32);
8334
8335         if (off < 0 || off >= sizeof(struct __sk_buff))
8336                 return false;
8337
8338         if (type == BPF_WRITE)
8339                 return false;
8340
8341         switch (off) {
8342         case bpf_ctx_range(struct __sk_buff, data):
8343                 if (size != size_default)
8344                         return false;
8345                 info->reg_type = PTR_TO_PACKET;
8346                 return true;
8347         case bpf_ctx_range(struct __sk_buff, data_end):
8348                 if (size != size_default)
8349                         return false;
8350                 info->reg_type = PTR_TO_PACKET_END;
8351                 return true;
8352         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8353                 if (size != sizeof(__u64))
8354                         return false;
8355                 info->reg_type = PTR_TO_FLOW_KEYS;
8356                 return true;
8357         default:
8358                 return false;
8359         }
8360 }
8361
8362 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8363                                              const struct bpf_insn *si,
8364                                              struct bpf_insn *insn_buf,
8365                                              struct bpf_prog *prog,
8366                                              u32 *target_size)
8367
8368 {
8369         struct bpf_insn *insn = insn_buf;
8370
8371         switch (si->off) {
8372         case offsetof(struct __sk_buff, data):
8373                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8374                                       si->dst_reg, si->src_reg,
8375                                       offsetof(struct bpf_flow_dissector, data));
8376                 break;
8377
8378         case offsetof(struct __sk_buff, data_end):
8379                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8380                                       si->dst_reg, si->src_reg,
8381                                       offsetof(struct bpf_flow_dissector, data_end));
8382                 break;
8383
8384         case offsetof(struct __sk_buff, flow_keys):
8385                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8386                                       si->dst_reg, si->src_reg,
8387                                       offsetof(struct bpf_flow_dissector, flow_keys));
8388                 break;
8389         }
8390
8391         return insn - insn_buf;
8392 }
8393
8394 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8395                                                   struct bpf_insn *insn)
8396 {
8397         /* si->dst_reg = skb_shinfo(SKB); */
8398 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8399         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8400                               BPF_REG_AX, si->src_reg,
8401                               offsetof(struct sk_buff, end));
8402         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8403                               si->dst_reg, si->src_reg,
8404                               offsetof(struct sk_buff, head));
8405         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8406 #else
8407         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8408                               si->dst_reg, si->src_reg,
8409                               offsetof(struct sk_buff, end));
8410 #endif
8411
8412         return insn;
8413 }
8414
8415 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8416                                   const struct bpf_insn *si,
8417                                   struct bpf_insn *insn_buf,
8418                                   struct bpf_prog *prog, u32 *target_size)
8419 {
8420         struct bpf_insn *insn = insn_buf;
8421         int off;
8422
8423         switch (si->off) {
8424         case offsetof(struct __sk_buff, len):
8425                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8426                                       bpf_target_off(struct sk_buff, len, 4,
8427                                                      target_size));
8428                 break;
8429
8430         case offsetof(struct __sk_buff, protocol):
8431                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8432                                       bpf_target_off(struct sk_buff, protocol, 2,
8433                                                      target_size));
8434                 break;
8435
8436         case offsetof(struct __sk_buff, vlan_proto):
8437                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8438                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8439                                                      target_size));
8440                 break;
8441
8442         case offsetof(struct __sk_buff, priority):
8443                 if (type == BPF_WRITE)
8444                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8445                                               bpf_target_off(struct sk_buff, priority, 4,
8446                                                              target_size));
8447                 else
8448                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8449                                               bpf_target_off(struct sk_buff, priority, 4,
8450                                                              target_size));
8451                 break;
8452
8453         case offsetof(struct __sk_buff, ingress_ifindex):
8454                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8455                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8456                                                      target_size));
8457                 break;
8458
8459         case offsetof(struct __sk_buff, ifindex):
8460                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8461                                       si->dst_reg, si->src_reg,
8462                                       offsetof(struct sk_buff, dev));
8463                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8464                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8465                                       bpf_target_off(struct net_device, ifindex, 4,
8466                                                      target_size));
8467                 break;
8468
8469         case offsetof(struct __sk_buff, hash):
8470                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8471                                       bpf_target_off(struct sk_buff, hash, 4,
8472                                                      target_size));
8473                 break;
8474
8475         case offsetof(struct __sk_buff, mark):
8476                 if (type == BPF_WRITE)
8477                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8478                                               bpf_target_off(struct sk_buff, mark, 4,
8479                                                              target_size));
8480                 else
8481                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8482                                               bpf_target_off(struct sk_buff, mark, 4,
8483                                                              target_size));
8484                 break;
8485
8486         case offsetof(struct __sk_buff, pkt_type):
8487                 *target_size = 1;
8488                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8489                                       PKT_TYPE_OFFSET());
8490                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8491 #ifdef __BIG_ENDIAN_BITFIELD
8492                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8493 #endif
8494                 break;
8495
8496         case offsetof(struct __sk_buff, queue_mapping):
8497                 if (type == BPF_WRITE) {
8498                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8499                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8500                                               bpf_target_off(struct sk_buff,
8501                                                              queue_mapping,
8502                                                              2, target_size));
8503                 } else {
8504                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8505                                               bpf_target_off(struct sk_buff,
8506                                                              queue_mapping,
8507                                                              2, target_size));
8508                 }
8509                 break;
8510
8511         case offsetof(struct __sk_buff, vlan_present):
8512                 *target_size = 1;
8513                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8514                                       PKT_VLAN_PRESENT_OFFSET());
8515                 if (PKT_VLAN_PRESENT_BIT)
8516                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8517                 if (PKT_VLAN_PRESENT_BIT < 7)
8518                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8519                 break;
8520
8521         case offsetof(struct __sk_buff, vlan_tci):
8522                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8523                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8524                                                      target_size));
8525                 break;
8526
8527         case offsetof(struct __sk_buff, cb[0]) ...
8528              offsetofend(struct __sk_buff, cb[4]) - 1:
8529                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8530                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8531                               offsetof(struct qdisc_skb_cb, data)) %
8532                              sizeof(__u64));
8533
8534                 prog->cb_access = 1;
8535                 off  = si->off;
8536                 off -= offsetof(struct __sk_buff, cb[0]);
8537                 off += offsetof(struct sk_buff, cb);
8538                 off += offsetof(struct qdisc_skb_cb, data);
8539                 if (type == BPF_WRITE)
8540                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8541                                               si->src_reg, off);
8542                 else
8543                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8544                                               si->src_reg, off);
8545                 break;
8546
8547         case offsetof(struct __sk_buff, tc_classid):
8548                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8549
8550                 off  = si->off;
8551                 off -= offsetof(struct __sk_buff, tc_classid);
8552                 off += offsetof(struct sk_buff, cb);
8553                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8554                 *target_size = 2;
8555                 if (type == BPF_WRITE)
8556                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8557                                               si->src_reg, off);
8558                 else
8559                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8560                                               si->src_reg, off);
8561                 break;
8562
8563         case offsetof(struct __sk_buff, data):
8564                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8565                                       si->dst_reg, si->src_reg,
8566                                       offsetof(struct sk_buff, data));
8567                 break;
8568
8569         case offsetof(struct __sk_buff, data_meta):
8570                 off  = si->off;
8571                 off -= offsetof(struct __sk_buff, data_meta);
8572                 off += offsetof(struct sk_buff, cb);
8573                 off += offsetof(struct bpf_skb_data_end, data_meta);
8574                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8575                                       si->src_reg, off);
8576                 break;
8577
8578         case offsetof(struct __sk_buff, data_end):
8579                 off  = si->off;
8580                 off -= offsetof(struct __sk_buff, data_end);
8581                 off += offsetof(struct sk_buff, cb);
8582                 off += offsetof(struct bpf_skb_data_end, data_end);
8583                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8584                                       si->src_reg, off);
8585                 break;
8586
8587         case offsetof(struct __sk_buff, tc_index):
8588 #ifdef CONFIG_NET_SCHED
8589                 if (type == BPF_WRITE)
8590                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8591                                               bpf_target_off(struct sk_buff, tc_index, 2,
8592                                                              target_size));
8593                 else
8594                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8595                                               bpf_target_off(struct sk_buff, tc_index, 2,
8596                                                              target_size));
8597 #else
8598                 *target_size = 2;
8599                 if (type == BPF_WRITE)
8600                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8601                 else
8602                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8603 #endif
8604                 break;
8605
8606         case offsetof(struct __sk_buff, napi_id):
8607 #if defined(CONFIG_NET_RX_BUSY_POLL)
8608                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8609                                       bpf_target_off(struct sk_buff, napi_id, 4,
8610                                                      target_size));
8611                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8612                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8613 #else
8614                 *target_size = 4;
8615                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8616 #endif
8617                 break;
8618         case offsetof(struct __sk_buff, family):
8619                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8620
8621                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8622                                       si->dst_reg, si->src_reg,
8623                                       offsetof(struct sk_buff, sk));
8624                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8625                                       bpf_target_off(struct sock_common,
8626                                                      skc_family,
8627                                                      2, target_size));
8628                 break;
8629         case offsetof(struct __sk_buff, remote_ip4):
8630                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8631
8632                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8633                                       si->dst_reg, si->src_reg,
8634                                       offsetof(struct sk_buff, sk));
8635                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8636                                       bpf_target_off(struct sock_common,
8637                                                      skc_daddr,
8638                                                      4, target_size));
8639                 break;
8640         case offsetof(struct __sk_buff, local_ip4):
8641                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8642                                           skc_rcv_saddr) != 4);
8643
8644                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8645                                       si->dst_reg, si->src_reg,
8646                                       offsetof(struct sk_buff, sk));
8647                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8648                                       bpf_target_off(struct sock_common,
8649                                                      skc_rcv_saddr,
8650                                                      4, target_size));
8651                 break;
8652         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8653              offsetof(struct __sk_buff, remote_ip6[3]):
8654 #if IS_ENABLED(CONFIG_IPV6)
8655                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8656                                           skc_v6_daddr.s6_addr32[0]) != 4);
8657
8658                 off = si->off;
8659                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8660
8661                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8662                                       si->dst_reg, si->src_reg,
8663                                       offsetof(struct sk_buff, sk));
8664                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8665                                       offsetof(struct sock_common,
8666                                                skc_v6_daddr.s6_addr32[0]) +
8667                                       off);
8668 #else
8669                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8670 #endif
8671                 break;
8672         case offsetof(struct __sk_buff, local_ip6[0]) ...
