net/core/filter.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Linux Socket Filter - Kernel level socket filtering
   4  *
   5  * Based on the design of the Berkeley Packet Filter. The new
   6  * internal format has been designed by PLUMgrid:
   7  *
   8  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   9  *
  10  * Authors:
  11  *
  12  *      Jay Schulist <jschlst@samba.org>
  13  *      Alexei Starovoitov <ast@plumgrid.com>
  14  *      Daniel Borkmann <dborkman@redhat.com>
  15  *
  16  * Andi Kleen - Fix a few bad bugs and races.
  17  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  18  */
  19
  20 #include <linux/atomic.h>
  21 #include <linux/module.h>
  22 #include <linux/types.h>
  23 #include <linux/mm.h>
  24 #include <linux/fcntl.h>
  25 #include <linux/socket.h>
  26 #include <linux/sock_diag.h>
  27 #include <linux/in.h>
  28 #include <linux/inet.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_packet.h>
  31 #include <linux/if_arp.h>
  32 #include <linux/gfp.h>
  33 #include <net/inet_common.h>
  34 #include <net/ip.h>
  35 #include <net/protocol.h>
  36 #include <net/netlink.h>
  37 #include <linux/skbuff.h>
  38 #include <linux/skmsg.h>
  39 #include <net/sock.h>
  40 #include <net/flow_dissector.h>
  41 #include <linux/errno.h>
  42 #include <linux/timer.h>
  43 #include <linux/uaccess.h>
  44 #include <asm/unaligned.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         ret = skb_cow(skb, len_diff);
3245         if (unlikely(ret < 0))
3246                 return ret;
3247
3248         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3249         if (unlikely(ret < 0))
3250                 return ret;
3251
3252         if (skb_is_gso(skb)) {
3253                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3254
3255                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3256                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3257                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3258                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3259                 }
3260         }
3261
3262         skb->protocol = htons(ETH_P_IPV6);
3263         skb_clear_hash(skb);
3264
3265         return 0;
3266 }
3267
3268 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3269 {
3270         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3271         u32 off = skb_mac_header_len(skb);
3272         int ret;
3273
3274         ret = skb_unclone(skb, GFP_ATOMIC);
3275         if (unlikely(ret < 0))
3276                 return ret;
3277
3278         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3279         if (unlikely(ret < 0))
3280                 return ret;
3281
3282         if (skb_is_gso(skb)) {
3283                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3284
3285                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3286                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3287                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3288                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3289                 }
3290         }
3291
3292         skb->protocol = htons(ETH_P_IP);
3293         skb_clear_hash(skb);
3294
3295         return 0;
3296 }
3297
3298 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3299 {
3300         __be16 from_proto = skb->protocol;
3301
3302         if (from_proto == htons(ETH_P_IP) &&
3303               to_proto == htons(ETH_P_IPV6))
3304                 return bpf_skb_proto_4_to_6(skb);
3305
3306         if (from_proto == htons(ETH_P_IPV6) &&
3307               to_proto == htons(ETH_P_IP))
3308                 return bpf_skb_proto_6_to_4(skb);
3309
3310         return -ENOTSUPP;
3311 }
3312
3313 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3314            u64, flags)
3315 {
3316         int ret;
3317
3318         if (unlikely(flags))
3319                 return -EINVAL;
3320
3321         /* General idea is that this helper does the basic groundwork
3322          * needed for changing the protocol, and eBPF program fills the
3323          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3324          * and other helpers, rather than passing a raw buffer here.
3325          *
3326          * The rationale is to keep this minimal and without a need to
3327          * deal with raw packet data. F.e. even if we would pass buffers
3328          * here, the program still needs to call the bpf_lX_csum_replace()
3329          * helpers anyway. Plus, this way we keep also separation of
3330          * concerns, since f.e. bpf_skb_store_bytes() should only take
3331          * care of stores.
3332          *
3333          * Currently, additional options and extension header space are
3334          * not supported, but flags register is reserved so we can adapt
3335          * that. For offloads, we mark packet as dodgy, so that headers
3336          * need to be verified first.
3337          */
3338         ret = bpf_skb_proto_xlat(skb, proto);
3339         bpf_compute_data_pointers(skb);
3340         return ret;
3341 }
3342
3343 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3344         .func           = bpf_skb_change_proto,
3345         .gpl_only       = false,
3346         .ret_type       = RET_INTEGER,
3347         .arg1_type      = ARG_PTR_TO_CTX,
3348         .arg2_type      = ARG_ANYTHING,
3349         .arg3_type      = ARG_ANYTHING,
3350 };
3351
3352 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3353 {
3354         /* We only allow a restricted subset to be changed for now. */
3355         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3356                      !skb_pkt_type_ok(pkt_type)))
3357                 return -EINVAL;
3358
3359         skb->pkt_type = pkt_type;
3360         return 0;
3361 }
3362
3363 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3364         .func           = bpf_skb_change_type,
3365         .gpl_only       = false,
3366         .ret_type       = RET_INTEGER,
3367         .arg1_type      = ARG_PTR_TO_CTX,
3368         .arg2_type      = ARG_ANYTHING,
3369 };
3370
3371 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3372 {
3373         switch (skb->protocol) {
3374         case htons(ETH_P_IP):
3375                 return sizeof(struct iphdr);
3376         case htons(ETH_P_IPV6):
3377                 return sizeof(struct ipv6hdr);
3378         default:
3379                 return ~0U;
3380         }
3381 }
3382
3383 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3384                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3385
3386 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3387                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3388                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3389                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3390                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3391                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3392                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3393
3394 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3395                             u64 flags)
3396 {
3397         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3398         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3399         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3400         unsigned int gso_type = SKB_GSO_DODGY;
3401         int ret;
3402
3403         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3404                 /* udp gso_size delineates datagrams, only allow if fixed */
3405                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3406                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3407                         return -ENOTSUPP;
3408         }
3409
3410         ret = skb_cow_head(skb, len_diff);
3411         if (unlikely(ret < 0))
3412                 return ret;
3413
3414         if (encap) {
3415                 if (skb->protocol != htons(ETH_P_IP) &&
3416                     skb->protocol != htons(ETH_P_IPV6))
3417                         return -ENOTSUPP;
3418
3419                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3420                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3421                         return -EINVAL;
3422
3423                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3424                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3425                         return -EINVAL;
3426
3427                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3428                     inner_mac_len < ETH_HLEN)
3429                         return -EINVAL;
3430
3431                 if (skb->encapsulation)
3432                         return -EALREADY;
3433
3434                 mac_len = skb->network_header - skb->mac_header;
3435                 inner_net = skb->network_header;
3436                 if (inner_mac_len > len_diff)
3437                         return -EINVAL;
3438                 inner_trans = skb->transport_header;
3439         }
3440
3441         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3442         if (unlikely(ret < 0))
3443                 return ret;
3444
3445         if (encap) {
3446                 skb->inner_mac_header = inner_net - inner_mac_len;
3447                 skb->inner_network_header = inner_net;
3448                 skb->inner_transport_header = inner_trans;
3449
3450                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3451                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3452                 else
3453                         skb_set_inner_protocol(skb, skb->protocol);
3454
3455                 skb->encapsulation = 1;
3456                 skb_set_network_header(skb, mac_len);
3457
3458                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3459                         gso_type |= SKB_GSO_UDP_TUNNEL;
3460                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3461                         gso_type |= SKB_GSO_GRE;
3462                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3463                         gso_type |= SKB_GSO_IPXIP6;
3464                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3465                         gso_type |= SKB_GSO_IPXIP4;
3466
3467                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3468                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3469                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3470                                         sizeof(struct ipv6hdr) :
3471                                         sizeof(struct iphdr);
3472
3473                         skb_set_transport_header(skb, mac_len + nh_len);
3474                 }
3475
3476                 /* Match skb->protocol to new outer l3 protocol */
3477                 if (skb->protocol == htons(ETH_P_IP) &&
3478                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3479                         skb->protocol = htons(ETH_P_IPV6);
3480                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3481                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3482                         skb->protocol = htons(ETH_P_IP);
3483         }
3484
3485         if (skb_is_gso(skb)) {
3486                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3487
3488                 /* Due to header grow, MSS needs to be downgraded. */
3489                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3490                         skb_decrease_gso_size(shinfo, len_diff);
3491
3492                 /* Header must be checked, and gso_segs recomputed. */
3493                 shinfo->gso_type |= gso_type;
3494                 shinfo->gso_segs = 0;
3495         }
3496
3497         return 0;
3498 }
3499
3500 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3501                               u64 flags)
3502 {
3503         int ret;
3504
3505         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3506                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3507                 return -EINVAL;
3508
3509         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3510                 /* udp gso_size delineates datagrams, only allow if fixed */
3511                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3512                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3513                         return -ENOTSUPP;
3514         }
3515
3516         ret = skb_unclone(skb, GFP_ATOMIC);
3517         if (unlikely(ret < 0))
3518                 return ret;
3519
3520         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3521         if (unlikely(ret < 0))
3522                 return ret;
3523
3524         if (skb_is_gso(skb)) {
3525                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3526
3527                 /* Due to header shrink, MSS can be upgraded. */
3528                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3529                         skb_increase_gso_size(shinfo, len_diff);
3530
3531                 /* Header must be checked, and gso_segs recomputed. */
3532                 shinfo->gso_type |= SKB_GSO_DODGY;
3533                 shinfo->gso_segs = 0;
3534         }
3535
3536         return 0;
3537 }
3538
3539 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3540
3541 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3542            u32, mode, u64, flags)
3543 {
3544         u32 len_diff_abs = abs(len_diff);
3545         bool shrink = len_diff < 0;
3546         int ret = 0;
3547
3548         if (unlikely(flags || mode))
3549                 return -EINVAL;
3550         if (unlikely(len_diff_abs > 0xfffU))
3551                 return -EFAULT;
3552
3553         if (!shrink) {
3554                 ret = skb_cow(skb, len_diff);
3555                 if (unlikely(ret < 0))
3556                         return ret;
3557                 __skb_push(skb, len_diff_abs);
3558                 memset(skb->data, 0, len_diff_abs);
3559         } else {
3560                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3561                         return -ENOMEM;
3562                 __skb_pull(skb, len_diff_abs);
3563         }
3564         if (tls_sw_has_ctx_rx(skb->sk)) {
3565                 struct strp_msg *rxm = strp_msg(skb);
3566
3567                 rxm->full_len += len_diff;
3568         }
3569         return ret;
3570 }
3571
3572 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3573         .func           = sk_skb_adjust_room,
3574         .gpl_only       = false,
3575         .ret_type       = RET_INTEGER,
3576         .arg1_type      = ARG_PTR_TO_CTX,
3577         .arg2_type      = ARG_ANYTHING,
3578         .arg3_type      = ARG_ANYTHING,
3579         .arg4_type      = ARG_ANYTHING,
3580 };
3581
3582 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3583            u32, mode, u64, flags)
3584 {
3585         u32 len_cur, len_diff_abs = abs(len_diff);
3586         u32 len_min = bpf_skb_net_base_len(skb);
3587         u32 len_max = BPF_SKB_MAX_LEN;
3588         __be16 proto = skb->protocol;
3589         bool shrink = len_diff < 0;
3590         u32 off;
3591         int ret;
3592
3593         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3594                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3595                 return -EINVAL;
3596         if (unlikely(len_diff_abs > 0xfffU))
3597                 return -EFAULT;
3598         if (unlikely(proto != htons(ETH_P_IP) &&
3599                      proto != htons(ETH_P_IPV6)))
3600                 return -ENOTSUPP;
3601
3602         off = skb_mac_header_len(skb);
3603         switch (mode) {
3604         case BPF_ADJ_ROOM_NET:
3605                 off += bpf_skb_net_base_len(skb);
3606                 break;
3607         case BPF_ADJ_ROOM_MAC:
3608                 break;
3609         default:
3610                 return -ENOTSUPP;
3611         }
3612
3613         len_cur = skb->len - skb_network_offset(skb);
3614         if ((shrink && (len_diff_abs >= len_cur ||
3615                         len_cur - len_diff_abs < len_min)) ||
3616             (!shrink && (skb->len + len_diff_abs > len_max &&
3617                          !skb_is_gso(skb))))
3618                 return -ENOTSUPP;
3619
3620         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3621                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3622         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3623                 __skb_reset_checksum_unnecessary(skb);
3624
3625         bpf_compute_data_pointers(skb);
3626         return ret;
3627 }
3628
3629 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3630         .func           = bpf_skb_adjust_room,
3631         .gpl_only       = false,
3632         .ret_type       = RET_INTEGER,
3633         .arg1_type      = ARG_PTR_TO_CTX,
3634         .arg2_type      = ARG_ANYTHING,
3635         .arg3_type      = ARG_ANYTHING,
3636         .arg4_type      = ARG_ANYTHING,
3637 };
3638
3639 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3640 {
3641         u32 min_len = skb_network_offset(skb);
3642
3643         if (skb_transport_header_was_set(skb))
3644                 min_len = skb_transport_offset(skb);
3645         if (skb->ip_summed == CHECKSUM_PARTIAL)
3646                 min_len = skb_checksum_start_offset(skb) +
3647                           skb->csum_offset + sizeof(__sum16);
3648         return min_len;
3649 }
3650
3651 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3652 {
3653         unsigned int old_len = skb->len;
3654         int ret;
3655
3656         ret = __skb_grow_rcsum(skb, new_len);
3657         if (!ret)
3658                 memset(skb->data + old_len, 0, new_len - old_len);
3659         return ret;
3660 }
3661
3662 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3663 {
3664         return __skb_trim_rcsum(skb, new_len);
3665 }
3666
3667 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3668                                         u64 flags)
3669 {
3670         u32 max_len = BPF_SKB_MAX_LEN;
3671         u32 min_len = __bpf_skb_min_len(skb);
3672         int ret;
3673
3674         if (unlikely(flags || new_len > max_len || new_len < min_len))
3675                 return -EINVAL;
3676         if (skb->encapsulation)
3677                 return -ENOTSUPP;
3678
3679         /* The basic idea of this helper is that it's performing the
3680          * needed work to either grow or trim an skb, and eBPF program
3681          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3682          * bpf_lX_csum_replace() and others rather than passing a raw
3683          * buffer here. This one is a slow path helper and intended
3684          * for replies with control messages.
3685          *
3686          * Like in bpf_skb_change_proto(), we want to keep this rather
3687          * minimal and without protocol specifics so that we are able
3688          * to separate concerns as in bpf_skb_store_bytes() should only
3689          * be the one responsible for writing buffers.
3690          *
3691          * It's really expected to be a slow path operation here for
3692          * control message replies, so we're implicitly linearizing,
3693          * uncloning and drop offloads from the skb by this.
3694          */
3695         ret = __bpf_try_make_writable(skb, skb->len);
3696         if (!ret) {
3697                 if (new_len > skb->len)
3698                         ret = bpf_skb_grow_rcsum(skb, new_len);
3699                 else if (new_len < skb->len)
3700                         ret = bpf_skb_trim_rcsum(skb, new_len);
3701                 if (!ret && skb_is_gso(skb))
3702                         skb_gso_reset(skb);
3703         }
3704         return ret;
3705 }
3706
3707 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3708            u64, flags)
3709 {
3710         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3711
3712         bpf_compute_data_pointers(skb);
3713         return ret;
3714 }
3715
3716 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3717         .func           = bpf_skb_change_tail,
3718         .gpl_only       = false,
3719         .ret_type       = RET_INTEGER,
3720         .arg1_type      = ARG_PTR_TO_CTX,
3721         .arg2_type      = ARG_ANYTHING,
3722         .arg3_type      = ARG_ANYTHING,
3723 };
3724
3725 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3726            u64, flags)
3727 {
3728         return __bpf_skb_change_tail(skb, new_len, flags);
3729 }
3730
3731 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3732         .func           = sk_skb_change_tail,
3733         .gpl_only       = false,
3734         .ret_type       = RET_INTEGER,
3735         .arg1_type      = ARG_PTR_TO_CTX,
3736         .arg2_type      = ARG_ANYTHING,
3737         .arg3_type      = ARG_ANYTHING,
3738 };
3739
3740 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3741                                         u64 flags)
3742 {
3743         u32 max_len = BPF_SKB_MAX_LEN;
3744         u32 new_len = skb->len + head_room;
3745         int ret;
3746
3747         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3748                      new_len < skb->len))
3749                 return -EINVAL;
3750
3751         ret = skb_cow(skb, head_room);
3752         if (likely(!ret)) {
3753                 /* Idea for this helper is that we currently only
3754                  * allow to expand on mac header. This means that
3755                  * skb->protocol network header, etc, stay as is.
3756                  * Compared to bpf_skb_change_tail(), we're more
3757                  * flexible due to not needing to linearize or
3758                  * reset GSO. Intention for this helper is to be
3759                  * used by an L3 skb that needs to push mac header
3760                  * for redirection into L2 device.
3761                  */
3762                 __skb_push(skb, head_room);
3763                 memset(skb->data, 0, head_room);
3764                 skb_reset_mac_header(skb);
3765                 skb_reset_mac_len(skb);
3766         }
3767
3768         return ret;
3769 }
3770
3771 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3772            u64, flags)
3773 {
3774         int ret = __bpf_skb_change_head(skb, head_room, flags);
3775
3776         bpf_compute_data_pointers(skb);
3777         return ret;
3778 }
3779
3780 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3781         .func           = bpf_skb_change_head,
3782         .gpl_only       = false,
3783         .ret_type       = RET_INTEGER,
3784         .arg1_type      = ARG_PTR_TO_CTX,
3785         .arg2_type      = ARG_ANYTHING,
3786         .arg3_type      = ARG_ANYTHING,
3787 };
3788
3789 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3790            u64, flags)
3791 {
3792         return __bpf_skb_change_head(skb, head_room, flags);
3793 }
3794
3795 static const struct bpf_func_proto sk_skb_change_head_proto = {
3796         .func           = sk_skb_change_head,
3797         .gpl_only       = false,
3798         .ret_type       = RET_INTEGER,
3799         .arg1_type      = ARG_PTR_TO_CTX,
3800         .arg2_type      = ARG_ANYTHING,
3801         .arg3_type      = ARG_ANYTHING,
3802 };
3803 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3804 {
3805         return xdp_data_meta_unsupported(xdp) ? 0 :
3806                xdp->data - xdp->data_meta;
3807 }
3808
3809 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3810 {
3811         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3812         unsigned long metalen = xdp_get_metalen(xdp);
3813         void *data_start = xdp_frame_end + metalen;
3814         void *data = xdp->data + offset;
3815
3816         if (unlikely(data < data_start ||
3817                      data > xdp->data_end - ETH_HLEN))
3818                 return -EINVAL;
3819
3820         if (metalen)
3821                 memmove(xdp->data_meta + offset,
3822                         xdp->data_meta, metalen);
3823         xdp->data_meta += offset;
3824         xdp->data = data;
3825
3826         return 0;
3827 }
3828
3829 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3830         .func           = bpf_xdp_adjust_head,
3831         .gpl_only       = false,
3832         .ret_type       = RET_INTEGER,
3833         .arg1_type      = ARG_PTR_TO_CTX,
3834         .arg2_type      = ARG_ANYTHING,
3835 };
3836
3837 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3838 {
3839         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
3840         void *data_end = xdp->data_end + offset;
3841
3842         /* Notice that xdp_data_hard_end have reserved some tailroom */
3843         if (unlikely(data_end > data_hard_end))
3844                 return -EINVAL;
3845
3846         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
3847         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
3848                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
3849                 return -EINVAL;
3850         }
3851
3852         if (unlikely(data_end < xdp->data + ETH_HLEN))
3853                 return -EINVAL;
3854
3855         /* Clear memory area on grow, can contain uninit kernel memory */
3856         if (offset > 0)
3857                 memset(xdp->data_end, 0, offset);
3858
3859         xdp->data_end = data_end;
3860
3861         return 0;
3862 }
3863
3864 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3865         .func           = bpf_xdp_adjust_tail,
3866         .gpl_only       = false,
3867         .ret_type       = RET_INTEGER,
3868         .arg1_type      = ARG_PTR_TO_CTX,
3869         .arg2_type      = ARG_ANYTHING,
3870 };
3871
3872 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3873 {
3874         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3875         void *meta = xdp->data_meta + offset;
3876         unsigned long metalen = xdp->data - meta;
3877
3878         if (xdp_data_meta_unsupported(xdp))
3879                 return -ENOTSUPP;
3880         if (unlikely(meta < xdp_frame_end ||
3881                      meta > xdp->data))
3882                 return -EINVAL;
3883         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3884                      (metalen > 32)))
3885                 return -EACCES;
3886
3887         xdp->data_meta = meta;
3888
3889         return 0;
3890 }
3891
3892 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3893         .func           = bpf_xdp_adjust_meta,
3894         .gpl_only       = false,
3895         .ret_type       = RET_INTEGER,
3896         .arg1_type      = ARG_PTR_TO_CTX,
3897         .arg2_type      = ARG_ANYTHING,
3898 };
3899
3900 /* XDP_REDIRECT works by a three-step process, implemented in the functions
3901  * below:
3902  *
3903  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
3904  *    of the redirect and store it (along with some other metadata) in a per-CPU
3905  *    struct bpf_redirect_info.
3906  *
3907  * 2. When the program returns the XDP_REDIRECT return code, the driver will
3908  *    call xdp_do_redirect() which will use the information in struct
3909  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
3910  *    bulk queue structure.
3911  *
3912  * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
3913  *    which will flush all the different bulk queues, thus completing the
3914  *    redirect.
3915  *
3916  * Pointers to the map entries will be kept around for this whole sequence of
3917  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
3918  * the core code; instead, the RCU protection relies on everything happening
3919  * inside a single NAPI poll sequence, which means it's between a pair of calls
3920  * to local_bh_disable()/local_bh_enable().
3921  *
3922  * The map entries are marked as __rcu and the map code makes sure to
3923  * dereference those pointers with rcu_dereference_check() in a way that works
3924  * for both sections that to hold an rcu_read_lock() and sections that are
3925  * called from NAPI without a separate rcu_read_lock(). The code below does not
3926  * use RCU annotations, but relies on those in the map code.
3927  */
3928 void xdp_do_flush(void)
3929 {
3930         __dev_flush();
3931         __cpu_map_flush();
3932         __xsk_map_flush();
3933 }
3934 EXPORT_SYMBOL_GPL(xdp_do_flush);
3935
3936 void bpf_clear_redirect_map(struct bpf_map *map)
3937 {
3938         struct bpf_redirect_info *ri;
3939         int cpu;
3940
3941         for_each_possible_cpu(cpu) {
3942                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3943                 /* Avoid polluting remote cacheline due to writes if
3944                  * not needed. Once we pass this test, we need the
3945                  * cmpxchg() to make sure it hasn't been changed in
3946                  * the meantime by remote CPU.
3947                  */
3948                 if (unlikely(READ_ONCE(ri->map) == map))
3949                         cmpxchg(&ri->map, map, NULL);
3950         }
3951 }
3952
3953 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3954                     struct bpf_prog *xdp_prog)
3955 {
3956         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3957         enum bpf_map_type map_type = ri->map_type;
3958         void *fwd = ri->tgt_value;
3959         u32 map_id = ri->map_id;
3960         struct bpf_map *map;
3961         int err;
3962
3963         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
3964         ri->map_type = BPF_MAP_TYPE_UNSPEC;
3965
3966         switch (map_type) {
3967         case BPF_MAP_TYPE_DEVMAP:
3968                 fallthrough;
3969         case BPF_MAP_TYPE_DEVMAP_HASH:
3970                 map = READ_ONCE(ri->map);
3971                 if (unlikely(map)) {
3972                         WRITE_ONCE(ri->map, NULL);
3973                         err = dev_map_enqueue_multi(xdp, dev, map,
3974                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
3975                 } else {
3976                         err = dev_map_enqueue(fwd, xdp, dev);
3977                 }
3978                 break;
3979         case BPF_MAP_TYPE_CPUMAP:
3980                 err = cpu_map_enqueue(fwd, xdp, dev);
3981                 break;
3982         case BPF_MAP_TYPE_XSKMAP:
3983                 err = __xsk_map_redirect(fwd, xdp);
3984                 break;
3985         case BPF_MAP_TYPE_UNSPEC:
3986                 if (map_id == INT_MAX) {
3987                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
3988                         if (unlikely(!fwd)) {
3989                                 err = -EINVAL;
3990                                 break;
3991                         }
3992                         err = dev_xdp_enqueue(fwd, xdp, dev);
3993                         break;
3994                 }
3995                 fallthrough;
3996         default:
3997                 err = -EBADRQC;
3998         }
3999
4000         if (unlikely(err))
4001                 goto err;
4002
4003         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4004         return 0;
4005 err:
4006         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4007         return err;
4008 }
4009 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4010
4011 static int xdp_do_generic_redirect_map(struct net_device *dev,
4012                                        struct sk_buff *skb,
4013                                        struct xdp_buff *xdp,
4014                                        struct bpf_prog *xdp_prog,
4015                                        void *fwd,
4016                                        enum bpf_map_type map_type, u32 map_id)
4017 {
4018         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4019         struct bpf_map *map;
4020         int err;
4021
4022         switch (map_type) {
4023         case BPF_MAP_TYPE_DEVMAP:
4024                 fallthrough;
4025         case BPF_MAP_TYPE_DEVMAP_HASH:
4026                 map = READ_ONCE(ri->map);
4027                 if (unlikely(map)) {
4028                         WRITE_ONCE(ri->map, NULL);
4029                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4030                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4031                 } else {
4032                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4033                 }
4034                 if (unlikely(err))
4035                         goto err;
4036                 break;
4037         case BPF_MAP_TYPE_XSKMAP:
4038                 err = xsk_generic_rcv(fwd, xdp);
4039                 if (err)
4040                         goto err;
4041                 consume_skb(skb);
4042                 break;
4043         default:
4044                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
4045                 err = -EBADRQC;
4046                 goto err;
4047         }
4048
4049         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4050         return 0;
4051 err:
4052         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4053         return err;
4054 }
4055
4056 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4057                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4058 {
4059         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4060         enum bpf_map_type map_type = ri->map_type;
4061         void *fwd = ri->tgt_value;
4062         u32 map_id = ri->map_id;
4063         int err;
4064
4065         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4066         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4067
4068         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4069                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4070                 if (unlikely(!fwd)) {
4071                         err = -EINVAL;
4072                         goto err;
4073                 }
4074
4075                 err = xdp_ok_fwd_dev(fwd, skb->len);
4076                 if (unlikely(err))
4077                         goto err;
4078
4079                 skb->dev = fwd;
4080                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4081                 generic_xdp_tx(skb, xdp_prog);
4082                 return 0;
4083         }
4084
4085         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4086 err:
4087         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4088         return err;
4089 }
4090
4091 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4092 {
4093         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4094
4095         if (unlikely(flags))
4096                 return XDP_ABORTED;
4097
4098         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4099          * by map_idr) is used for ifindex based XDP redirect.