8673              offsetof(struct __sk_buff, local_ip6[3]):
8674 #if IS_ENABLED(CONFIG_IPV6)
8675                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8676                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8677
8678                 off = si->off;
8679                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8680
8681                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8682                                       si->dst_reg, si->src_reg,
8683                                       offsetof(struct sk_buff, sk));
8684                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8685                                       offsetof(struct sock_common,
8686                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8687                                       off);
8688 #else
8689                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8690 #endif
8691                 break;
8692
8693         case offsetof(struct __sk_buff, remote_port):
8694                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8695
8696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8697                                       si->dst_reg, si->src_reg,
8698                                       offsetof(struct sk_buff, sk));
8699                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8700                                       bpf_target_off(struct sock_common,
8701                                                      skc_dport,
8702                                                      2, target_size));
8703 #ifndef __BIG_ENDIAN_BITFIELD
8704                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8705 #endif
8706                 break;
8707
8708         case offsetof(struct __sk_buff, local_port):
8709                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8710
8711                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8712                                       si->dst_reg, si->src_reg,
8713                                       offsetof(struct sk_buff, sk));
8714                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8715                                       bpf_target_off(struct sock_common,
8716                                                      skc_num, 2, target_size));
8717                 break;
8718
8719         case offsetof(struct __sk_buff, tstamp):
8720                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8721
8722                 if (type == BPF_WRITE)
8723                         *insn++ = BPF_STX_MEM(BPF_DW,
8724                                               si->dst_reg, si->src_reg,
8725                                               bpf_target_off(struct sk_buff,
8726                                                              tstamp, 8,
8727                                                              target_size));
8728                 else
8729                         *insn++ = BPF_LDX_MEM(BPF_DW,
8730                                               si->dst_reg, si->src_reg,
8731                                               bpf_target_off(struct sk_buff,
8732                                                              tstamp, 8,
8733                                                              target_size));
8734                 break;
8735
8736         case offsetof(struct __sk_buff, gso_segs):
8737                 insn = bpf_convert_shinfo_access(si, insn);
8738                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8739                                       si->dst_reg, si->dst_reg,
8740                                       bpf_target_off(struct skb_shared_info,
8741                                                      gso_segs, 2,
8742                                                      target_size));
8743                 break;
8744         case offsetof(struct __sk_buff, gso_size):
8745                 insn = bpf_convert_shinfo_access(si, insn);
8746                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8747                                       si->dst_reg, si->dst_reg,
8748                                       bpf_target_off(struct skb_shared_info,
8749                                                      gso_size, 2,
8750                                                      target_size));
8751                 break;
8752         case offsetof(struct __sk_buff, wire_len):
8753                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8754
8755                 off = si->off;
8756                 off -= offsetof(struct __sk_buff, wire_len);
8757                 off += offsetof(struct sk_buff, cb);
8758                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8759                 *target_size = 4;
8760                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8761                 break;
8762
8763         case offsetof(struct __sk_buff, sk):
8764                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8765                                       si->dst_reg, si->src_reg,
8766                                       offsetof(struct sk_buff, sk));
8767                 break;
8768         }
8769
8770         return insn - insn_buf;
8771 }
8772
8773 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8774                                 const struct bpf_insn *si,
8775                                 struct bpf_insn *insn_buf,
8776                                 struct bpf_prog *prog, u32 *target_size)
8777 {
8778         struct bpf_insn *insn = insn_buf;
8779         int off;
8780
8781         switch (si->off) {
8782         case offsetof(struct bpf_sock, bound_dev_if):
8783                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8784
8785                 if (type == BPF_WRITE)
8786                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8787                                         offsetof(struct sock, sk_bound_dev_if));
8788                 else
8789                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8790                                       offsetof(struct sock, sk_bound_dev_if));
8791                 break;
8792
8793         case offsetof(struct bpf_sock, mark):
8794                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8795
8796                 if (type == BPF_WRITE)
8797                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8798                                         offsetof(struct sock, sk_mark));
8799                 else
8800                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8801                                       offsetof(struct sock, sk_mark));
8802                 break;
8803
8804         case offsetof(struct bpf_sock, priority):
8805                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8806
8807                 if (type == BPF_WRITE)
8808                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8809                                         offsetof(struct sock, sk_priority));
8810                 else
8811                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8812                                       offsetof(struct sock, sk_priority));
8813                 break;
8814
8815         case offsetof(struct bpf_sock, family):
8816                 *insn++ = BPF_LDX_MEM(
8817                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8818                         si->dst_reg, si->src_reg,
8819                         bpf_target_off(struct sock_common,
8820                                        skc_family,
8821                                        sizeof_field(struct sock_common,
8822                                                     skc_family),
8823                                        target_size));
8824                 break;
8825
8826         case offsetof(struct bpf_sock, type):
8827                 *insn++ = BPF_LDX_MEM(
8828                         BPF_FIELD_SIZEOF(struct sock, sk_type),
8829                         si->dst_reg, si->src_reg,
8830                         bpf_target_off(struct sock, sk_type,
8831                                        sizeof_field(struct sock, sk_type),
8832                                        target_size));
8833                 break;
8834
8835         case offsetof(struct bpf_sock, protocol):
8836                 *insn++ = BPF_LDX_MEM(
8837                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8838                         si->dst_reg, si->src_reg,
8839                         bpf_target_off(struct sock, sk_protocol,
8840                                        sizeof_field(struct sock, sk_protocol),
8841                                        target_size));
8842                 break;
8843
8844         case offsetof(struct bpf_sock, src_ip4):
8845                 *insn++ = BPF_LDX_MEM(
8846                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8847                         bpf_target_off(struct sock_common, skc_rcv_saddr,
8848                                        sizeof_field(struct sock_common,
8849                                                     skc_rcv_saddr),
8850                                        target_size));
8851                 break;
8852
8853         case offsetof(struct bpf_sock, dst_ip4):
8854                 *insn++ = BPF_LDX_MEM(
8855                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8856                         bpf_target_off(struct sock_common, skc_daddr,
8857                                        sizeof_field(struct sock_common,
8858                                                     skc_daddr),
8859                                        target_size));
8860                 break;
8861
8862         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8863 #if IS_ENABLED(CONFIG_IPV6)
8864                 off = si->off;
8865                 off -= offsetof(struct bpf_sock, src_ip6[0]);
8866                 *insn++ = BPF_LDX_MEM(
8867                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8868                         bpf_target_off(
8869                                 struct sock_common,
8870                                 skc_v6_rcv_saddr.s6_addr32[0],
8871                                 sizeof_field(struct sock_common,
8872                                              skc_v6_rcv_saddr.s6_addr32[0]),
8873                                 target_size) + off);
8874 #else
8875                 (void)off;
8876                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8877 #endif
8878                 break;
8879
8880         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8881 #if IS_ENABLED(CONFIG_IPV6)
8882                 off = si->off;
8883                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8884                 *insn++ = BPF_LDX_MEM(
8885                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8886                         bpf_target_off(struct sock_common,
8887                                        skc_v6_daddr.s6_addr32[0],
8888                                        sizeof_field(struct sock_common,
8889                                                     skc_v6_daddr.s6_addr32[0]),
8890                                        target_size) + off);
8891 #else
8892                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8893                 *target_size = 4;
8894 #endif
8895                 break;
8896
8897         case offsetof(struct bpf_sock, src_port):
8898                 *insn++ = BPF_LDX_MEM(
8899                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8900                         si->dst_reg, si->src_reg,
8901                         bpf_target_off(struct sock_common, skc_num,
8902                                        sizeof_field(struct sock_common,
8903                                                     skc_num),
8904                                        target_size));
8905                 break;
8906
8907         case offsetof(struct bpf_sock, dst_port):
8908                 *insn++ = BPF_LDX_MEM(
8909                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8910                         si->dst_reg, si->src_reg,
8911                         bpf_target_off(struct sock_common, skc_dport,
8912                                        sizeof_field(struct sock_common,
8913                                                     skc_dport),
8914                                        target_size));
8915                 break;
8916
8917         case offsetof(struct bpf_sock, state):
8918                 *insn++ = BPF_LDX_MEM(
8919                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8920                         si->dst_reg, si->src_reg,
8921                         bpf_target_off(struct sock_common, skc_state,
8922                                        sizeof_field(struct sock_common,
8923                                                     skc_state),
8924                                        target_size));
8925                 break;
8926         case offsetof(struct bpf_sock, rx_queue_mapping):
8927 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8928                 *insn++ = BPF_LDX_MEM(
8929                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8930                         si->dst_reg, si->src_reg,
8931                         bpf_target_off(struct sock, sk_rx_queue_mapping,
8932                                        sizeof_field(struct sock,
8933                                                     sk_rx_queue_mapping),
8934                                        target_size));
8935                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8936                                       1);
8937                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8938 #else
8939                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8940                 *target_size = 2;
8941 #endif
8942                 break;
8943         }
8944
8945         return insn - insn_buf;
8946 }
8947
8948 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8949                                          const struct bpf_insn *si,
8950                                          struct bpf_insn *insn_buf,
8951                                          struct bpf_prog *prog, u32 *target_size)
8952 {
8953         struct bpf_insn *insn = insn_buf;
8954
8955         switch (si->off) {
8956         case offsetof(struct __sk_buff, ifindex):
8957                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8958                                       si->dst_reg, si->src_reg,
8959                                       offsetof(struct sk_buff, dev));
8960                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8961                                       bpf_target_off(struct net_device, ifindex, 4,
8962                                                      target_size));
8963                 break;
8964         default:
8965                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
8966                                               target_size);
8967         }
8968
8969         return insn - insn_buf;
8970 }
8971
8972 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
8973                                   const struct bpf_insn *si,
8974                                   struct bpf_insn *insn_buf,
8975                                   struct bpf_prog *prog, u32 *target_size)
8976 {
8977         struct bpf_insn *insn = insn_buf;
8978
8979         switch (si->off) {
8980         case offsetof(struct xdp_md, data):
8981                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
8982                                       si->dst_reg, si->src_reg,
8983                                       offsetof(struct xdp_buff, data));
8984                 break;
8985         case offsetof(struct xdp_md, data_meta):
8986                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
8987                                       si->dst_reg, si->src_reg,
8988                                       offsetof(struct xdp_buff, data_meta));
8989                 break;
8990         case offsetof(struct xdp_md, data_end):
8991                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
8992                                       si->dst_reg, si->src_reg,
8993                                       offsetof(struct xdp_buff, data_end));
8994                 break;
8995         case offsetof(struct xdp_md, ingress_ifindex):
8996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
8997                                       si->dst_reg, si->src_reg,
8998                                       offsetof(struct xdp_buff, rxq));
8999                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9000                                       si->dst_reg, si->dst_reg,
9001                                       offsetof(struct xdp_rxq_info, dev));
9002                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9003                                       offsetof(struct net_device, ifindex));
9004                 break;
9005         case offsetof(struct xdp_md, rx_queue_index):
9006                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9007                                       si->dst_reg, si->src_reg,
9008                                       offsetof(struct xdp_buff, rxq));
9009                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9010                                       offsetof(struct xdp_rxq_info,
9011                                                queue_index));
9012                 break;
9013         case offsetof(struct xdp_md, egress_ifindex):
9014                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9015                                       si->dst_reg, si->src_reg,
9016                                       offsetof(struct xdp_buff, txq));
9017                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9018                                       si->dst_reg, si->dst_reg,
9019                                       offsetof(struct xdp_txq_info, dev));
9020                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9021                                       offsetof(struct net_device, ifindex));
9022                 break;
9023         }
9024
9025         return insn - insn_buf;
9026 }
9027
9028 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9029  * context Structure, F is Field in context structure that contains a pointer
9030  * to Nested Structure of type NS that has the field NF.
9031  *
9032  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9033  * sure that SIZE is not greater than actual size of S.F.NF.
9034  *
9035  * If offset OFF is provided, the load happens from that offset relative to
9036  * offset of NF.
9037  */
9038 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9039         do {                                                                   \
9040                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9041                                       si->src_reg, offsetof(S, F));            \
9042                 *insn++ = BPF_LDX_MEM(                                         \
9043                         SIZE, si->dst_reg, si->dst_reg,                        \
9044                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9045                                        target_size)                            \
9046                                 + OFF);                                        \
9047         } while (0)
9048
9049 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9050         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9051                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9052
9053 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9054  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9055  *
9056  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9057  * "register" since two registers available in convert_ctx_access are not
9058  * enough: we can't override neither SRC, since it contains value to store, nor
9059  * DST since it contains pointer to context that may be used by later
9060  * instructions. But we need a temporary place to save pointer to nested
9061  * structure whose field we want to store to.
9062  */
9063 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9064         do {                                                                   \
9065                 int tmp_reg = BPF_REG_9;                                       \
9066                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9067                         --tmp_reg;                                             \
9068                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9069                         --tmp_reg;                                             \
9070                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9071                                       offsetof(S, TF));                        \
9072                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9073                                       si->dst_reg, offsetof(S, F));            \
9074                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9075                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9076                                        target_size)                            \
9077                                 + OFF);                                        \
9078                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9079                                       offsetof(S, TF));                        \
9080         } while (0)
9081
9082 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9083                                                       TF)                      \
9084         do {                                                                   \
9085                 if (type == BPF_WRITE) {                                       \
9086                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9087                                                          OFF, TF);             \
9088                 } else {                                                       \
9089                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9090                                 S, NS, F, NF, SIZE, OFF);  \
9091                 }                                                              \
9092         } while (0)
9093
9094 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9095         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9096                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9097
9098 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9099                                         const struct bpf_insn *si,
9100                                         struct bpf_insn *insn_buf,
9101                                         struct bpf_prog *prog, u32 *target_size)
9102 {
9103         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9104         struct bpf_insn *insn = insn_buf;
9105
9106         switch (si->off) {
9107         case offsetof(struct bpf_sock_addr, user_family):
9108                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9109                                             struct sockaddr, uaddr, sa_family);
9110                 break;
9111
9112         case offsetof(struct bpf_sock_addr, user_ip4):
9113                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9114                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9115                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9116                 break;
9117
9118         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9119                 off = si->off;
9120                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9121                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9122                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9123                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9124                         tmp_reg);
9125                 break;
9126
9127         case offsetof(struct bpf_sock_addr, user_port):
9128                 /* To get port we need to know sa_family first and then treat
9129                  * sockaddr as either sockaddr_in or sockaddr_in6.
9130                  * Though we can simplify since port field has same offset and
9131                  * size in both structures.
9132                  * Here we check this invariant and use just one of the
9133                  * structures if it's true.