4100          */
4101         ri->tgt_index = ifindex;
4102         ri->map_id = INT_MAX;
4103         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4104
4105         return XDP_REDIRECT;
4106 }
4107
4108 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4109         .func           = bpf_xdp_redirect,
4110         .gpl_only       = false,
4111         .ret_type       = RET_INTEGER,
4112         .arg1_type      = ARG_ANYTHING,
4113         .arg2_type      = ARG_ANYTHING,
4114 };
4115
4116 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4117            u64, flags)
4118 {
4119         return map->ops->map_redirect(map, ifindex, flags);
4120 }
4121
4122 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4123         .func           = bpf_xdp_redirect_map,
4124         .gpl_only       = false,
4125         .ret_type       = RET_INTEGER,
4126         .arg1_type      = ARG_CONST_MAP_PTR,
4127         .arg2_type      = ARG_ANYTHING,
4128         .arg3_type      = ARG_ANYTHING,
4129 };
4130
4131 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4132                                   unsigned long off, unsigned long len)
4133 {
4134         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4135
4136         if (unlikely(!ptr))
4137                 return len;
4138         if (ptr != dst_buff)
4139                 memcpy(dst_buff, ptr, len);
4140
4141         return 0;
4142 }
4143
4144 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4145            u64, flags, void *, meta, u64, meta_size)
4146 {
4147         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4148
4149         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4150                 return -EINVAL;
4151         if (unlikely(!skb || skb_size > skb->len))
4152                 return -EFAULT;
4153
4154         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4155                                 bpf_skb_copy);
4156 }
4157
4158 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4159         .func           = bpf_skb_event_output,
4160         .gpl_only       = true,
4161         .ret_type       = RET_INTEGER,
4162         .arg1_type      = ARG_PTR_TO_CTX,
4163         .arg2_type      = ARG_CONST_MAP_PTR,
4164         .arg3_type      = ARG_ANYTHING,
4165         .arg4_type      = ARG_PTR_TO_MEM,
4166         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4167 };
4168
4169 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4170
4171 const struct bpf_func_proto bpf_skb_output_proto = {
4172         .func           = bpf_skb_event_output,
4173         .gpl_only       = true,
4174         .ret_type       = RET_INTEGER,
4175         .arg1_type      = ARG_PTR_TO_BTF_ID,
4176         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4177         .arg2_type      = ARG_CONST_MAP_PTR,
4178         .arg3_type      = ARG_ANYTHING,
4179         .arg4_type      = ARG_PTR_TO_MEM,
4180         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4181 };
4182
4183 static unsigned short bpf_tunnel_key_af(u64 flags)
4184 {
4185         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4186 }
4187
4188 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4189            u32, size, u64, flags)
4190 {
4191         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4192         u8 compat[sizeof(struct bpf_tunnel_key)];
4193         void *to_orig = to;
4194         int err;
4195
4196         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
4197                 err = -EINVAL;
4198                 goto err_clear;
4199         }
4200         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4201                 err = -EPROTO;
4202                 goto err_clear;
4203         }
4204         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4205                 err = -EINVAL;
4206                 switch (size) {
4207                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4208                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4209                         goto set_compat;
4210                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4211                         /* Fixup deprecated structure layouts here, so we have
4212                          * a common path later on.
4213                          */
4214                         if (ip_tunnel_info_af(info) != AF_INET)
4215                                 goto err_clear;
4216 set_compat:
4217                         to = (struct bpf_tunnel_key *)compat;
4218                         break;
4219                 default:
4220                         goto err_clear;
4221                 }
4222         }
4223
4224         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4225         to->tunnel_tos = info->key.tos;
4226         to->tunnel_ttl = info->key.ttl;
4227         to->tunnel_ext = 0;
4228
4229         if (flags & BPF_F_TUNINFO_IPV6) {
4230                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4231                        sizeof(to->remote_ipv6));
4232                 to->tunnel_label = be32_to_cpu(info->key.label);
4233         } else {
4234                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4235                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4236                 to->tunnel_label = 0;
4237         }
4238
4239         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4240                 memcpy(to_orig, to, size);
4241
4242         return 0;
4243 err_clear:
4244         memset(to_orig, 0, size);
4245         return err;
4246 }
4247
4248 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4249         .func           = bpf_skb_get_tunnel_key,
4250         .gpl_only       = false,
4251         .ret_type       = RET_INTEGER,
4252         .arg1_type      = ARG_PTR_TO_CTX,
4253         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4254         .arg3_type      = ARG_CONST_SIZE,
4255         .arg4_type      = ARG_ANYTHING,
4256 };
4257
4258 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4259 {
4260         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4261         int err;
4262
4263         if (unlikely(!info ||
4264                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4265                 err = -ENOENT;
4266                 goto err_clear;
4267         }
4268         if (unlikely(size < info->options_len)) {
4269                 err = -ENOMEM;
4270                 goto err_clear;
4271         }
4272
4273         ip_tunnel_info_opts_get(to, info);
4274         if (size > info->options_len)
4275                 memset(to + info->options_len, 0, size - info->options_len);
4276
4277         return info->options_len;
4278 err_clear:
4279         memset(to, 0, size);
4280         return err;
4281 }
4282
4283 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4284         .func           = bpf_skb_get_tunnel_opt,
4285         .gpl_only       = false,
4286         .ret_type       = RET_INTEGER,
4287         .arg1_type      = ARG_PTR_TO_CTX,
4288         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4289         .arg3_type      = ARG_CONST_SIZE,
4290 };
4291
4292 static struct metadata_dst __percpu *md_dst;
4293
4294 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4295            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4296 {
4297         struct metadata_dst *md = this_cpu_ptr(md_dst);
4298         u8 compat[sizeof(struct bpf_tunnel_key)];
4299         struct ip_tunnel_info *info;
4300
4301         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4302                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4303                 return -EINVAL;
4304         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4305                 switch (size) {
4306                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4307                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4308                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4309                         /* Fixup deprecated structure layouts here, so we have
4310                          * a common path later on.
4311                          */
4312                         memcpy(compat, from, size);
4313                         memset(compat + size, 0, sizeof(compat) - size);
4314                         from = (const struct bpf_tunnel_key *) compat;
4315                         break;
4316                 default:
4317                         return -EINVAL;
4318                 }
4319         }
4320         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4321                      from->tunnel_ext))
4322                 return -EINVAL;
4323
4324         skb_dst_drop(skb);
4325         dst_hold((struct dst_entry *) md);
4326         skb_dst_set(skb, (struct dst_entry *) md);
4327
4328         info = &md->u.tun_info;
4329         memset(info, 0, sizeof(*info));
4330         info->mode = IP_TUNNEL_INFO_TX;
4331
4332         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4333         if (flags & BPF_F_DONT_FRAGMENT)
4334                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4335         if (flags & BPF_F_ZERO_CSUM_TX)
4336                 info->key.tun_flags &= ~TUNNEL_CSUM;
4337         if (flags & BPF_F_SEQ_NUMBER)
4338                 info->key.tun_flags |= TUNNEL_SEQ;
4339
4340         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4341         info->key.tos = from->tunnel_tos;
4342         info->key.ttl = from->tunnel_ttl;
4343
4344         if (flags & BPF_F_TUNINFO_IPV6) {
4345                 info->mode |= IP_TUNNEL_INFO_IPV6;
4346                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4347                        sizeof(from->remote_ipv6));
4348                 info->key.label = cpu_to_be32(from->tunnel_label) &
4349                                   IPV6_FLOWLABEL_MASK;
4350         } else {
4351                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4352         }
4353
4354         return 0;
4355 }
4356
4357 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4358         .func           = bpf_skb_set_tunnel_key,
4359         .gpl_only       = false,
4360         .ret_type       = RET_INTEGER,
4361         .arg1_type      = ARG_PTR_TO_CTX,
4362         .arg2_type      = ARG_PTR_TO_MEM,
4363         .arg3_type      = ARG_CONST_SIZE,
4364         .arg4_type      = ARG_ANYTHING,
4365 };
4366
4367 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4368            const u8 *, from, u32, size)
4369 {
4370         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4371         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4372
4373         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4374                 return -EINVAL;
4375         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4376                 return -ENOMEM;
4377
4378         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4379
4380         return 0;
4381 }
4382
4383 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4384         .func           = bpf_skb_set_tunnel_opt,
4385         .gpl_only       = false,
4386         .ret_type       = RET_INTEGER,
4387         .arg1_type      = ARG_PTR_TO_CTX,
4388         .arg2_type      = ARG_PTR_TO_MEM,
4389         .arg3_type      = ARG_CONST_SIZE,
4390 };
4391
4392 static const struct bpf_func_proto *
4393 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4394 {
4395         if (!md_dst) {
4396                 struct metadata_dst __percpu *tmp;
4397
4398                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4399                                                 METADATA_IP_TUNNEL,
4400                                                 GFP_KERNEL);
4401                 if (!tmp)
4402                         return NULL;
4403                 if (cmpxchg(&md_dst, NULL, tmp))
4404                         metadata_dst_free_percpu(tmp);
4405         }
4406
4407         switch (which) {
4408         case BPF_FUNC_skb_set_tunnel_key:
4409                 return &bpf_skb_set_tunnel_key_proto;
4410         case BPF_FUNC_skb_set_tunnel_opt:
4411                 return &bpf_skb_set_tunnel_opt_proto;
4412         default:
4413                 return NULL;
4414         }
4415 }
4416
4417 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4418            u32, idx)
4419 {
4420         struct bpf_array *array = container_of(map, struct bpf_array, map);
4421         struct cgroup *cgrp;
4422         struct sock *sk;
4423
4424         sk = skb_to_full_sk(skb);
4425         if (!sk || !sk_fullsock(sk))
4426                 return -ENOENT;
4427         if (unlikely(idx >= array->map.max_entries))
4428                 return -E2BIG;
4429
4430         cgrp = READ_ONCE(array->ptrs[idx]);
4431         if (unlikely(!cgrp))
4432                 return -EAGAIN;
4433
4434         return sk_under_cgroup_hierarchy(sk, cgrp);
4435 }
4436
4437 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4438         .func           = bpf_skb_under_cgroup,
4439         .gpl_only       = false,
4440         .ret_type       = RET_INTEGER,
4441         .arg1_type      = ARG_PTR_TO_CTX,
4442         .arg2_type      = ARG_CONST_MAP_PTR,
4443         .arg3_type      = ARG_ANYTHING,
4444 };
4445
4446 #ifdef CONFIG_SOCK_CGROUP_DATA
4447 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4448 {
4449         struct cgroup *cgrp;
4450
4451         sk = sk_to_full_sk(sk);
4452         if (!sk || !sk_fullsock(sk))
4453                 return 0;
4454
4455         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4456         return cgroup_id(cgrp);
4457 }
4458
4459 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4460 {
4461         return __bpf_sk_cgroup_id(skb->sk);
4462 }
4463
4464 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4465         .func           = bpf_skb_cgroup_id,
4466         .gpl_only       = false,
4467         .ret_type       = RET_INTEGER,
4468         .arg1_type      = ARG_PTR_TO_CTX,
4469 };
4470
4471 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4472                                               int ancestor_level)
4473 {
4474         struct cgroup *ancestor;
4475         struct cgroup *cgrp;
4476
4477         sk = sk_to_full_sk(sk);
4478         if (!sk || !sk_fullsock(sk))
4479                 return 0;
4480
4481         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4482         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4483         if (!ancestor)
4484                 return 0;
4485
4486         return cgroup_id(ancestor);
4487 }
4488
4489 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4490            ancestor_level)
4491 {
4492         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4493 }
4494
4495 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4496         .func           = bpf_skb_ancestor_cgroup_id,
4497         .gpl_only       = false,
4498         .ret_type       = RET_INTEGER,
4499         .arg1_type      = ARG_PTR_TO_CTX,
4500         .arg2_type      = ARG_ANYTHING,
4501 };
4502
4503 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4504 {
4505         return __bpf_sk_cgroup_id(sk);
4506 }
4507
4508 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4509         .func           = bpf_sk_cgroup_id,
4510         .gpl_only       = false,
4511         .ret_type       = RET_INTEGER,
4512         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4513 };
4514
4515 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4516 {
4517         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4518 }
4519
4520 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4521         .func           = bpf_sk_ancestor_cgroup_id,
4522         .gpl_only       = false,
4523         .ret_type       = RET_INTEGER,
4524         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4525         .arg2_type      = ARG_ANYTHING,
4526 };
4527 #endif
4528
4529 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
4530                                   unsigned long off, unsigned long len)
4531 {
4532         memcpy(dst_buff, src_buff + off, len);
4533         return 0;
4534 }
4535
4536 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4537            u64, flags, void *, meta, u64, meta_size)
4538 {
4539         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4540
4541         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4542                 return -EINVAL;
4543         if (unlikely(!xdp ||
4544                      xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
4545                 return -EFAULT;
4546
4547         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
4548                                 xdp_size, bpf_xdp_copy);
4549 }
4550
4551 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4552         .func           = bpf_xdp_event_output,
4553         .gpl_only       = true,
4554         .ret_type       = RET_INTEGER,
4555         .arg1_type      = ARG_PTR_TO_CTX,
4556         .arg2_type      = ARG_CONST_MAP_PTR,
4557         .arg3_type      = ARG_ANYTHING,
4558         .arg4_type      = ARG_PTR_TO_MEM,
4559         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4560 };
4561
4562 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4563
4564 const struct bpf_func_proto bpf_xdp_output_proto = {
4565         .func           = bpf_xdp_event_output,
4566         .gpl_only       = true,
4567         .ret_type       = RET_INTEGER,
4568         .arg1_type      = ARG_PTR_TO_BTF_ID,
4569         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4570         .arg2_type      = ARG_CONST_MAP_PTR,
4571         .arg3_type      = ARG_ANYTHING,
4572         .arg4_type      = ARG_PTR_TO_MEM,
4573         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4574 };
4575
4576 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4577 {
4578         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4579 }
4580
4581 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4582         .func           = bpf_get_socket_cookie,
4583         .gpl_only       = false,
4584         .ret_type       = RET_INTEGER,
4585         .arg1_type      = ARG_PTR_TO_CTX,
4586 };
4587
4588 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4589 {
4590         return __sock_gen_cookie(ctx->sk);
4591 }
4592
4593 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4594         .func           = bpf_get_socket_cookie_sock_addr,
4595         .gpl_only       = false,
4596         .ret_type       = RET_INTEGER,
4597         .arg1_type      = ARG_PTR_TO_CTX,
4598 };
4599
4600 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4601 {
4602         return __sock_gen_cookie(ctx);
4603 }
4604
4605 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4606         .func           = bpf_get_socket_cookie_sock,
4607         .gpl_only       = false,
4608         .ret_type       = RET_INTEGER,
4609         .arg1_type      = ARG_PTR_TO_CTX,
4610 };
4611
4612 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4613 {
4614         return sk ? sock_gen_cookie(sk) : 0;
4615 }
4616
4617 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4618         .func           = bpf_get_socket_ptr_cookie,
4619         .gpl_only       = false,
4620         .ret_type       = RET_INTEGER,
4621         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4622 };
4623
4624 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4625 {
4626         return __sock_gen_cookie(ctx->sk);
4627 }
4628
4629 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4630         .func           = bpf_get_socket_cookie_sock_ops,
4631         .gpl_only       = false,
4632         .ret_type       = RET_INTEGER,
4633         .arg1_type      = ARG_PTR_TO_CTX,
4634 };
4635
4636 static u64 __bpf_get_netns_cookie(struct sock *sk)
4637 {
4638         const struct net *net = sk ? sock_net(sk) : &init_net;
4639
4640         return net->net_cookie;
4641 }
4642