9134                  */
9135                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9136                              offsetof(struct sockaddr_in6, sin6_port));
9137                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9138                              sizeof_field(struct sockaddr_in6, sin6_port));
9139                 /* Account for sin6_port being smaller than user_port. */
9140                 port_size = min(port_size, BPF_LDST_BYTES(si));
9141                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9142                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9143                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9144                 break;
9145
9146         case offsetof(struct bpf_sock_addr, family):
9147                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9148                                             struct sock, sk, sk_family);
9149                 break;
9150
9151         case offsetof(struct bpf_sock_addr, type):
9152                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9153                                             struct sock, sk, sk_type);
9154                 break;
9155
9156         case offsetof(struct bpf_sock_addr, protocol):
9157                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9158                                             struct sock, sk, sk_protocol);
9159                 break;
9160
9161         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9162                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9163                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9164                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9165                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9166                 break;
9167
9168         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9169                                 msg_src_ip6[3]):
9170                 off = si->off;
9171                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9172                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9173                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9174                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9175                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9176                 break;
9177         case offsetof(struct bpf_sock_addr, sk):
9178                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9179                                       si->dst_reg, si->src_reg,
9180                                       offsetof(struct bpf_sock_addr_kern, sk));
9181                 break;
9182         }
9183
9184         return insn - insn_buf;
9185 }
9186
9187 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9188                                        const struct bpf_insn *si,
9189                                        struct bpf_insn *insn_buf,
9190                                        struct bpf_prog *prog,
9191                                        u32 *target_size)
9192 {
9193         struct bpf_insn *insn = insn_buf;
9194         int off;
9195
9196 /* Helper macro for adding read access to tcp_sock or sock fields. */
9197 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9198         do {                                                                  \
9199                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
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                 if (si->dst_reg == si->src_reg) {                             \
9207                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9208                                           offsetof(struct bpf_sock_ops_kern,  \
9209                                           temp));                             \
9210                         fullsock_reg = reg;                                   \
9211                         jmp += 2;                                             \
9212                 }                                                             \
9213                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9214                                                 struct bpf_sock_ops_kern,     \
9215                                                 is_fullsock),                 \
9216                                       fullsock_reg, si->src_reg,              \
9217                                       offsetof(struct bpf_sock_ops_kern,      \
9218                                                is_fullsock));                 \
9219                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9220                 if (si->dst_reg == si->src_reg)                               \
9221                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9222                                       offsetof(struct bpf_sock_ops_kern,      \
9223                                       temp));                                 \
9224                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9225                                                 struct bpf_sock_ops_kern, sk),\
9226                                       si->dst_reg, si->src_reg,               \
9227                                       offsetof(struct bpf_sock_ops_kern, sk));\
9228                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9229                                                        OBJ_FIELD),            \
9230                                       si->dst_reg, si->dst_reg,               \
9231                                       offsetof(OBJ, OBJ_FIELD));              \
9232                 if (si->dst_reg == si->src_reg) {                             \
9233                         *insn++ = BPF_JMP_A(1);                               \
9234                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9235                                       offsetof(struct bpf_sock_ops_kern,      \
9236                                       temp));                                 \
9237                 }                                                             \
9238         } while (0)
9239
9240 #define SOCK_OPS_GET_SK()                                                             \
9241         do {                                                                  \
9242                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9243                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9244                         reg--;                                                \
9245                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9246                         reg--;                                                \
9247                 if (si->dst_reg == si->src_reg) {                             \
9248                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9249                                           offsetof(struct bpf_sock_ops_kern,  \
9250                                           temp));                             \
9251                         fullsock_reg = reg;                                   \
9252                         jmp += 2;                                             \
9253                 }                                                             \
9254                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9255                                                 struct bpf_sock_ops_kern,     \
9256                                                 is_fullsock),                 \
9257                                       fullsock_reg, si->src_reg,              \
9258                                       offsetof(struct bpf_sock_ops_kern,      \
9259                                                is_fullsock));                 \
9260                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9261                 if (si->dst_reg == si->src_reg)                               \
9262                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9263                                       offsetof(struct bpf_sock_ops_kern,      \
9264                                       temp));                                 \
9265                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9266                                                 struct bpf_sock_ops_kern, sk),\
9267                                       si->dst_reg, si->src_reg,               \
9268                                       offsetof(struct bpf_sock_ops_kern, sk));\
9269                 if (si->dst_reg == si->src_reg) {                             \
9270                         *insn++ = BPF_JMP_A(1);                               \
9271                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9272                                       offsetof(struct bpf_sock_ops_kern,      \
9273                                       temp));                                 \
9274                 }                                                             \
9275         } while (0)
9276
9277 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9278                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9279
9280 /* Helper macro for adding write access to tcp_sock or sock fields.
9281  * The macro is called with two registers, dst_reg which contains a pointer
9282  * to ctx (context) and src_reg which contains the value that should be
9283  * stored. However, we need an additional register since we cannot overwrite
9284  * dst_reg because it may be used later in the program.
9285  * Instead we "borrow" one of the other register. We first save its value
9286  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9287  * it at the end of the macro.
9288  */
9289 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9290         do {                                                                  \
9291                 int reg = BPF_REG_9;                                          \
9292                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9293                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9294                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9295                         reg--;                                                \
9296                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9297                         reg--;                                                \
9298                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9299                                       offsetof(struct bpf_sock_ops_kern,      \
9300                                                temp));                        \
9301                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9302                                                 struct bpf_sock_ops_kern,     \
9303                                                 is_fullsock),                 \
9304                                       reg, si->dst_reg,                       \
9305                                       offsetof(struct bpf_sock_ops_kern,      \
9306                                                is_fullsock));                 \
9307                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9308                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9309                                                 struct bpf_sock_ops_kern, sk),\
9310                                       reg, si->dst_reg,                       \
9311                                       offsetof(struct bpf_sock_ops_kern, sk));\
9312                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9313                                       reg, si->src_reg,                       \
9314                                       offsetof(OBJ, OBJ_FIELD));              \
9315                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9316                                       offsetof(struct bpf_sock_ops_kern,      \
9317                                                temp));                        \
9318         } while (0)
9319
9320 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9321         do {                                                                  \
9322                 if (TYPE == BPF_WRITE)                                        \
9323                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9324                 else                                                          \
9325                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9326         } while (0)
9327
9328         if (insn > insn_buf)
9329                 return insn - insn_buf;
9330
9331         switch (si->off) {
9332         case offsetof(struct bpf_sock_ops, op):
9333                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9334                                                        op),
9335                                       si->dst_reg, si->src_reg,
9336                                       offsetof(struct bpf_sock_ops_kern, op));
9337                 break;
9338
9339         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9340              offsetof(struct bpf_sock_ops, replylong[3]):
9341                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9342                              sizeof_field(struct bpf_sock_ops_kern, reply));
9343                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9344                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9345                 off = si->off;
9346                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9347                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9348                 if (type == BPF_WRITE)
9349                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9350                                               off);
9351                 else
9352                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9353                                               off);
9354                 break;
9355
9356         case offsetof(struct bpf_sock_ops, family):
9357                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9358
9359                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9360                                               struct bpf_sock_ops_kern, sk),
9361                                       si->dst_reg, si->src_reg,
9362                                       offsetof(struct bpf_sock_ops_kern, sk));
9363                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9364                                       offsetof(struct sock_common, skc_family));
9365                 break;
9366
9367         case offsetof(struct bpf_sock_ops, remote_ip4):
9368                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9369
9370                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9371                                                 struct bpf_sock_ops_kern, sk),
9372                                       si->dst_reg, si->src_reg,
9373                                       offsetof(struct bpf_sock_ops_kern, sk));