4643 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4644 {
4645         return __bpf_get_netns_cookie(ctx);
4646 }
4647
4648 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4649         .func           = bpf_get_netns_cookie_sock,
4650         .gpl_only       = false,
4651         .ret_type       = RET_INTEGER,
4652         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4653 };
4654
4655 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4656 {
4657         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4658 }
4659
4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4661         .func           = bpf_get_netns_cookie_sock_addr,
4662         .gpl_only       = false,
4663         .ret_type       = RET_INTEGER,
4664         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4665 };
4666
4667 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
4668 {
4669         struct sock *sk = sk_to_full_sk(skb->sk);
4670         kuid_t kuid;
4671
4672         if (!sk || !sk_fullsock(sk))
4673                 return overflowuid;
4674         kuid = sock_net_uid(sock_net(sk), sk);
4675         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
4676 }
4677
4678 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
4679         .func           = bpf_get_socket_uid,
4680         .gpl_only       = false,
4681         .ret_type       = RET_INTEGER,
4682         .arg1_type      = ARG_PTR_TO_CTX,
4683 };
4684
4685 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
4686                            char *optval, int optlen)
4687 {
4688         char devname[IFNAMSIZ];
4689         int val, valbool;
4690         struct net *net;
4691         int ifindex;
4692         int ret = 0;
4693
4694         if (!sk_fullsock(sk))
4695                 return -EINVAL;
4696
4697         sock_owned_by_me(sk);
4698
4699         if (level == SOL_SOCKET) {
4700                 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE)
4701                         return -EINVAL;
4702                 val = *((int *)optval);
4703                 valbool = val ? 1 : 0;
4704
4705                 /* Only some socketops are supported */
4706                 switch (optname) {
4707                 case SO_RCVBUF:
4708                         val = min_t(u32, val, sysctl_rmem_max);
4709                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
4710                         WRITE_ONCE(sk->sk_rcvbuf,
4711                                    max_t(int, val * 2, SOCK_MIN_RCVBUF));
4712                         break;
4713                 case SO_SNDBUF:
4714                         val = min_t(u32, val, sysctl_wmem_max);
4715                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
4716                         WRITE_ONCE(sk->sk_sndbuf,
4717                                    max_t(int, val * 2, SOCK_MIN_SNDBUF));
4718                         break;
4719                 case SO_MAX_PACING_RATE: /* 32bit version */
4720                         if (val != ~0U)
4721                                 cmpxchg(&sk->sk_pacing_status,
4722                                         SK_PACING_NONE,
4723                                         SK_PACING_NEEDED);
4724                         sk->sk_max_pacing_rate = (val == ~0U) ?
4725                                                  ~0UL : (unsigned int)val;
4726                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
4727                                                  sk->sk_max_pacing_rate);
4728                         break;
4729                 case SO_PRIORITY:
4730                         sk->sk_priority = val;
4731                         break;
4732                 case SO_RCVLOWAT:
4733                         if (val < 0)
4734                                 val = INT_MAX;
4735                         WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
4736                         break;
4737                 case SO_MARK:
4738                         if (sk->sk_mark != val) {
4739                                 sk->sk_mark = val;
4740                                 sk_dst_reset(sk);
4741                         }
4742                         break;
4743                 case SO_BINDTODEVICE:
4744                         optlen = min_t(long, optlen, IFNAMSIZ - 1);
4745                         strncpy(devname, optval, optlen);
4746                         devname[optlen] = 0;
4747
4748                         ifindex = 0;
4749                         if (devname[0] != '\0') {
4750                                 struct net_device *dev;
4751
4752                                 ret = -ENODEV;
4753
4754                                 net = sock_net(sk);
4755                                 dev = dev_get_by_name(net, devname);
4756                                 if (!dev)
4757                                         break;
4758                                 ifindex = dev->ifindex;
4759                                 dev_put(dev);
4760                         }
4761                         fallthrough;
4762                 case SO_BINDTOIFINDEX:
4763                         if (optname == SO_BINDTOIFINDEX)
4764                                 ifindex = val;
4765                         ret = sock_bindtoindex(sk, ifindex, false);
4766                         break;
4767                 case SO_KEEPALIVE:
4768                         if (sk->sk_prot->keepalive)
4769                                 sk->sk_prot->keepalive(sk, valbool);
4770                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
4771                         break;
4772                 case SO_REUSEPORT:
4773                         sk->sk_reuseport = valbool;
4774                         break;
4775                 default:
4776                         ret = -EINVAL;
4777                 }
4778 #ifdef CONFIG_INET
4779         } else if (level == SOL_IP) {
4780                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4781                         return -EINVAL;
4782
4783                 val = *((int *)optval);
4784                 /* Only some options are supported */
4785                 switch (optname) {
4786                 case IP_TOS:
4787                         if (val < -1 || val > 0xff) {
4788                                 ret = -EINVAL;
4789                         } else {
4790                                 struct inet_sock *inet = inet_sk(sk);
4791
4792                                 if (val == -1)
4793                                         val = 0;
4794                                 inet->tos = val;
4795                         }
4796                         break;
4797                 default:
4798                         ret = -EINVAL;
4799                 }
4800 #if IS_ENABLED(CONFIG_IPV6)
4801         } else if (level == SOL_IPV6) {
4802                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4803                         return -EINVAL;
4804
4805                 val = *((int *)optval);
4806                 /* Only some options are supported */
4807                 switch (optname) {
4808                 case IPV6_TCLASS:
4809                         if (val < -1 || val > 0xff) {
4810                                 ret = -EINVAL;
4811                         } else {
4812                                 struct ipv6_pinfo *np = inet6_sk(sk);
4813
4814                                 if (val == -1)
4815                                         val = 0;
4816                                 np->tclass = val;
4817                         }
4818                         break;
4819                 default:
4820                         ret = -EINVAL;
4821                 }
4822 #endif
4823         } else if (level == SOL_TCP &&
4824                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4825                 if (optname == TCP_CONGESTION) {
4826                         char name[TCP_CA_NAME_MAX];
4827
4828                         strncpy(name, optval, min_t(long, optlen,
4829                                                     TCP_CA_NAME_MAX-1));
4830                         name[TCP_CA_NAME_MAX-1] = 0;
4831                         ret = tcp_set_congestion_control(sk, name, false, true);
4832                 } else {
4833                         struct inet_connection_sock *icsk = inet_csk(sk);
4834                         struct tcp_sock *tp = tcp_sk(sk);
4835                         unsigned long timeout;
4836
4837                         if (optlen != sizeof(int))
4838                                 return -EINVAL;
4839
4840                         val = *((int *)optval);
4841                         /* Only some options are supported */
4842                         switch (optname) {
4843                         case TCP_BPF_IW:
4844                                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
4845                                         ret = -EINVAL;
4846                                 else
4847                                         tp->snd_cwnd = val;
4848                                 break;
4849                         case TCP_BPF_SNDCWND_CLAMP:
4850                                 if (val <= 0) {
4851                                         ret = -EINVAL;
4852                                 } else {
4853                                         tp->snd_cwnd_clamp = val;
4854                                         tp->snd_ssthresh = val;
4855                                 }
4856                                 break;
4857                         case TCP_BPF_DELACK_MAX:
4858                                 timeout = usecs_to_jiffies(val);
4859                                 if (timeout > TCP_DELACK_MAX ||
4860                                     timeout < TCP_TIMEOUT_MIN)
4861                                         return -EINVAL;
4862                                 inet_csk(sk)->icsk_delack_max = timeout;
4863                                 break;
4864                         case TCP_BPF_RTO_MIN:
4865                                 timeout = usecs_to_jiffies(val);
4866                                 if (timeout > TCP_RTO_MIN ||
4867                                     timeout < TCP_TIMEOUT_MIN)
4868                                         return -EINVAL;
4869                                 inet_csk(sk)->icsk_rto_min = timeout;
4870                                 break;
4871                         case TCP_SAVE_SYN:
4872                                 if (val < 0 || val > 1)
4873                                         ret = -EINVAL;
4874                                 else
4875                                         tp->save_syn = val;
4876                                 break;
4877                         case TCP_KEEPIDLE:
4878                                 ret = tcp_sock_set_keepidle_locked(sk, val);
4879                                 break;
4880                         case TCP_KEEPINTVL:
4881                                 if (val < 1 || val > MAX_TCP_KEEPINTVL)
4882                                         ret = -EINVAL;
4883                                 else
4884                                         tp->keepalive_intvl = val * HZ;
4885                                 break;
4886                         case TCP_KEEPCNT:
4887                                 if (val < 1 || val > MAX_TCP_KEEPCNT)
4888                                         ret = -EINVAL;
4889                                 else
4890                                         tp->keepalive_probes = val;
4891                                 break;
4892                         case TCP_SYNCNT:
4893                                 if (val < 1 || val > MAX_TCP_SYNCNT)
4894                                         ret = -EINVAL;
4895                                 else
4896                                         icsk->icsk_syn_retries = val;
4897                                 break;
4898                         case TCP_USER_TIMEOUT:
4899                                 if (val < 0)
4900                                         ret = -EINVAL;
4901                                 else
4902                                         icsk->icsk_user_timeout = val;
4903                                 break;
4904                         case TCP_NOTSENT_LOWAT:
4905                                 tp->notsent_lowat = val;
4906                                 sk->sk_write_space(sk);
4907                                 break;
4908                         case TCP_WINDOW_CLAMP:
4909                                 ret = tcp_set_window_clamp(sk, val);
4910                                 break;
4911                         default:
4912                                 ret = -EINVAL;
4913                         }
4914                 }
4915 #endif
4916         } else {
4917                 ret = -EINVAL;
4918         }
4919         return ret;
4920 }
4921
4922 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
4923                            char *optval, int optlen)
4924 {
4925         if (!sk_fullsock(sk))
4926                 goto err_clear;
4927
4928         sock_owned_by_me(sk);
4929
4930         if (level == SOL_SOCKET) {
4931                 if (optlen != sizeof(int))
4932                         goto err_clear;
4933
4934                 switch (optname) {
4935                 case SO_MARK:
4936                         *((int *)optval) = sk->sk_mark;
4937                         break;
4938                 case SO_PRIORITY:
4939                         *((int *)optval) = sk->sk_priority;
4940                         break;
4941                 case SO_BINDTOIFINDEX:
4942                         *((int *)optval) = sk->sk_bound_dev_if;
4943                         break;
4944                 case SO_REUSEPORT:
4945                         *((int *)optval) = sk->sk_reuseport;
4946                         break;
4947                 default:
4948                         goto err_clear;
4949                 }
4950 #ifdef CONFIG_INET
4951         } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4952                 struct inet_connection_sock *icsk;
4953                 struct tcp_sock *tp;
4954
4955                 switch (optname) {
4956                 case TCP_CONGESTION:
4957                         icsk = inet_csk(sk);
4958
4959                         if (!icsk->icsk_ca_ops || optlen <= 1)
4960                                 goto err_clear;
4961                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4962                         optval[optlen - 1] = 0;
4963                         break;
4964                 case TCP_SAVED_SYN:
4965                         tp = tcp_sk(sk);
4966
4967                         if (optlen <= 0 || !tp->saved_syn ||
4968                             optlen > tcp_saved_syn_len(tp->saved_syn))
4969                                 goto err_clear;
4970                         memcpy(optval, tp->saved_syn->data, optlen);
4971                         break;
4972                 default:
4973                         goto err_clear;
4974                 }
4975         } else if (level == SOL_IP) {
4976                 struct inet_sock *inet = inet_sk(sk);
4977
4978                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4979                         goto err_clear;
4980
4981                 /* Only some options are supported */
4982                 switch (optname) {
4983                 case IP_TOS:
4984                         *((int *)optval) = (int)inet->tos;
4985                         break;
4986                 default:
4987                         goto err_clear;
4988                 }
4989 #if IS_ENABLED(CONFIG_IPV6)
4990         } else if (level == SOL_IPV6) {
4991                 struct ipv6_pinfo *np = inet6_sk(sk);
4992
4993                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4994                         goto err_clear;
4995
4996                 /* Only some options are supported */
4997                 switch (optname) {
4998                 case IPV6_TCLASS:
4999                         *((int *)optval) = (int)np->tclass;
5000                         break;
5001                 default:
5002                         goto err_clear;
5003                 }
5004 #endif
5005 #endif
5006         } else {
5007                 goto err_clear;
5008         }
5009         return 0;
5010 err_clear:
5011         memset(optval, 0, optlen);
5012         return -EINVAL;
5013 }
5014
5015 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5016            int, level, int, optname, char *, optval, int, optlen)
5017 {
5018         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5019 }
5020
5021 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5022         .func           = bpf_sock_addr_setsockopt,
5023         .gpl_only       = false,
5024         .ret_type       = RET_INTEGER,
5025         .arg1_type      = ARG_PTR_TO_CTX,
5026         .arg2_type      = ARG_ANYTHING,
5027         .arg3_type      = ARG_ANYTHING,
5028         .arg4_type      = ARG_PTR_TO_MEM,
5029         .arg5_type      = ARG_CONST_SIZE,
5030 };
5031
5032 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5033            int, level, int, optname, char *, optval, int, optlen)
5034 {
5035         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5036 }
5037
5038 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5039         .func           = bpf_sock_addr_getsockopt,
5040         .gpl_only       = false,
5041         .ret_type       = RET_INTEGER,
5042         .arg1_type      = ARG_PTR_TO_CTX,
5043         .arg2_type      = ARG_ANYTHING,
5044         .arg3_type      = ARG_ANYTHING,
5045         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5046         .arg5_type      = ARG_CONST_SIZE,
5047 };
5048
5049 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5050            int, level, int, optname, char *, optval, int, optlen)
5051 {
5052         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5053 }
5054
5055 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5056         .func           = bpf_sock_ops_setsockopt,
5057         .gpl_only       = false,
5058         .ret_type       = RET_INTEGER,
5059         .arg1_type      = ARG_PTR_TO_CTX,
5060         .arg2_type      = ARG_ANYTHING,
5061         .arg3_type      = ARG_ANYTHING,
5062         .arg4_type      = ARG_PTR_TO_MEM,
5063         .arg5_type      = ARG_CONST_SIZE,
5064 };
5065
5066 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5067                                 int optname, const u8 **start)
5068 {
5069         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5070         const u8 *hdr_start;
5071         int ret;
5072
5073         if (syn_skb) {
5074                 /* sk is a request_sock here */
5075
5076                 if (optname == TCP_BPF_SYN) {
5077                         hdr_start = syn_skb->data;
5078                         ret = tcp_hdrlen(syn_skb);
5079                 } else if (optname == TCP_BPF_SYN_IP) {
5080                         hdr_start = skb_network_header(syn_skb);
5081                         ret = skb_network_header_len(syn_skb) +
5082                                 tcp_hdrlen(syn_skb);
5083                 } else {
5084                         /* optname == TCP_BPF_SYN_MAC */
5085                         hdr_start = skb_mac_header(syn_skb);
5086                         ret = skb_mac_header_len(syn_skb) +
5087                                 skb_network_header_len(syn_skb) +
5088                                 tcp_hdrlen(syn_skb);
5089                 }
5090         } else {
5091                 struct sock *sk = bpf_sock->sk;
5092                 struct saved_syn *saved_syn;
5093
5094                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5095                         /* synack retransmit. bpf_sock->syn_skb will
5096                          * not be available.  It has to resort to
5097                          * saved_syn (if it is saved).