9374                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9375                                       offsetof(struct sock_common, skc_daddr));
9376                 break;
9377
9378         case offsetof(struct bpf_sock_ops, local_ip4):
9379                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9380                                           skc_rcv_saddr) != 4);
9381
9382                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9383                                               struct bpf_sock_ops_kern, sk),
9384                                       si->dst_reg, si->src_reg,
9385                                       offsetof(struct bpf_sock_ops_kern, sk));
9386                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9387                                       offsetof(struct sock_common,
9388                                                skc_rcv_saddr));
9389                 break;
9390
9391         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9392              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9393 #if IS_ENABLED(CONFIG_IPV6)
9394                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9395                                           skc_v6_daddr.s6_addr32[0]) != 4);
9396
9397                 off = si->off;
9398                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9399                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9400                                                 struct bpf_sock_ops_kern, sk),
9401                                       si->dst_reg, si->src_reg,
9402                                       offsetof(struct bpf_sock_ops_kern, sk));
9403                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9404                                       offsetof(struct sock_common,
9405                                                skc_v6_daddr.s6_addr32[0]) +
9406                                       off);
9407 #else
9408                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9409 #endif
9410                 break;
9411
9412         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9413              offsetof(struct bpf_sock_ops, local_ip6[3]):
9414 #if IS_ENABLED(CONFIG_IPV6)
9415                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9416                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9417
9418                 off = si->off;
9419                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9420                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9421                                                 struct bpf_sock_ops_kern, sk),
9422                                       si->dst_reg, si->src_reg,
9423                                       offsetof(struct bpf_sock_ops_kern, sk));
9424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9425                                       offsetof(struct sock_common,
9426                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9427                                       off);
9428 #else
9429                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9430 #endif
9431                 break;
9432
9433         case offsetof(struct bpf_sock_ops, remote_port):
9434                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9435
9436                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9437                                                 struct bpf_sock_ops_kern, sk),
9438                                       si->dst_reg, si->src_reg,
9439                                       offsetof(struct bpf_sock_ops_kern, sk));
9440                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9441                                       offsetof(struct sock_common, skc_dport));
9442 #ifndef __BIG_ENDIAN_BITFIELD
9443                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9444 #endif
9445                 break;
9446
9447         case offsetof(struct bpf_sock_ops, local_port):
9448                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9449
9450                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9451                                                 struct bpf_sock_ops_kern, sk),
9452                                       si->dst_reg, si->src_reg,
9453                                       offsetof(struct bpf_sock_ops_kern, sk));
9454                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9455                                       offsetof(struct sock_common, skc_num));
9456                 break;
9457
9458         case offsetof(struct bpf_sock_ops, is_fullsock):
9459                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9460                                                 struct bpf_sock_ops_kern,
9461                                                 is_fullsock),
9462                                       si->dst_reg, si->src_reg,
9463                                       offsetof(struct bpf_sock_ops_kern,
9464                                                is_fullsock));
9465                 break;
9466
9467         case offsetof(struct bpf_sock_ops, state):
9468                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9469
9470                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9471                                                 struct bpf_sock_ops_kern, sk),
9472                                       si->dst_reg, si->src_reg,
9473                                       offsetof(struct bpf_sock_ops_kern, sk));
9474                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9475                                       offsetof(struct sock_common, skc_state));
9476                 break;
9477
9478         case offsetof(struct bpf_sock_ops, rtt_min):
9479                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9480                              sizeof(struct minmax));
9481                 BUILD_BUG_ON(sizeof(struct minmax) <
9482                              sizeof(struct minmax_sample));
9483
9484                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9485                                                 struct bpf_sock_ops_kern, sk),
9486                                       si->dst_reg, si->src_reg,
9487                                       offsetof(struct bpf_sock_ops_kern, sk));
9488                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9489                                       offsetof(struct tcp_sock, rtt_min) +
9490                                       sizeof_field(struct minmax_sample, t));
9491                 break;
9492
9493         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9494                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9495                                    struct tcp_sock);
9496                 break;
9497
9498         case offsetof(struct bpf_sock_ops, sk_txhash):
9499                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9500                                           struct sock, type);
9501                 break;
9502         case offsetof(struct bpf_sock_ops, snd_cwnd):
9503                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9504                 break;
9505         case offsetof(struct bpf_sock_ops, srtt_us):
9506                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9507                 break;
9508         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9509                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9510                 break;
9511         case offsetof(struct bpf_sock_ops, rcv_nxt):
9512                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9513                 break;
9514         case offsetof(struct bpf_sock_ops, snd_nxt):
9515                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9516                 break;
9517         case offsetof(struct bpf_sock_ops, snd_una):
9518                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9519                 break;
9520         case offsetof(struct bpf_sock_ops, mss_cache):
9521                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9522                 break;
9523         case offsetof(struct bpf_sock_ops, ecn_flags):
9524                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9525                 break;
9526         case offsetof(struct bpf_sock_ops, rate_delivered):
9527                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9528                 break;
9529         case offsetof(struct bpf_sock_ops, rate_interval_us):
9530                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9531                 break;
9532         case offsetof(struct bpf_sock_ops, packets_out):
9533                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9534                 break;
9535         case offsetof(struct bpf_sock_ops, retrans_out):
9536                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9537                 break;
9538         case offsetof(struct bpf_sock_ops, total_retrans):
9539                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9540                 break;
9541         case offsetof(struct bpf_sock_ops, segs_in):
9542                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9543                 break;
9544         case offsetof(struct bpf_sock_ops, data_segs_in):
9545                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9546                 break;
9547         case offsetof(struct bpf_sock_ops, segs_out):
9548                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9549                 break;
9550         case offsetof(struct bpf_sock_ops, data_segs_out):
9551                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9552                 break;
9553         case offsetof(struct bpf_sock_ops, lost_out):
9554                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9555                 break;
9556         case offsetof(struct bpf_sock_ops, sacked_out):
9557                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9558                 break;
9559         case offsetof(struct bpf_sock_ops, bytes_received):
9560                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9561                 break;
9562         case offsetof(struct bpf_sock_ops, bytes_acked):
9563                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9564                 break;
9565         case offsetof(struct bpf_sock_ops, sk):
9566                 SOCK_OPS_GET_SK();
9567                 break;
9568         case offsetof(struct bpf_sock_ops, skb_data_end):
9569                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9570                                                        skb_data_end),
9571                                       si->dst_reg, si->src_reg,
9572                                       offsetof(struct bpf_sock_ops_kern,
9573                                                skb_data_end));
9574                 break;
9575         case offsetof(struct bpf_sock_ops, skb_data):
9576                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9577                                                        skb),
9578                                       si->dst_reg, si->src_reg,
9579                                       offsetof(struct bpf_sock_ops_kern,
9580                                                skb));
9581                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9582                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9583                                       si->dst_reg, si->dst_reg,
9584                                       offsetof(struct sk_buff, data));
9585                 break;
9586         case offsetof(struct bpf_sock_ops, skb_len):
9587                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9588                                                        skb),
9589                                       si->dst_reg, si->src_reg,
9590                                       offsetof(struct bpf_sock_ops_kern,
9591                                                skb));
9592                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9593                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9594                                       si->dst_reg, si->dst_reg,
9595                                       offsetof(struct sk_buff, len));
9596                 break;
9597         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9598                 off = offsetof(struct sk_buff, cb);
9599                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9600                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9601                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9602                                                        skb),
9603                                       si->dst_reg, si->src_reg,
9604                                       offsetof(struct bpf_sock_ops_kern,
9605                                                skb));
9606                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9607                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9608                                                        tcp_flags),
9609                                       si->dst_reg, si->dst_reg, off);
9610                 break;
9611         }
9612         return insn - insn_buf;
9613 }
9614
9615 /* data_end = skb->data + skb_headlen() */
9616 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9617                                                     struct bpf_insn *insn)
9618 {
9619         /* si->dst_reg = skb->data */
9620         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9621                               si->dst_reg, si->src_reg,
9622                               offsetof(struct sk_buff, data));
9623         /* AX = skb->len */
9624         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9625                               BPF_REG_AX, si->src_reg,
9626                               offsetof(struct sk_buff, len));
9627         /* si->dst_reg = skb->data + skb->len */
9628         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9629         /* AX = skb->data_len */
9630         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9631                               BPF_REG_AX, si->src_reg,
9632                               offsetof(struct sk_buff, data_len));
9633         /* si->dst_reg = skb->data + skb->len - skb->data_len */
9634         *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9635
9636         return insn;
9637 }
9638
9639 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9640                                      const struct bpf_insn *si,
9641                                      struct bpf_insn *insn_buf,
9642                                      struct bpf_prog *prog, u32 *target_size)
9643 {
9644         struct bpf_insn *insn = insn_buf;
9645
9646         switch (si->off) {
9647         case offsetof(struct __sk_buff, data_end):
9648                 insn = bpf_convert_data_end_access(si, insn);