5098                          */
5099                         saved_syn = inet_reqsk(sk)->saved_syn;
5100                 else
5101                         saved_syn = tcp_sk(sk)->saved_syn;
5102
5103                 if (!saved_syn)
5104                         return -ENOENT;
5105
5106                 if (optname == TCP_BPF_SYN) {
5107                         hdr_start = saved_syn->data +
5108                                 saved_syn->mac_hdrlen +
5109                                 saved_syn->network_hdrlen;
5110                         ret = saved_syn->tcp_hdrlen;
5111                 } else if (optname == TCP_BPF_SYN_IP) {
5112                         hdr_start = saved_syn->data +
5113                                 saved_syn->mac_hdrlen;
5114                         ret = saved_syn->network_hdrlen +
5115                                 saved_syn->tcp_hdrlen;
5116                 } else {
5117                         /* optname == TCP_BPF_SYN_MAC */
5118
5119                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5120                         if (!saved_syn->mac_hdrlen)
5121                                 return -ENOENT;
5122
5123                         hdr_start = saved_syn->data;
5124                         ret = saved_syn->mac_hdrlen +
5125                                 saved_syn->network_hdrlen +
5126                                 saved_syn->tcp_hdrlen;
5127                 }
5128         }
5129
5130         *start = hdr_start;
5131         return ret;
5132 }
5133
5134 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5135            int, level, int, optname, char *, optval, int, optlen)
5136 {
5137         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5138             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5139                 int ret, copy_len = 0;
5140                 const u8 *start;
5141
5142                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5143                 if (ret > 0) {
5144                         copy_len = ret;
5145                         if (optlen < copy_len) {
5146                                 copy_len = optlen;
5147                                 ret = -ENOSPC;
5148                         }
5149
5150                         memcpy(optval, start, copy_len);
5151                 }
5152
5153                 /* Zero out unused buffer at the end */
5154                 memset(optval + copy_len, 0, optlen - copy_len);
5155
5156                 return ret;
5157         }
5158
5159         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5160 }
5161
5162 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5163         .func           = bpf_sock_ops_getsockopt,
5164         .gpl_only       = false,
5165         .ret_type       = RET_INTEGER,
5166         .arg1_type      = ARG_PTR_TO_CTX,
5167         .arg2_type      = ARG_ANYTHING,
5168         .arg3_type      = ARG_ANYTHING,
5169         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5170         .arg5_type      = ARG_CONST_SIZE,
5171 };
5172
5173 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5174            int, argval)
5175 {
5176         struct sock *sk = bpf_sock->sk;
5177         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5178
5179         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5180                 return -EINVAL;
5181
5182         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5183
5184         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5185 }
5186
5187 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5188         .func           = bpf_sock_ops_cb_flags_set,
5189         .gpl_only       = false,
5190         .ret_type       = RET_INTEGER,
5191         .arg1_type      = ARG_PTR_TO_CTX,
5192         .arg2_type      = ARG_ANYTHING,
5193 };
5194
5195 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5196 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5197
5198 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5199            int, addr_len)
5200 {
5201 #ifdef CONFIG_INET
5202         struct sock *sk = ctx->sk;
5203         u32 flags = BIND_FROM_BPF;
5204         int err;
5205
5206         err = -EINVAL;
5207         if (addr_len < offsetofend(struct sockaddr, sa_family))
5208                 return err;
5209         if (addr->sa_family == AF_INET) {
5210                 if (addr_len < sizeof(struct sockaddr_in))
5211                         return err;
5212                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5213                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5214                 return __inet_bind(sk, addr, addr_len, flags);
5215 #if IS_ENABLED(CONFIG_IPV6)
5216         } else if (addr->sa_family == AF_INET6) {
5217                 if (addr_len < SIN6_LEN_RFC2133)
5218                         return err;
5219                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5220                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5221                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5222                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5223                  */
5224                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5225 #endif /* CONFIG_IPV6 */
5226         }
5227 #endif /* CONFIG_INET */
5228
5229         return -EAFNOSUPPORT;
5230 }
5231
5232 static const struct bpf_func_proto bpf_bind_proto = {
5233         .func           = bpf_bind,
5234         .gpl_only       = false,
5235         .ret_type       = RET_INTEGER,
5236         .arg1_type      = ARG_PTR_TO_CTX,
5237         .arg2_type      = ARG_PTR_TO_MEM,
5238         .arg3_type      = ARG_CONST_SIZE,
5239 };
5240
5241 #ifdef CONFIG_XFRM
5242 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5243            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5244 {
5245         const struct sec_path *sp = skb_sec_path(skb);
5246         const struct xfrm_state *x;
5247
5248         if (!sp || unlikely(index >= sp->len || flags))
5249                 goto err_clear;
5250
5251         x = sp->xvec[index];
5252
5253         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5254                 goto err_clear;
5255
5256         to->reqid = x->props.reqid;
5257         to->spi = x->id.spi;
5258         to->family = x->props.family;
5259         to->ext = 0;
5260
5261         if (to->family == AF_INET6) {
5262                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5263                        sizeof(to->remote_ipv6));
5264         } else {
5265                 to->remote_ipv4 = x->props.saddr.a4;
5266                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5267         }
5268
5269         return 0;
5270 err_clear:
5271         memset(to, 0, size);
5272         return -EINVAL;
5273 }
5274
5275 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5276         .func           = bpf_skb_get_xfrm_state,
5277         .gpl_only       = false,
5278         .ret_type       = RET_INTEGER,
5279         .arg1_type      = ARG_PTR_TO_CTX,
5280         .arg2_type      = ARG_ANYTHING,
5281         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5282         .arg4_type      = ARG_CONST_SIZE,
5283         .arg5_type      = ARG_ANYTHING,
5284 };
5285 #endif
5286
5287 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5288 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5289                                   const struct neighbour *neigh,
5290                                   const struct net_device *dev, u32 mtu)
5291 {
5292         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5293         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5294         params->h_vlan_TCI = 0;
5295         params->h_vlan_proto = 0;
5296         if (mtu)
5297                 params->mtu_result = mtu; /* union with tot_len */
5298
5299         return 0;
5300 }
5301 #endif
5302
5303 #if IS_ENABLED(CONFIG_INET)
5304 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5305                                u32 flags, bool check_mtu)
5306 {
5307         struct fib_nh_common *nhc;
5308         struct in_device *in_dev;
5309         struct neighbour *neigh;
5310         struct net_device *dev;
5311         struct fib_result res;
5312         struct flowi4 fl4;
5313         u32 mtu = 0;
5314         int err;
5315
5316         dev = dev_get_by_index_rcu(net, params->ifindex);
5317         if (unlikely(!dev))
5318                 return -ENODEV;
5319
5320         /* verify forwarding is enabled on this interface */
5321         in_dev = __in_dev_get_rcu(dev);
5322         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5323                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5324
5325         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5326                 fl4.flowi4_iif = 1;
5327                 fl4.flowi4_oif = params->ifindex;
5328         } else {
5329                 fl4.flowi4_iif = params->ifindex;
5330                 fl4.flowi4_oif = 0;
5331         }
5332         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5333         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5334         fl4.flowi4_flags = 0;
5335
5336         fl4.flowi4_proto = params->l4_protocol;
5337         fl4.daddr = params->ipv4_dst;
5338         fl4.saddr = params->ipv4_src;
5339         fl4.fl4_sport = params->sport;
5340         fl4.fl4_dport = params->dport;
5341         fl4.flowi4_multipath_hash = 0;
5342
5343         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5344                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5345                 struct fib_table *tb;
5346
5347                 tb = fib_get_table(net, tbid);
5348                 if (unlikely(!tb))
5349                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5350
5351                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5352         } else {
5353                 fl4.flowi4_mark = 0;
5354                 fl4.flowi4_secid = 0;
5355                 fl4.flowi4_tun_key.tun_id = 0;
5356                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5357
5358                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5359         }
5360
5361         if (err) {
5362                 /* map fib lookup errors to RTN_ type */
5363                 if (err == -EINVAL)
5364                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5365                 if (err == -EHOSTUNREACH)
5366                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5367                 if (err == -EACCES)
5368                         return BPF_FIB_LKUP_RET_PROHIBIT;
5369
5370                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5371         }
5372
5373         if (res.type != RTN_UNICAST)
5374                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5375
5376         if (fib_info_num_path(res.fi) > 1)
5377                 fib_select_path(net, &res, &fl4, NULL);
5378
5379         if (check_mtu) {
5380                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5381                 if (params->tot_len > mtu) {
5382                         params->mtu_result = mtu; /* union with tot_len */
5383                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5384                 }
5385         }
5386
5387         nhc = res.nhc;
5388
5389         /* do not handle lwt encaps right now */
5390         if (nhc->nhc_lwtstate)
5391                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5392
5393         dev = nhc->nhc_dev;
5394
5395         params->rt_metric = res.fi->fib_priority;
5396         params->ifindex = dev->ifindex;
5397
5398         /* xdp and cls_bpf programs are run in RCU-bh so
5399          * rcu_read_lock_bh is not needed here
5400          */
5401         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5402                 if (nhc->nhc_gw_family)
5403                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5404
5405                 neigh = __ipv4_neigh_lookup_noref(dev,
5406                                                  (__force u32)params->ipv4_dst);
5407         } else {
5408                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5409
5410                 params->family = AF_INET6;
5411                 *dst = nhc->nhc_gw.ipv6;
5412                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5413         }
5414
5415         if (!neigh)
5416                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5417
5418         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5419 }
5420 #endif
5421
5422 #if IS_ENABLED(CONFIG_IPV6)
5423 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5424                                u32 flags, bool check_mtu)
5425 {
5426         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5427         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5428         struct fib6_result res = {};
5429         struct neighbour *neigh;
5430         struct net_device *dev;
5431         struct inet6_dev *idev;
5432         struct flowi6 fl6;
5433         int strict = 0;
5434         int oif, err;
5435         u32 mtu = 0;
5436
5437         /* link local addresses are never forwarded */
5438         if (rt6_need_strict(dst) || rt6_need_strict(src))
5439                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5440
5441         dev = dev_get_by_index_rcu(net, params->ifindex);
5442         if (unlikely(!dev))
5443                 return -ENODEV;
5444
5445         idev = __in6_dev_get_safely(dev);
5446         if (unlikely(!idev || !idev->cnf.forwarding))
5447                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5448
5449         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5450                 fl6.flowi6_iif = 1;
5451                 oif = fl6.flowi6_oif = params->ifindex;
5452         } else {
5453                 oif = fl6.flowi6_iif = params->ifindex;
5454                 fl6.flowi6_oif = 0;
5455                 strict = RT6_LOOKUP_F_HAS_SADDR;
5456         }
5457         fl6.flowlabel = params->flowinfo;
5458         fl6.flowi6_scope = 0;
5459         fl6.flowi6_flags = 0;
5460         fl6.mp_hash = 0;
5461
5462         fl6.flowi6_proto = params->l4_protocol;
5463         fl6.daddr = *dst;
5464         fl6.saddr = *src;
5465         fl6.fl6_sport = params->sport;
5466         fl6.fl6_dport = params->dport;
5467
5468         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5469                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5470                 struct fib6_table *tb;
5471
5472                 tb = ipv6_stub->fib6_get_table(net, tbid);
5473                 if (unlikely(!tb))
5474                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5475
5476                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5477                                                    strict);
5478         } else {
5479                 fl6.flowi6_mark = 0;
5480                 fl6.flowi6_secid = 0;
5481                 fl6.flowi6_tun_key.tun_id = 0;
5482                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5483
5484                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5485         }
5486
5487         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5488                      res.f6i == net->ipv6.fib6_null_entry))
5489                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5490
5491         switch (res.fib6_type) {
5492         /* only unicast is forwarded */
5493         case RTN_UNICAST:
5494                 break;
5495         case RTN_BLACKHOLE:
5496                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5497         case RTN_UNREACHABLE:
5498                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5499         case RTN_PROHIBIT:
5500                 return BPF_FIB_LKUP_RET_PROHIBIT;
5501         default:
5502                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5503         }
5504
5505         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5506                                     fl6.flowi6_oif != 0, NULL, strict);
5507
5508         if (check_mtu) {
5509                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5510                 if (params->tot_len > mtu) {
5511                         params->mtu_result = mtu; /* union with tot_len */
5512                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5513                 }
5514         }
5515
5516         if (res.nh->fib_nh_lws)
5517                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5518
5519         if (res.nh->fib_nh_gw_family)
5520                 *dst = res.nh->fib_nh_gw6;
5521
5522         dev = res.nh->fib_nh_dev;
5523         params->rt_metric = res.f6i->fib6_metric;
5524         params->ifindex = dev->ifindex;
5525
5526         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5527          * not needed here.