9649                 break;
9650         default:
9651                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9652                                               target_size);
9653         }
9654
9655         return insn - insn_buf;
9656 }
9657
9658 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9659                                      const struct bpf_insn *si,
9660                                      struct bpf_insn *insn_buf,
9661                                      struct bpf_prog *prog, u32 *target_size)
9662 {
9663         struct bpf_insn *insn = insn_buf;
9664 #if IS_ENABLED(CONFIG_IPV6)
9665         int off;
9666 #endif
9667
9668         /* convert ctx uses the fact sg element is first in struct */
9669         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9670
9671         switch (si->off) {
9672         case offsetof(struct sk_msg_md, data):
9673                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9674                                       si->dst_reg, si->src_reg,
9675                                       offsetof(struct sk_msg, data));
9676                 break;
9677         case offsetof(struct sk_msg_md, data_end):
9678                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9679                                       si->dst_reg, si->src_reg,
9680                                       offsetof(struct sk_msg, data_end));
9681                 break;
9682         case offsetof(struct sk_msg_md, family):
9683                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9684
9685                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9686                                               struct sk_msg, sk),
9687                                       si->dst_reg, si->src_reg,
9688                                       offsetof(struct sk_msg, sk));
9689                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9690                                       offsetof(struct sock_common, skc_family));
9691                 break;
9692
9693         case offsetof(struct sk_msg_md, remote_ip4):
9694                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9695
9696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9697                                                 struct sk_msg, sk),
9698                                       si->dst_reg, si->src_reg,
9699                                       offsetof(struct sk_msg, sk));
9700                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9701                                       offsetof(struct sock_common, skc_daddr));
9702                 break;
9703
9704         case offsetof(struct sk_msg_md, local_ip4):
9705                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9706                                           skc_rcv_saddr) != 4);
9707
9708                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9709                                               struct sk_msg, sk),
9710                                       si->dst_reg, si->src_reg,
9711                                       offsetof(struct sk_msg, sk));
9712                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9713                                       offsetof(struct sock_common,
9714                                                skc_rcv_saddr));
9715                 break;
9716
9717         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9718              offsetof(struct sk_msg_md, remote_ip6[3]):
9719 #if IS_ENABLED(CONFIG_IPV6)
9720                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9721                                           skc_v6_daddr.s6_addr32[0]) != 4);
9722
9723                 off = si->off;
9724                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9725                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9726                                                 struct sk_msg, sk),
9727                                       si->dst_reg, si->src_reg,
9728                                       offsetof(struct sk_msg, sk));
9729                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9730                                       offsetof(struct sock_common,
9731                                                skc_v6_daddr.s6_addr32[0]) +
9732                                       off);
9733 #else
9734                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9735 #endif
9736                 break;
9737
9738         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9739              offsetof(struct sk_msg_md, local_ip6[3]):
9740 #if IS_ENABLED(CONFIG_IPV6)
9741                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9742                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9743
9744                 off = si->off;
9745                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9746                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9747                                                 struct sk_msg, sk),
9748                                       si->dst_reg, si->src_reg,
9749                                       offsetof(struct sk_msg, sk));
9750                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9751                                       offsetof(struct sock_common,
9752                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9753                                       off);
9754 #else
9755                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9756 #endif
9757                 break;
9758
9759         case offsetof(struct sk_msg_md, remote_port):
9760                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9761
9762                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9763                                                 struct sk_msg, sk),
9764                                       si->dst_reg, si->src_reg,
9765                                       offsetof(struct sk_msg, sk));
9766                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9767                                       offsetof(struct sock_common, skc_dport));
9768 #ifndef __BIG_ENDIAN_BITFIELD
9769                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9770 #endif
9771                 break;
9772
9773         case offsetof(struct sk_msg_md, local_port):
9774                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9775
9776                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9777                                                 struct sk_msg, sk),
9778                                       si->dst_reg, si->src_reg,
9779                                       offsetof(struct sk_msg, sk));
9780                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9781                                       offsetof(struct sock_common, skc_num));
9782                 break;
9783
9784         case offsetof(struct sk_msg_md, size):
9785                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9786                                       si->dst_reg, si->src_reg,
9787                                       offsetof(struct sk_msg_sg, size));
9788                 break;
9789
9790         case offsetof(struct sk_msg_md, sk):
9791                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9792                                       si->dst_reg, si->src_reg,
9793                                       offsetof(struct sk_msg, sk));
9794                 break;
9795         }
9796
9797         return insn - insn_buf;
9798 }
9799
9800 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9801         .get_func_proto         = sk_filter_func_proto,
9802         .is_valid_access        = sk_filter_is_valid_access,
9803         .convert_ctx_access     = bpf_convert_ctx_access,
9804         .gen_ld_abs             = bpf_gen_ld_abs,
9805 };
9806
9807 const struct bpf_prog_ops sk_filter_prog_ops = {
9808         .test_run               = bpf_prog_test_run_skb,
9809 };
9810
9811 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9812         .get_func_proto         = tc_cls_act_func_proto,
9813         .is_valid_access        = tc_cls_act_is_valid_access,
9814         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
9815         .gen_prologue           = tc_cls_act_prologue,
9816         .gen_ld_abs             = bpf_gen_ld_abs,
9817         .check_kfunc_call       = bpf_prog_test_check_kfunc_call,
9818 };
9819
9820 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9821         .test_run               = bpf_prog_test_run_skb,
9822 };
9823
9824 const struct bpf_verifier_ops xdp_verifier_ops = {
9825         .get_func_proto         = xdp_func_proto,
9826         .is_valid_access        = xdp_is_valid_access,
9827         .convert_ctx_access     = xdp_convert_ctx_access,
9828         .gen_prologue           = bpf_noop_prologue,
9829 };
9830
9831 const struct bpf_prog_ops xdp_prog_ops = {
9832         .test_run               = bpf_prog_test_run_xdp,
9833 };
9834
9835 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9836         .get_func_proto         = cg_skb_func_proto,
9837         .is_valid_access        = cg_skb_is_valid_access,
9838         .convert_ctx_access     = bpf_convert_ctx_access,
9839 };
9840
9841 const struct bpf_prog_ops cg_skb_prog_ops = {
9842         .test_run               = bpf_prog_test_run_skb,
9843 };
9844
9845 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9846         .get_func_proto         = lwt_in_func_proto,
9847         .is_valid_access        = lwt_is_valid_access,
9848         .convert_ctx_access     = bpf_convert_ctx_access,
9849 };
9850
9851 const struct bpf_prog_ops lwt_in_prog_ops = {
9852         .test_run               = bpf_prog_test_run_skb,
9853 };
9854
9855 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9856         .get_func_proto         = lwt_out_func_proto,
9857         .is_valid_access        = lwt_is_valid_access,
9858         .convert_ctx_access     = bpf_convert_ctx_access,
9859 };
9860
9861 const struct bpf_prog_ops lwt_out_prog_ops = {
9862         .test_run               = bpf_prog_test_run_skb,
9863 };
9864
9865 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9866         .get_func_proto         = lwt_xmit_func_proto,
9867         .is_valid_access        = lwt_is_valid_access,
9868         .convert_ctx_access     = bpf_convert_ctx_access,
9869         .gen_prologue           = tc_cls_act_prologue,
9870 };
9871
9872 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9873         .test_run               = bpf_prog_test_run_skb,
9874 };
9875
9876 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9877         .get_func_proto         = lwt_seg6local_func_proto,
9878         .is_valid_access        = lwt_is_valid_access,
9879         .convert_ctx_access     = bpf_convert_ctx_access,
9880 };
9881
9882 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9883         .test_run               = bpf_prog_test_run_skb,
9884 };
9885
9886 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9887         .get_func_proto         = sock_filter_func_proto,
9888         .is_valid_access        = sock_filter_is_valid_access,
9889         .convert_ctx_access     = bpf_sock_convert_ctx_access,
9890 };
9891
9892 const struct bpf_prog_ops cg_sock_prog_ops = {
9893 };
9894
9895 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9896         .get_func_proto         = sock_addr_func_proto,
9897         .is_valid_access        = sock_addr_is_valid_access,
9898         .convert_ctx_access     = sock_addr_convert_ctx_access,
9899 };
9900
9901 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9902 };
9903
9904 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9905         .get_func_proto         = sock_ops_func_proto,
9906         .is_valid_access        = sock_ops_is_valid_access,
9907         .convert_ctx_access     = sock_ops_convert_ctx_access,
9908 };
9909
9910 const struct bpf_prog_ops sock_ops_prog_ops = {
9911 };
9912
9913 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9914         .get_func_proto         = sk_skb_func_proto,
9915         .is_valid_access        = sk_skb_is_valid_access,
9916         .convert_ctx_access     = sk_skb_convert_ctx_access,
9917         .gen_prologue           = sk_skb_prologue,
9918 };
9919
9920 const struct bpf_prog_ops sk_skb_prog_ops = {
9921 };
9922
9923 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9924         .get_func_proto         = sk_msg_func_proto,
9925         .is_valid_access        = sk_msg_is_valid_access,
9926         .convert_ctx_access     = sk_msg_convert_ctx_access,
9927         .gen_prologue           = bpf_noop_prologue,
9928 };
9929
9930 const struct bpf_prog_ops sk_msg_prog_ops = {
9931 };
9932
9933 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9934         .get_func_proto         = flow_dissector_func_proto,
9935         .is_valid_access        = flow_dissector_is_valid_access,
9936         .convert_ctx_access     = flow_dissector_convert_ctx_access,
9937 };
9938
9939 const struct bpf_prog_ops flow_dissector_prog_ops = {
9940         .test_run               = bpf_prog_test_run_flow_dissector,
9941 };
9942
9943 int sk_detach_filter(struct sock *sk)
9944 {
9945         int ret = -ENOENT;
9946         struct sk_filter *filter;
9947
9948         if (sock_flag(sk, SOCK_FILTER_LOCKED))
9949                 return -EPERM;
9950
9951         filter = rcu_dereference_protected(sk->sk_filter,
9952                                            lockdep_sock_is_held(sk));
9953         if (filter) {
9954                 RCU_INIT_POINTER(sk->sk_filter, NULL);
9955                 sk_filter_uncharge(sk, filter);
9956                 ret = 0;
9957         }
9958
9959         return ret;
9960 }
9961 EXPORT_SYMBOL_GPL(sk_detach_filter);
9962
9963 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
9964                   unsigned int len)
9965 {
9966         struct sock_fprog_kern *fprog;
9967         struct sk_filter *filter;
9968         int ret = 0;
9969
9970         lock_sock(sk);
9971         filter = rcu_dereference_protected(sk->sk_filter,
9972                                            lockdep_sock_is_held(sk));
9973         if (!filter)
9974                 goto out;
9975
9976         /* We're copying the filter that has been originally attached,
9977          * so no conversion/decode needed anymore. eBPF programs that
9978          * have no original program cannot be dumped through this.
9979          */
9980         ret = -EACCES;
9981         fprog = filter->prog->orig_prog;
9982         if (!fprog)
9983                 goto out;
9984
9985         ret = fprog->len;
9986         if (!len)
9987                 /* User space only enquires number of filter blocks. */
9988                 goto out;
9989
9990         ret = -EINVAL;
9991         if (len < fprog->len)
9992                 goto out;
9993
9994         ret = -EFAULT;
9995         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
9996                 goto out;
9997
9998         /* Instead of bytes, the API requests to return the number
9999          * of filter blocks.