5528          */
5529         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5530         if (!neigh)
5531                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5532
5533         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5534 }
5535 #endif
5536
5537 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5538            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5539 {
5540         if (plen < sizeof(*params))
5541                 return -EINVAL;
5542
5543         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5544                 return -EINVAL;
5545
5546         switch (params->family) {
5547 #if IS_ENABLED(CONFIG_INET)
5548         case AF_INET:
5549                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5550                                            flags, true);
5551 #endif
5552 #if IS_ENABLED(CONFIG_IPV6)
5553         case AF_INET6:
5554                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5555                                            flags, true);
5556 #endif
5557         }
5558         return -EAFNOSUPPORT;
5559 }
5560
5561 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5562         .func           = bpf_xdp_fib_lookup,
5563         .gpl_only       = true,
5564         .ret_type       = RET_INTEGER,
5565         .arg1_type      = ARG_PTR_TO_CTX,
5566         .arg2_type      = ARG_PTR_TO_MEM,
5567         .arg3_type      = ARG_CONST_SIZE,
5568         .arg4_type      = ARG_ANYTHING,
5569 };
5570
5571 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5572            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5573 {
5574         struct net *net = dev_net(skb->dev);
5575         int rc = -EAFNOSUPPORT;
5576         bool check_mtu = false;
5577
5578         if (plen < sizeof(*params))
5579                 return -EINVAL;
5580
5581         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5582                 return -EINVAL;
5583
5584         if (params->tot_len)
5585                 check_mtu = true;
5586
5587         switch (params->family) {
5588 #if IS_ENABLED(CONFIG_INET)
5589         case AF_INET:
5590                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5591                 break;
5592 #endif
5593 #if IS_ENABLED(CONFIG_IPV6)
5594         case AF_INET6:
5595                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5596                 break;
5597 #endif
5598         }
5599
5600         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5601                 struct net_device *dev;
5602
5603                 /* When tot_len isn't provided by user, check skb
5604                  * against MTU of FIB lookup resulting net_device
5605                  */
5606                 dev = dev_get_by_index_rcu(net, params->ifindex);
5607                 if (!is_skb_forwardable(dev, skb))
5608                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5609
5610                 params->mtu_result = dev->mtu; /* union with tot_len */
5611         }
5612
5613         return rc;
5614 }
5615
5616 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5617         .func           = bpf_skb_fib_lookup,
5618         .gpl_only       = true,
5619         .ret_type       = RET_INTEGER,
5620         .arg1_type      = ARG_PTR_TO_CTX,
5621         .arg2_type      = ARG_PTR_TO_MEM,
5622         .arg3_type      = ARG_CONST_SIZE,
5623         .arg4_type      = ARG_ANYTHING,
5624 };
5625
5626 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5627                                             u32 ifindex)
5628 {
5629         struct net *netns = dev_net(dev_curr);
5630
5631         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5632         if (ifindex == 0)
5633                 return dev_curr;
5634
5635         return dev_get_by_index_rcu(netns, ifindex);
5636 }
5637
5638 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5639            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5640 {
5641         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5642         struct net_device *dev = skb->dev;
5643         int skb_len, dev_len;
5644         int mtu;
5645
5646         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5647                 return -EINVAL;
5648
5649         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5650                 return -EINVAL;
5651
5652         dev = __dev_via_ifindex(dev, ifindex);
5653         if (unlikely(!dev))
5654                 return -ENODEV;
5655
5656         mtu = READ_ONCE(dev->mtu);
5657
5658         dev_len = mtu + dev->hard_header_len;
5659
5660         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5661         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
5662
5663         skb_len += len_diff; /* minus result pass check */
5664         if (skb_len <= dev_len) {
5665                 ret = BPF_MTU_CHK_RET_SUCCESS;
5666                 goto out;
5667         }
5668         /* At this point, skb->len exceed MTU, but as it include length of all
5669          * segments, it can still be below MTU.  The SKB can possibly get
5670          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
5671          * must choose if segs are to be MTU checked.
5672          */
5673         if (skb_is_gso(skb)) {
5674                 ret = BPF_MTU_CHK_RET_SUCCESS;
5675
5676                 if (flags & BPF_MTU_CHK_SEGS &&
5677                     !skb_gso_validate_network_len(skb, mtu))
5678                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
5679         }
5680 out:
5681         /* BPF verifier guarantees valid pointer */
5682         *mtu_len = mtu;
5683
5684         return ret;
5685 }
5686
5687 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
5688            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5689 {
5690         struct net_device *dev = xdp->rxq->dev;
5691         int xdp_len = xdp->data_end - xdp->data;
5692         int ret = BPF_MTU_CHK_RET_SUCCESS;
5693         int mtu, dev_len;
5694
5695         /* XDP variant doesn't support multi-buffer segment check (yet) */
5696         if (unlikely(flags))
5697                 return -EINVAL;
5698
5699         dev = __dev_via_ifindex(dev, ifindex);
5700         if (unlikely(!dev))
5701                 return -ENODEV;
5702
5703         mtu = READ_ONCE(dev->mtu);
5704
5705         /* Add L2-header as dev MTU is L3 size */
5706         dev_len = mtu + dev->hard_header_len;
5707
5708         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
5709         if (*mtu_len)
5710                 xdp_len = *mtu_len + dev->hard_header_len;
5711
5712         xdp_len += len_diff; /* minus result pass check */
5713         if (xdp_len > dev_len)
5714                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5715
5716         /* BPF verifier guarantees valid pointer */
5717         *mtu_len = mtu;
5718
5719         return ret;
5720 }
5721
5722 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
5723         .func           = bpf_skb_check_mtu,
5724         .gpl_only       = true,
5725         .ret_type       = RET_INTEGER,
5726         .arg1_type      = ARG_PTR_TO_CTX,
5727         .arg2_type      = ARG_ANYTHING,
5728         .arg3_type      = ARG_PTR_TO_INT,
5729         .arg4_type      = ARG_ANYTHING,
5730         .arg5_type      = ARG_ANYTHING,
5731 };
5732
5733 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
5734         .func           = bpf_xdp_check_mtu,
5735         .gpl_only       = true,
5736         .ret_type       = RET_INTEGER,
5737         .arg1_type      = ARG_PTR_TO_CTX,
5738         .arg2_type      = ARG_ANYTHING,
5739         .arg3_type      = ARG_PTR_TO_INT,
5740         .arg4_type      = ARG_ANYTHING,
5741         .arg5_type      = ARG_ANYTHING,
5742 };
5743
5744 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5745 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
5746 {
5747         int err;
5748         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
5749
5750         if (!seg6_validate_srh(srh, len, false))
5751                 return -EINVAL;
5752
5753         switch (type) {
5754         case BPF_LWT_ENCAP_SEG6_INLINE:
5755                 if (skb->protocol != htons(ETH_P_IPV6))
5756                         return -EBADMSG;
5757
5758                 err = seg6_do_srh_inline(skb, srh);
5759                 break;
5760         case BPF_LWT_ENCAP_SEG6:
5761                 skb_reset_inner_headers(skb);
5762                 skb->encapsulation = 1;
5763                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
5764                 break;
5765         default:
5766                 return -EINVAL;
5767         }
5768
5769         bpf_compute_data_pointers(skb);
5770         if (err)
5771                 return err;
5772
5773         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
5774         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
5775
5776         return seg6_lookup_nexthop(skb, NULL, 0);
5777 }
5778 #endif /* CONFIG_IPV6_SEG6_BPF */
5779
5780 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5781 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
5782                              bool ingress)
5783 {
5784         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
5785 }
5786 #endif
5787
5788 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
5789            u32, len)
5790 {
5791         switch (type) {
5792 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5793         case BPF_LWT_ENCAP_SEG6:
5794         case BPF_LWT_ENCAP_SEG6_INLINE:
5795                 return bpf_push_seg6_encap(skb, type, hdr, len);
5796 #endif
5797 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5798         case BPF_LWT_ENCAP_IP:
5799                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
5800 #endif
5801         default:
5802                 return -EINVAL;
5803         }
5804 }
5805
5806 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
5807            void *, hdr, u32, len)
5808 {
5809         switch (type) {
5810 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
5811         case BPF_LWT_ENCAP_IP:
5812                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
5813 #endif
5814         default:
5815                 return -EINVAL;
5816         }
5817 }
5818
5819 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
5820         .func           = bpf_lwt_in_push_encap,
5821         .gpl_only       = false,
5822         .ret_type       = RET_INTEGER,
5823         .arg1_type      = ARG_PTR_TO_CTX,
5824         .arg2_type      = ARG_ANYTHING,
5825         .arg3_type      = ARG_PTR_TO_MEM,
5826         .arg4_type      = ARG_CONST_SIZE
5827 };
5828
5829 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
5830         .func           = bpf_lwt_xmit_push_encap,
5831         .gpl_only       = false,
5832         .ret_type       = RET_INTEGER,
5833         .arg1_type      = ARG_PTR_TO_CTX,
5834         .arg2_type      = ARG_ANYTHING,
5835         .arg3_type      = ARG_PTR_TO_MEM,
5836         .arg4_type      = ARG_CONST_SIZE
5837 };
5838
5839 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5840 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
5841            const void *, from, u32, len)
5842 {
5843         struct seg6_bpf_srh_state *srh_state =
5844                 this_cpu_ptr(&seg6_bpf_srh_states);
5845         struct ipv6_sr_hdr *srh = srh_state->srh;
5846         void *srh_tlvs, *srh_end, *ptr;
5847         int srhoff = 0;
5848
5849         if (srh == NULL)
5850                 return -EINVAL;
5851
5852         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
5853         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
5854
5855         ptr = skb->data + offset;
5856         if (ptr >= srh_tlvs && ptr + len <= srh_end)
5857                 srh_state->valid = false;
5858         else if (ptr < (void *)&srh->flags ||
5859                  ptr + len > (void *)&srh->segments)
5860                 return -EFAULT;
5861
5862         if (unlikely(bpf_try_make_writable(skb, offset + len)))
5863                 return -EFAULT;
5864         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
5865                 return -EINVAL;
5866         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5867
5868         memcpy(skb->data + offset, from, len);
5869         return 0;
5870 }
5871
5872 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
5873         .func           = bpf_lwt_seg6_store_bytes,
5874         .gpl_only       = false,
5875         .ret_type       = RET_INTEGER,
5876         .arg1_type      = ARG_PTR_TO_CTX,
5877         .arg2_type      = ARG_ANYTHING,
5878         .arg3_type      = ARG_PTR_TO_MEM,
5879         .arg4_type      = ARG_CONST_SIZE
5880 };
5881
5882 static void bpf_update_srh_state(struct sk_buff *skb)
5883 {
5884         struct seg6_bpf_srh_state *srh_state =
5885                 this_cpu_ptr(&seg6_bpf_srh_states);
5886         int srhoff = 0;
5887
5888         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
5889                 srh_state->srh = NULL;
5890         } else {
5891                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
5892                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
5893                 srh_state->valid = true;
5894         }
5895 }
5896
5897 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
5898            u32, action, void *, param, u32, param_len)
5899 {
5900         struct seg6_bpf_srh_state *srh_state =
5901                 this_cpu_ptr(&seg6_bpf_srh_states);
5902         int hdroff = 0;
5903         int err;
5904
5905         switch (action) {
5906         case SEG6_LOCAL_ACTION_END_X:
5907                 if (!seg6_bpf_has_valid_srh(skb))
5908                         return -EBADMSG;
5909                 if (param_len != sizeof(struct in6_addr))
5910                         return -EINVAL;
5911                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
5912         case SEG6_LOCAL_ACTION_END_T:
5913                 if (!seg6_bpf_has_valid_srh(skb))
5914                         return -EBADMSG;
5915                 if (param_len != sizeof(int))
5916                         return -EINVAL;
5917                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5918         case SEG6_LOCAL_ACTION_END_DT6:
5919                 if (!seg6_bpf_has_valid_srh(skb))
5920                         return -EBADMSG;
5921                 if (param_len != sizeof(int))
5922                         return -EINVAL;
5923
5924                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
5925                         return -EBADMSG;
5926                 if (!pskb_pull(skb, hdroff))
5927                         return -EBADMSG;
5928
5929                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
5930                 skb_reset_network_header(skb);
5931                 skb_reset_transport_header(skb);
5932                 skb->encapsulation = 0;
5933
5934                 bpf_compute_data_pointers(skb);
5935                 bpf_update_srh_state(skb);
5936                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
5937         case SEG6_LOCAL_ACTION_END_B6:
5938                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5939                         return -EBADMSG;
5940                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
5941                                           param, param_len);
5942                 if (!err)
5943                         bpf_update_srh_state(skb);
5944
5945                 return err;
5946         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
5947                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
5948                         return -EBADMSG;
5949                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
5950                                           param, param_len);
5951                 if (!err)
5952                         bpf_update_srh_state(skb);
5953
5954                 return err;
5955         default:
5956                 return -EINVAL;
5957         }
5958 }
5959
5960 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
5961         .func           = bpf_lwt_seg6_action,
5962         .gpl_only       = false,
5963         .ret_type       = RET_INTEGER,
5964         .arg1_type      = ARG_PTR_TO_CTX,
5965         .arg2_type      = ARG_ANYTHING,
5966         .arg3_type      = ARG_PTR_TO_MEM,
5967         .arg4_type      = ARG_CONST_SIZE
5968 };
5969
5970 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
5971            s32, len)
5972 {
5973         struct seg6_bpf_srh_state *srh_state =
5974                 this_cpu_ptr(&seg6_bpf_srh_states);
5975         struct ipv6_sr_hdr *srh = srh_state->srh;
5976         void *srh_end, *srh_tlvs, *ptr;
5977         struct ipv6hdr *hdr;
5978         int srhoff = 0;
5979         int ret;
5980
5981         if (unlikely(srh == NULL))
5982                 return -EINVAL;
5983
5984         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
5985                         ((srh->first_segment + 1) << 4));
5986         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