10000          */
10001         ret = fprog->len;
10002 out:
10003         release_sock(sk);
10004         return ret;
10005 }
10006
10007 #ifdef CONFIG_INET
10008 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10009                                     struct sock_reuseport *reuse,
10010                                     struct sock *sk, struct sk_buff *skb,
10011                                     u32 hash)
10012 {
10013         reuse_kern->skb = skb;
10014         reuse_kern->sk = sk;
10015         reuse_kern->selected_sk = NULL;
10016         reuse_kern->data_end = skb->data + skb_headlen(skb);
10017         reuse_kern->hash = hash;
10018         reuse_kern->reuseport_id = reuse->reuseport_id;
10019         reuse_kern->bind_inany = reuse->bind_inany;
10020 }
10021
10022 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10023                                   struct bpf_prog *prog, struct sk_buff *skb,
10024                                   u32 hash)
10025 {
10026         struct sk_reuseport_kern reuse_kern;
10027         enum sk_action action;
10028
10029         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
10030         action = BPF_PROG_RUN(prog, &reuse_kern);
10031
10032         if (action == SK_PASS)
10033                 return reuse_kern.selected_sk;
10034         else
10035                 return ERR_PTR(-ECONNREFUSED);
10036 }
10037
10038 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10039            struct bpf_map *, map, void *, key, u32, flags)
10040 {
10041         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10042         struct sock_reuseport *reuse;
10043         struct sock *selected_sk;
10044
10045         selected_sk = map->ops->map_lookup_elem(map, key);
10046         if (!selected_sk)
10047                 return -ENOENT;
10048
10049         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10050         if (!reuse) {
10051                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10052                 if (sk_is_refcounted(selected_sk))
10053                         sock_put(selected_sk);
10054
10055                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10056                  * The only (!reuse) case here is - the sk has already been
10057                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10058                  *
10059                  * Other maps (e.g. sock_map) do not provide this guarantee and
10060                  * the sk may never be in the reuseport group to begin with.
10061                  */
10062                 return is_sockarray ? -ENOENT : -EINVAL;
10063         }
10064
10065         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10066                 struct sock *sk = reuse_kern->sk;
10067
10068                 if (sk->sk_protocol != selected_sk->sk_protocol)
10069                         return -EPROTOTYPE;
10070                 else if (sk->sk_family != selected_sk->sk_family)
10071                         return -EAFNOSUPPORT;
10072
10073                 /* Catch all. Likely bound to a different sockaddr. */
10074                 return -EBADFD;
10075         }
10076
10077         reuse_kern->selected_sk = selected_sk;
10078
10079         return 0;
10080 }
10081
10082 static const struct bpf_func_proto sk_select_reuseport_proto = {
10083         .func           = sk_select_reuseport,
10084         .gpl_only       = false,
10085         .ret_type       = RET_INTEGER,
10086         .arg1_type      = ARG_PTR_TO_CTX,
10087         .arg2_type      = ARG_CONST_MAP_PTR,
10088         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10089         .arg4_type      = ARG_ANYTHING,
10090 };
10091
10092 BPF_CALL_4(sk_reuseport_load_bytes,
10093            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10094            void *, to, u32, len)
10095 {
10096         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10097 }
10098
10099 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10100         .func           = sk_reuseport_load_bytes,
10101         .gpl_only       = false,
10102         .ret_type       = RET_INTEGER,
10103         .arg1_type      = ARG_PTR_TO_CTX,
10104         .arg2_type      = ARG_ANYTHING,
10105         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10106         .arg4_type      = ARG_CONST_SIZE,
10107 };
10108
10109 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10110            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10111            void *, to, u32, len, u32, start_header)
10112 {
10113         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10114                                                len, start_header);
10115 }
10116
10117 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10118         .func           = sk_reuseport_load_bytes_relative,
10119         .gpl_only       = false,
10120         .ret_type       = RET_INTEGER,
10121         .arg1_type      = ARG_PTR_TO_CTX,
10122         .arg2_type      = ARG_ANYTHING,
10123         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10124         .arg4_type      = ARG_CONST_SIZE,
10125         .arg5_type      = ARG_ANYTHING,
10126 };
10127
10128 static const struct bpf_func_proto *
10129 sk_reuseport_func_proto(enum bpf_func_id func_id,
10130                         const struct bpf_prog *prog)
10131 {
10132         switch (func_id) {
10133         case BPF_FUNC_sk_select_reuseport:
10134                 return &sk_select_reuseport_proto;
10135         case BPF_FUNC_skb_load_bytes:
10136                 return &sk_reuseport_load_bytes_proto;
10137         case BPF_FUNC_skb_load_bytes_relative:
10138                 return &sk_reuseport_load_bytes_relative_proto;
10139         default:
10140                 return bpf_base_func_proto(func_id);
10141         }
10142 }
10143
10144 static bool
10145 sk_reuseport_is_valid_access(int off, int size,
10146                              enum bpf_access_type type,
10147                              const struct bpf_prog *prog,
10148                              struct bpf_insn_access_aux *info)
10149 {
10150         const u32 size_default = sizeof(__u32);
10151
10152         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10153             off % size || type != BPF_READ)
10154                 return false;
10155
10156         switch (off) {
10157         case offsetof(struct sk_reuseport_md, data):
10158                 info->reg_type = PTR_TO_PACKET;
10159                 return size == sizeof(__u64);
10160
10161         case offsetof(struct sk_reuseport_md, data_end):
10162                 info->reg_type = PTR_TO_PACKET_END;
10163                 return size == sizeof(__u64);
10164
10165         case offsetof(struct sk_reuseport_md, hash):
10166                 return size == size_default;
10167
10168         /* Fields that allow narrowing */
10169         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10170                 if (size < sizeof_field(struct sk_buff, protocol))
10171                         return false;
10172                 fallthrough;
10173         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10174         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10175         case bpf_ctx_range(struct sk_reuseport_md, len):
10176                 bpf_ctx_record_field_size(info, size_default);
10177                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10178
10179         default:
10180                 return false;
10181         }
10182 }
10183
10184 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10185         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10186                               si->dst_reg, si->src_reg,                 \
10187                               bpf_target_off(struct sk_reuseport_kern, F, \
10188                                              sizeof_field(struct sk_reuseport_kern, F), \
10189                                              target_size));             \
10190         })
10191
10192 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10193         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10194                                     struct sk_buff,                     \
10195                                     skb,                                \
10196                                     SKB_FIELD)
10197
10198 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10199         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10200                                     struct sock,                        \
10201                                     sk,                                 \
10202                                     SK_FIELD)
10203
10204 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10205                                            const struct bpf_insn *si,
10206                                            struct bpf_insn *insn_buf,
10207                                            struct bpf_prog *prog,
10208                                            u32 *target_size)
10209 {
10210         struct bpf_insn *insn = insn_buf;
10211
10212         switch (si->off) {
10213         case offsetof(struct sk_reuseport_md, data):
10214                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10215                 break;
10216
10217         case offsetof(struct sk_reuseport_md, len):
10218                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10219                 break;
10220
10221         case offsetof(struct sk_reuseport_md, eth_protocol):
10222                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10223                 break;
10224
10225         case offsetof(struct sk_reuseport_md, ip_protocol):
10226                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10227                 break;
10228
10229         case offsetof(struct sk_reuseport_md, data_end):
10230                 SK_REUSEPORT_LOAD_FIELD(data_end);
10231                 break;
10232
10233         case offsetof(struct sk_reuseport_md, hash):
10234                 SK_REUSEPORT_LOAD_FIELD(hash);
10235                 break;
10236
10237         case offsetof(struct sk_reuseport_md, bind_inany):
10238                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10239                 break;
10240         }
10241
10242         return insn - insn_buf;
10243 }
10244
10245 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10246         .get_func_proto         = sk_reuseport_func_proto,
10247         .is_valid_access        = sk_reuseport_is_valid_access,
10248         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10249 };
10250
10251 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10252 };
10253
10254 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10255 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10256
10257 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10258            struct sock *, sk, u64, flags)
10259 {
10260         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10261                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10262                 return -EINVAL;
10263         if (unlikely(sk && sk_is_refcounted(sk)))
10264                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10265         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10266                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10267
10268         /* Check if socket is suitable for packet L3/L4 protocol */
10269         if (sk && sk->sk_protocol != ctx->protocol)
10270                 return -EPROTOTYPE;
10271         if (sk && sk->sk_family != ctx->family &&
10272             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10273                 return -EAFNOSUPPORT;
10274
10275         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10276                 return -EEXIST;
10277
10278         /* Select socket as lookup result */
10279         ctx->selected_sk = sk;
10280         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10281         return 0;
10282 }
10283
10284 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10285         .func           = bpf_sk_lookup_assign,
10286         .gpl_only       = false,
10287         .ret_type       = RET_INTEGER,
10288         .arg1_type      = ARG_PTR_TO_CTX,
10289         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10290         .arg3_type      = ARG_ANYTHING,
10291 };
10292
10293 static const struct bpf_func_proto *
10294 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10295 {
10296         switch (func_id) {
10297         case BPF_FUNC_perf_event_output:
10298                 return &bpf_event_output_data_proto;
10299         case BPF_FUNC_sk_assign:
10300                 return &bpf_sk_lookup_assign_proto;
10301         case BPF_FUNC_sk_release:
10302                 return &bpf_sk_release_proto;
10303         default:
10304                 return bpf_sk_base_func_proto(func_id);
10305         }
10306 }
10307