5987                         srh_state->hdrlen);
5988         ptr = skb->data + offset;
5989
5990         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
5991                 return -EFAULT;
5992         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
5993                 return -EFAULT;
5994
5995         if (len > 0) {
5996                 ret = skb_cow_head(skb, len);
5997                 if (unlikely(ret < 0))
5998                         return ret;
5999
6000                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6001         } else {
6002                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6003         }
6004
6005         bpf_compute_data_pointers(skb);
6006         if (unlikely(ret < 0))
6007                 return ret;
6008
6009         hdr = (struct ipv6hdr *)skb->data;
6010         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6011
6012         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6013                 return -EINVAL;
6014         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6015         srh_state->hdrlen += len;
6016         srh_state->valid = false;
6017         return 0;
6018 }
6019
6020 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6021         .func           = bpf_lwt_seg6_adjust_srh,
6022         .gpl_only       = false,
6023         .ret_type       = RET_INTEGER,
6024         .arg1_type      = ARG_PTR_TO_CTX,
6025         .arg2_type      = ARG_ANYTHING,
6026         .arg3_type      = ARG_ANYTHING,
6027 };
6028 #endif /* CONFIG_IPV6_SEG6_BPF */
6029
6030 #ifdef CONFIG_INET
6031 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6032                               int dif, int sdif, u8 family, u8 proto)
6033 {
6034         bool refcounted = false;
6035         struct sock *sk = NULL;
6036
6037         if (family == AF_INET) {
6038                 __be32 src4 = tuple->ipv4.saddr;
6039                 __be32 dst4 = tuple->ipv4.daddr;
6040
6041                 if (proto == IPPROTO_TCP)
6042                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6043                                            src4, tuple->ipv4.sport,
6044                                            dst4, tuple->ipv4.dport,
6045                                            dif, sdif, &refcounted);
6046                 else
6047                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6048                                                dst4, tuple->ipv4.dport,
6049                                                dif, sdif, &udp_table, NULL);
6050 #if IS_ENABLED(CONFIG_IPV6)
6051         } else {
6052                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6053                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6054
6055                 if (proto == IPPROTO_TCP)
6056                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6057                                             src6, tuple->ipv6.sport,
6058                                             dst6, ntohs(tuple->ipv6.dport),
6059                                             dif, sdif, &refcounted);
6060                 else if (likely(ipv6_bpf_stub))
6061                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6062                                                             src6, tuple->ipv6.sport,
6063                                                             dst6, tuple->ipv6.dport,
6064                                                             dif, sdif,
6065                                                             &udp_table, NULL);
6066 #endif
6067         }
6068
6069         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6070                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6071                 sk = NULL;
6072         }
6073         return sk;
6074 }
6075
6076 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6077  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6078  * Returns the socket as an 'unsigned long' to simplify the casting in the
6079  * callers to satisfy BPF_CALL declarations.
6080  */
6081 static struct sock *
6082 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6083                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6084                  u64 flags)
6085 {
6086         struct sock *sk = NULL;
6087         u8 family = AF_UNSPEC;
6088         struct net *net;
6089         int sdif;
6090
6091         if (len == sizeof(tuple->ipv4))
6092                 family = AF_INET;
6093         else if (len == sizeof(tuple->ipv6))
6094                 family = AF_INET6;
6095         else
6096                 return NULL;
6097
6098         if (unlikely(family == AF_UNSPEC || flags ||
6099                      !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6100                 goto out;
6101
6102         if (family == AF_INET)
6103                 sdif = inet_sdif(skb);
6104         else
6105                 sdif = inet6_sdif(skb);
6106
6107         if ((s32)netns_id < 0) {
6108                 net = caller_net;
6109                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6110         } else {
6111                 net = get_net_ns_by_id(caller_net, netns_id);
6112                 if (unlikely(!net))
6113                         goto out;
6114                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6115                 put_net(net);
6116         }
6117
6118 out:
6119         return sk;
6120 }
6121
6122 static struct sock *
6123 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6124                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6125                 u64 flags)
6126 {
6127         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6128                                            ifindex, proto, netns_id, flags);
6129
6130         if (sk) {
6131                 sk = sk_to_full_sk(sk);
6132                 if (!sk_fullsock(sk)) {
6133                         sock_gen_put(sk);
6134                         return NULL;
6135                 }
6136         }
6137
6138         return sk;
6139 }
6140
6141 static struct sock *
6142 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6143                u8 proto, u64 netns_id, u64 flags)
6144 {
6145         struct net *caller_net;
6146         int ifindex;
6147
6148         if (skb->dev) {
6149                 caller_net = dev_net(skb->dev);
6150                 ifindex = skb->dev->ifindex;
6151         } else {
6152                 caller_net = sock_net(skb->sk);
6153                 ifindex = 0;
6154         }
6155
6156         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6157                                 netns_id, flags);
6158 }
6159
6160 static struct sock *
6161 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6162               u8 proto, u64 netns_id, u64 flags)
6163 {
6164         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6165                                          flags);
6166
6167         if (sk) {
6168                 sk = sk_to_full_sk(sk);
6169                 if (!sk_fullsock(sk)) {
6170                         sock_gen_put(sk);
6171                         return NULL;
6172                 }
6173         }
6174
6175         return sk;
6176 }
6177
6178 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6179            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6180 {
6181         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6182                                              netns_id, flags);
6183 }
6184
6185 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6186         .func           = bpf_skc_lookup_tcp,
6187         .gpl_only       = false,
6188         .pkt_access     = true,
6189         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6190         .arg1_type      = ARG_PTR_TO_CTX,
6191         .arg2_type      = ARG_PTR_TO_MEM,
6192         .arg3_type      = ARG_CONST_SIZE,
6193         .arg4_type      = ARG_ANYTHING,
6194         .arg5_type      = ARG_ANYTHING,
6195 };
6196
6197 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6198            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6199 {
6200         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6201                                             netns_id, flags);
6202 }
6203
6204 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6205         .func           = bpf_sk_lookup_tcp,
6206         .gpl_only       = false,
6207         .pkt_access     = true,
6208         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6209         .arg1_type      = ARG_PTR_TO_CTX,
6210         .arg2_type      = ARG_PTR_TO_MEM,
6211         .arg3_type      = ARG_CONST_SIZE,
6212         .arg4_type      = ARG_ANYTHING,
6213         .arg5_type      = ARG_ANYTHING,
6214 };
6215
6216 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6217            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6218 {
6219         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6220                                             netns_id, flags);
6221 }
6222
6223 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6224         .func           = bpf_sk_lookup_udp,
6225         .gpl_only       = false,
6226         .pkt_access     = true,
6227         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6228         .arg1_type      = ARG_PTR_TO_CTX,
6229         .arg2_type      = ARG_PTR_TO_MEM,
6230         .arg3_type      = ARG_CONST_SIZE,
6231         .arg4_type      = ARG_ANYTHING,
6232         .arg5_type      = ARG_ANYTHING,
6233 };
6234
6235 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6236 {
6237         if (sk && sk_is_refcounted(sk))
6238                 sock_gen_put(sk);
6239         return 0;
6240 }
6241
6242 static const struct bpf_func_proto bpf_sk_release_proto = {
6243         .func           = bpf_sk_release,
6244         .gpl_only       = false,
6245         .ret_type       = RET_INTEGER,
6246         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6247 };
6248
6249 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6250            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6251 {
6252         struct net *caller_net = dev_net(ctx->rxq->dev);
6253         int ifindex = ctx->rxq->dev->ifindex;
6254
6255         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6256                                               ifindex, IPPROTO_UDP, netns_id,
6257                                               flags);
6258 }
6259
6260 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6261         .func           = bpf_xdp_sk_lookup_udp,
6262         .gpl_only       = false,
6263         .pkt_access     = true,
6264         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6265         .arg1_type      = ARG_PTR_TO_CTX,
6266         .arg2_type      = ARG_PTR_TO_MEM,
6267         .arg3_type      = ARG_CONST_SIZE,
6268         .arg4_type      = ARG_ANYTHING,
6269         .arg5_type      = ARG_ANYTHING,
6270 };
6271
6272 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6273            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6274 {
6275         struct net *caller_net = dev_net(ctx->rxq->dev);
6276         int ifindex = ctx->rxq->dev->ifindex;
6277
6278         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6279                                                ifindex, IPPROTO_TCP, netns_id,
6280                                                flags);
6281 }
6282
6283 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6284         .func           = bpf_xdp_skc_lookup_tcp,
6285         .gpl_only       = false,
6286         .pkt_access     = true,
6287         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6288         .arg1_type      = ARG_PTR_TO_CTX,
6289         .arg2_type      = ARG_PTR_TO_MEM,
6290         .arg3_type      = ARG_CONST_SIZE,
6291         .arg4_type      = ARG_ANYTHING,
6292         .arg5_type      = ARG_ANYTHING,
6293 };
6294
6295 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6296            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6297 {
6298         struct net *caller_net = dev_net(ctx->rxq->dev);
6299         int ifindex = ctx->rxq->dev->ifindex;
6300
6301         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6302                                               ifindex, IPPROTO_TCP, netns_id,
6303                                               flags);
6304 }
6305
6306 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6307         .func           = bpf_xdp_sk_lookup_tcp,
6308         .gpl_only       = false,
6309         .pkt_access     = true,
6310         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6311         .arg1_type      = ARG_PTR_TO_CTX,
6312         .arg2_type      = ARG_PTR_TO_MEM,
6313         .arg3_type      = ARG_CONST_SIZE,
6314         .arg4_type      = ARG_ANYTHING,
6315         .arg5_type      = ARG_ANYTHING,
6316 };
6317
6318 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6319            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6320 {
6321         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6322                                                sock_net(ctx->sk), 0,
6323                                                IPPROTO_TCP, netns_id, flags);
6324 }
6325
6326 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6327         .func           = bpf_sock_addr_skc_lookup_tcp,
6328         .gpl_only       = false,
6329         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6330         .arg1_type      = ARG_PTR_TO_CTX,
6331         .arg2_type      = ARG_PTR_TO_MEM,
6332         .arg3_type      = ARG_CONST_SIZE,
6333         .arg4_type      = ARG_ANYTHING,
6334         .arg5_type      = ARG_ANYTHING,
6335 };
6336
6337 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6338            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6339 {
6340         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6341                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6342                                               netns_id, flags);
6343 }
6344
6345 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6346         .func           = bpf_sock_addr_sk_lookup_tcp,
6347         .gpl_only       = false,
6348         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6349         .arg1_type      = ARG_PTR_TO_CTX,
6350         .arg2_type      = ARG_PTR_TO_MEM,
6351         .arg3_type      = ARG_CONST_SIZE,
6352         .arg4_type      = ARG_ANYTHING,
6353         .arg5_type      = ARG_ANYTHING,
6354 };
6355
6356 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6357            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6358 {
6359         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6360                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6361                                               netns_id, flags);
6362 }
6363
6364 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6365         .func           = bpf_sock_addr_sk_lookup_udp,
6366         .gpl_only       = false,
6367         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6368         .arg1_type      = ARG_PTR_TO_CTX,
6369         .arg2_type      = ARG_PTR_TO_MEM,
6370         .arg3_type      = ARG_CONST_SIZE,
6371         .arg4_type      = ARG_ANYTHING,
6372         .arg5_type      = ARG_ANYTHING,
6373 };
6374
6375 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6376                                   struct bpf_insn_access_aux *info)
6377 {
6378         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6379                                           icsk_retransmits))
6380                 return false;
6381
6382         if (off % size != 0)
6383                 return false;
6384
6385         switch (off) {
6386         case offsetof(struct bpf_tcp_sock, bytes_received):
6387         case offsetof(struct bpf_tcp_sock, bytes_acked):
6388                 return size == sizeof(__u64);
6389         default:
6390                 return size == sizeof(__u32);
6391         }
6392 }
6393
6394 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6395                                     const struct bpf_insn *si,
6396                                     struct bpf_insn *insn_buf,
6397                                     struct bpf_prog *prog, u32 *target_size)
6398 {
6399         struct bpf_insn *insn = insn_buf;
6400
6401 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6402         do {                                                            \
6403                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6404                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6405                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6406                                       si->dst_reg, si->src_reg,         \
6407                                       offsetof(struct tcp_sock, FIELD)); \
6408         } while (0)
6409
6410 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6411         do {                                                            \