10308 static bool sk_lookup_is_valid_access(int off, int size,
10309                                       enum bpf_access_type type,
10310                                       const struct bpf_prog *prog,
10311                                       struct bpf_insn_access_aux *info)
10312 {
10313         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10314                 return false;
10315         if (off % size != 0)
10316                 return false;
10317         if (type != BPF_READ)
10318                 return false;
10319
10320         switch (off) {
10321         case offsetof(struct bpf_sk_lookup, sk):
10322                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10323                 return size == sizeof(__u64);
10324
10325         case bpf_ctx_range(struct bpf_sk_lookup, family):
10326         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10327         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10328         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10329         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10330         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10331         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10332         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10333                 bpf_ctx_record_field_size(info, sizeof(__u32));
10334                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10335
10336         default:
10337                 return false;
10338         }
10339 }
10340
10341 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10342                                         const struct bpf_insn *si,
10343                                         struct bpf_insn *insn_buf,
10344                                         struct bpf_prog *prog,
10345                                         u32 *target_size)
10346 {
10347         struct bpf_insn *insn = insn_buf;
10348
10349         switch (si->off) {
10350         case offsetof(struct bpf_sk_lookup, sk):
10351                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10352                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10353                 break;
10354
10355         case offsetof(struct bpf_sk_lookup, family):
10356                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10357                                       bpf_target_off(struct bpf_sk_lookup_kern,
10358                                                      family, 2, target_size));
10359                 break;
10360
10361         case offsetof(struct bpf_sk_lookup, protocol):
10362                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10363                                       bpf_target_off(struct bpf_sk_lookup_kern,
10364                                                      protocol, 2, target_size));
10365                 break;
10366
10367         case offsetof(struct bpf_sk_lookup, remote_ip4):
10368                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10369                                       bpf_target_off(struct bpf_sk_lookup_kern,
10370                                                      v4.saddr, 4, target_size));
10371                 break;
10372
10373         case offsetof(struct bpf_sk_lookup, local_ip4):
10374                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10375                                       bpf_target_off(struct bpf_sk_lookup_kern,
10376                                                      v4.daddr, 4, target_size));
10377                 break;
10378
10379         case bpf_ctx_range_till(struct bpf_sk_lookup,
10380                                 remote_ip6[0], remote_ip6[3]): {
10381 #if IS_ENABLED(CONFIG_IPV6)
10382                 int off = si->off;
10383
10384                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10385                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10386                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10387                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10388                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10389                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10390 #else
10391                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10392 #endif
10393                 break;
10394         }
10395         case bpf_ctx_range_till(struct bpf_sk_lookup,
10396                                 local_ip6[0], local_ip6[3]): {
10397 #if IS_ENABLED(CONFIG_IPV6)
10398                 int off = si->off;
10399
10400                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10401                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10402                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10403                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10404                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10405                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10406 #else
10407                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10408 #endif
10409                 break;
10410         }
10411         case offsetof(struct bpf_sk_lookup, remote_port):
10412                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10413                                       bpf_target_off(struct bpf_sk_lookup_kern,
10414                                                      sport, 2, target_size));
10415                 break;
10416
10417         case offsetof(struct bpf_sk_lookup, local_port):
10418                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10419                                       bpf_target_off(struct bpf_sk_lookup_kern,
10420                                                      dport, 2, target_size));
10421                 break;
10422         }
10423
10424         return insn - insn_buf;
10425 }
10426
10427 const struct bpf_prog_ops sk_lookup_prog_ops = {
10428         .test_run = bpf_prog_test_run_sk_lookup,
10429 };
10430
10431 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10432         .get_func_proto         = sk_lookup_func_proto,
10433         .is_valid_access        = sk_lookup_is_valid_access,
10434         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10435 };
10436
10437 #endif /* CONFIG_INET */
10438
10439 DEFINE_BPF_DISPATCHER(xdp)
10440
10441 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10442 {
10443         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10444 }
10445
10446 #ifdef CONFIG_DEBUG_INFO_BTF
10447 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10448 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10449 BTF_SOCK_TYPE_xxx
10450 #undef BTF_SOCK_TYPE
10451 #else
10452 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10453 #endif
10454
10455 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10456 {
10457         /* tcp6_sock type is not generated in dwarf and hence btf,
10458          * trigger an explicit type generation here.
10459          */
10460         BTF_TYPE_EMIT(struct tcp6_sock);
10461         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10462             sk->sk_family == AF_INET6)
10463                 return (unsigned long)sk;
10464
10465         return (unsigned long)NULL;
10466 }
10467
10468 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10469         .func                   = bpf_skc_to_tcp6_sock,
10470         .gpl_only               = false,
10471         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10472         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10473         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10474 };
10475
10476 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10477 {
10478         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10479                 return (unsigned long)sk;
10480
10481         return (unsigned long)NULL;
10482 }
10483
10484 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10485         .func                   = bpf_skc_to_tcp_sock,
10486         .gpl_only               = false,
10487         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10488         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10489         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10490 };
10491
10492 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10493 {
10494         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10495          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10496          */
10497         BTF_TYPE_EMIT(struct inet_timewait_sock);
10498         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10499
10500 #ifdef CONFIG_INET
10501         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10502                 return (unsigned long)sk;
10503 #endif
10504
10505 #if IS_BUILTIN(CONFIG_IPV6)
10506         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10507                 return (unsigned long)sk;
10508 #endif
10509
10510         return (unsigned long)NULL;
10511 }
10512
10513 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10514         .func                   = bpf_skc_to_tcp_timewait_sock,
10515         .gpl_only               = false,
10516         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10517         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10518         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10519 };
10520
10521 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10522 {
10523 #ifdef CONFIG_INET
10524         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10525                 return (unsigned long)sk;
10526 #endif
10527
10528 #if IS_BUILTIN(CONFIG_IPV6)
10529         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10530                 return (unsigned long)sk;
10531 #endif
10532
10533         return (unsigned long)NULL;
10534 }
10535
10536 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10537         .func                   = bpf_skc_to_tcp_request_sock,
10538         .gpl_only               = false,
10539         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10540         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10541         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10542 };
10543
10544 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10545 {
10546         /* udp6_sock type is not generated in dwarf and hence btf,
10547          * trigger an explicit type generation here.
10548          */
10549         BTF_TYPE_EMIT(struct udp6_sock);
10550         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10551             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10552                 return (unsigned long)sk;
10553
10554         return (unsigned long)NULL;
10555 }
10556
10557 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10558         .func                   = bpf_skc_to_udp6_sock,
10559         .gpl_only               = false,
10560         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10561         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10562         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10563 };
10564
10565 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10566 {
10567         return (unsigned long)sock_from_file(file);
10568 }
10569
10570 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10571 BTF_ID(struct, socket)
10572 BTF_ID(struct, file)
10573
10574 const struct bpf_func_proto bpf_sock_from_file_proto = {
10575         .func           = bpf_sock_from_file,
10576         .gpl_only       = false,
10577         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10578         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10579         .arg1_type      = ARG_PTR_TO_BTF_ID,
10580         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10581 };
10582
10583 static const struct bpf_func_proto *
10584 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10585 {
10586         const struct bpf_func_proto *func;
10587
10588         switch (func_id) {
10589         case BPF_FUNC_skc_to_tcp6_sock:
10590                 func = &bpf_skc_to_tcp6_sock_proto;
10591                 break;
10592         case BPF_FUNC_skc_to_tcp_sock:
10593                 func = &bpf_skc_to_tcp_sock_proto;
10594                 break;
10595         case BPF_FUNC_skc_to_tcp_timewait_sock:
10596                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10597                 break;
10598         case BPF_FUNC_skc_to_tcp_request_sock:
10599                 func = &bpf_skc_to_tcp_request_sock_proto;
10600                 break;
10601         case BPF_FUNC_skc_to_udp6_sock:
10602                 func = &bpf_skc_to_udp6_sock_proto;
10603                 break;
10604         default:
10605                 return bpf_base_func_proto(func_id);
10606         }
10607
10608         if (!perfmon_capable())
10609                 return NULL;
10610
10611         return func;
10612 }