6412                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6413                                           FIELD) >                      \
6414                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6415                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6416                                         struct inet_connection_sock,    \
6417                                         FIELD),                         \
6418                                       si->dst_reg, si->src_reg,         \
6419                                       offsetof(                         \
6420                                         struct inet_connection_sock,    \
6421                                         FIELD));                        \
6422         } while (0)
6423
6424         if (insn > insn_buf)
6425                 return insn - insn_buf;
6426
6427         switch (si->off) {
6428         case offsetof(struct bpf_tcp_sock, rtt_min):
6429                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6430                              sizeof(struct minmax));
6431                 BUILD_BUG_ON(sizeof(struct minmax) <
6432                              sizeof(struct minmax_sample));
6433
6434                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6435                                       offsetof(struct tcp_sock, rtt_min) +
6436                                       offsetof(struct minmax_sample, v));
6437                 break;
6438         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6439                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6440                 break;
6441         case offsetof(struct bpf_tcp_sock, srtt_us):
6442                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6443                 break;
6444         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6445                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6446                 break;
6447         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6448                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6449                 break;
6450         case offsetof(struct bpf_tcp_sock, snd_nxt):
6451                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6452                 break;
6453         case offsetof(struct bpf_tcp_sock, snd_una):
6454                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6455                 break;
6456         case offsetof(struct bpf_tcp_sock, mss_cache):
6457                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6458                 break;
6459         case offsetof(struct bpf_tcp_sock, ecn_flags):
6460                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6461                 break;
6462         case offsetof(struct bpf_tcp_sock, rate_delivered):
6463                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6464                 break;
6465         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6466                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6467                 break;
6468         case offsetof(struct bpf_tcp_sock, packets_out):
6469                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6470                 break;
6471         case offsetof(struct bpf_tcp_sock, retrans_out):
6472                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6473                 break;
6474         case offsetof(struct bpf_tcp_sock, total_retrans):
6475                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6476                 break;
6477         case offsetof(struct bpf_tcp_sock, segs_in):
6478                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6479                 break;
6480         case offsetof(struct bpf_tcp_sock, data_segs_in):
6481                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6482                 break;
6483         case offsetof(struct bpf_tcp_sock, segs_out):
6484                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6485                 break;
6486         case offsetof(struct bpf_tcp_sock, data_segs_out):
6487                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6488                 break;
6489         case offsetof(struct bpf_tcp_sock, lost_out):
6490                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6491                 break;
6492         case offsetof(struct bpf_tcp_sock, sacked_out):
6493                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6494                 break;
6495         case offsetof(struct bpf_tcp_sock, bytes_received):
6496                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6497                 break;
6498         case offsetof(struct bpf_tcp_sock, bytes_acked):
6499                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6500                 break;
6501         case offsetof(struct bpf_tcp_sock, dsack_dups):
6502                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6503                 break;
6504         case offsetof(struct bpf_tcp_sock, delivered):
6505                 BPF_TCP_SOCK_GET_COMMON(delivered);
6506                 break;
6507         case offsetof(struct bpf_tcp_sock, delivered_ce):
6508                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6509                 break;
6510         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6511                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6512                 break;
6513         }
6514
6515         return insn - insn_buf;
6516 }
6517
6518 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6519 {
6520         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6521                 return (unsigned long)sk;
6522
6523         return (unsigned long)NULL;
6524 }
6525
6526 const struct bpf_func_proto bpf_tcp_sock_proto = {
6527         .func           = bpf_tcp_sock,
6528         .gpl_only       = false,
6529         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6530         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6531 };
6532
6533 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6534 {
6535         sk = sk_to_full_sk(sk);
6536
6537         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6538                 return (unsigned long)sk;
6539
6540         return (unsigned long)NULL;
6541 }
6542
6543 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6544         .func           = bpf_get_listener_sock,
6545         .gpl_only       = false,
6546         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6547         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6548 };
6549
6550 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6551 {
6552         unsigned int iphdr_len;
6553
6554         switch (skb_protocol(skb, true)) {
6555         case cpu_to_be16(ETH_P_IP):
6556                 iphdr_len = sizeof(struct iphdr);
6557                 break;
6558         case cpu_to_be16(ETH_P_IPV6):
6559                 iphdr_len = sizeof(struct ipv6hdr);
6560                 break;
6561         default:
6562                 return 0;
6563         }
6564
6565         if (skb_headlen(skb) < iphdr_len)
6566                 return 0;
6567
6568         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6569                 return 0;
6570
6571         return INET_ECN_set_ce(skb);
6572 }
6573
6574 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6575                                   struct bpf_insn_access_aux *info)
6576 {
6577         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6578                 return false;
6579
6580         if (off % size != 0)
6581                 return false;
6582
6583         switch (off) {
6584         default:
6585                 return size == sizeof(__u32);
6586         }
6587 }
6588
6589 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6590                                     const struct bpf_insn *si,
6591                                     struct bpf_insn *insn_buf,
6592                                     struct bpf_prog *prog, u32 *target_size)
6593 {
6594         struct bpf_insn *insn = insn_buf;
6595
6596 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6597         do {                                                            \
6598                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6599                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6600                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6601                                       si->dst_reg, si->src_reg,         \
6602                                       offsetof(struct xdp_sock, FIELD)); \
6603         } while (0)
6604
6605         switch (si->off) {
6606         case offsetof(struct bpf_xdp_sock, queue_id):
6607                 BPF_XDP_SOCK_GET(queue_id);
6608                 break;
6609         }
6610
6611         return insn - insn_buf;
6612 }
6613
6614 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6615         .func           = bpf_skb_ecn_set_ce,
6616         .gpl_only       = false,
6617         .ret_type       = RET_INTEGER,
6618         .arg1_type      = ARG_PTR_TO_CTX,
6619 };
6620
6621 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6622            struct tcphdr *, th, u32, th_len)
6623 {
6624 #ifdef CONFIG_SYN_COOKIES
6625         u32 cookie;
6626         int ret;
6627
6628         if (unlikely(!sk || th_len < sizeof(*th)))
6629                 return -EINVAL;
6630
6631         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6632         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6633                 return -EINVAL;
6634
6635         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6636                 return -EINVAL;
6637
6638         if (!th->ack || th->rst || th->syn)
6639                 return -ENOENT;
6640
6641         if (tcp_synq_no_recent_overflow(sk))
6642                 return -ENOENT;
6643
6644         cookie = ntohl(th->ack_seq) - 1;
6645
6646         switch (sk->sk_family) {
6647         case AF_INET:
6648                 if (unlikely(iph_len < sizeof(struct iphdr)))
6649                         return -EINVAL;
6650
6651                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
6652                 break;
6653
6654 #if IS_BUILTIN(CONFIG_IPV6)
6655         case AF_INET6:
6656                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6657                         return -EINVAL;
6658
6659                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
6660                 break;
6661 #endif /* CONFIG_IPV6 */
6662
6663         default:
6664                 return -EPROTONOSUPPORT;
6665         }
6666
6667         if (ret > 0)
6668                 return 0;
6669
6670         return -ENOENT;
6671 #else
6672         return -ENOTSUPP;
6673 #endif
6674 }
6675
6676 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
6677         .func           = bpf_tcp_check_syncookie,
6678         .gpl_only       = true,
6679         .pkt_access     = true,
6680         .ret_type       = RET_INTEGER,
6681         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6682         .arg2_type      = ARG_PTR_TO_MEM,
6683         .arg3_type      = ARG_CONST_SIZE,
6684         .arg4_type      = ARG_PTR_TO_MEM,
6685         .arg5_type      = ARG_CONST_SIZE,
6686 };
6687
6688 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6689            struct tcphdr *, th, u32, th_len)
6690 {
6691 #ifdef CONFIG_SYN_COOKIES
6692         u32 cookie;
6693         u16 mss;
6694
6695         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
6696                 return -EINVAL;
6697
6698         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6699                 return -EINVAL;
6700
6701         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies)
6702                 return -ENOENT;
6703
6704         if (!th->syn || th->ack || th->fin || th->rst)
6705                 return -EINVAL;
6706
6707         if (unlikely(iph_len < sizeof(struct iphdr)))
6708                 return -EINVAL;
6709
6710         /* Both struct iphdr and struct ipv6hdr have the version field at the
6711          * same offset so we can cast to the shorter header (struct iphdr).
6712          */
6713         switch (((struct iphdr *)iph)->version) {
6714         case 4:
6715                 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only)
6716                         return -EINVAL;
6717
6718                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
6719                 break;
6720
6721 #if IS_BUILTIN(CONFIG_IPV6)
6722         case 6:
6723                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
6724                         return -EINVAL;
6725
6726                 if (sk->sk_family != AF_INET6)
6727                         return -EINVAL;
6728
6729                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
6730                 break;
6731 #endif /* CONFIG_IPV6 */
6732
6733         default:
6734                 return -EPROTONOSUPPORT;
6735         }
6736         if (mss == 0)
6737                 return -ENOENT;
6738
6739         return cookie | ((u64)mss << 32);
6740 #else
6741         return -EOPNOTSUPP;
6742 #endif /* CONFIG_SYN_COOKIES */
6743 }
6744
6745 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
6746         .func           = bpf_tcp_gen_syncookie,
6747         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
6748         .pkt_access     = true,
6749         .ret_type       = RET_INTEGER,
6750         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6751         .arg2_type      = ARG_PTR_TO_MEM,
6752         .arg3_type      = ARG_CONST_SIZE,
6753         .arg4_type      = ARG_PTR_TO_MEM,
6754         .arg5_type      = ARG_CONST_SIZE,
6755 };
6756
6757 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
6758 {
6759         if (!sk || flags != 0)
6760                 return -EINVAL;
6761         if (!skb_at_tc_ingress(skb))
6762                 return -EOPNOTSUPP;
6763         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
6764                 return -ENETUNREACH;
6765         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
6766                 return -ESOCKTNOSUPPORT;
6767         if (sk_is_refcounted(sk) &&
6768             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
6769                 return -ENOENT;
6770
6771         skb_orphan(skb);
6772         skb->sk = sk;
6773         skb->destructor = sock_pfree;
6774
6775         return 0;
6776 }
6777
6778 static const struct bpf_func_proto bpf_sk_assign_proto = {
6779         .func           = bpf_sk_assign,
6780         .gpl_only       = false,
6781         .ret_type       = RET_INTEGER,
6782         .arg1_type      = ARG_PTR_TO_CTX,
6783         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
6784         .arg3_type      = ARG_ANYTHING,
6785 };
6786
6787 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
6788                                     u8 search_kind, const u8 *magic,
6789                                     u8 magic_len, bool *eol)
6790 {
6791         u8 kind, kind_len;
6792
6793         *eol = false;
6794
6795         while (op < opend) {
6796                 kind = op[0];
6797
6798                 if (kind == TCPOPT_EOL) {
6799                         *eol = true;
6800                         return ERR_PTR(-ENOMSG);
6801                 } else if (kind == TCPOPT_NOP) {
6802                         op++;
6803                         continue;
6804                 }
6805
6806                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
6807                         /* Something is wrong in the received header.
6808                          * Follow the TCP stack's tcp_parse_options()
6809                          * and just bail here.
6810                          */
6811                         return ERR_PTR(-EFAULT);
6812
6813                 kind_len = op[1];
6814                 if (search_kind == kind) {
6815                         if (!magic_len)
6816                                 return op;
6817
6818                         if (magic_len > kind_len - 2)
6819                                 return ERR_PTR(-ENOMSG);
6820
6821                         if (!memcmp(&op[2], magic, magic_len))
6822                                 return op;
6823                 }
6824
6825                 op += kind_len;
6826         }
6827
6828         return ERR_PTR(-ENOMSG);
6829 }
6830
6831 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6832            void *, search_res, u32, len, u64, flags)
6833 {
6834         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
6835         const u8 *op, *opend, *magic, *search = search_res;
6836         u8 search_kind, search_len, copy_len, magic_len;
6837         int ret;
6838
6839         /* 2 byte is the minimal option len except TCPOPT_NOP and
6840          * TCPOPT_EOL which are useless for the bpf prog to learn
6841          * and this helper disallow loading them also.
6842          */
6843         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
6844                 return -EINVAL;
6845
6846         search_kind = search[0];
6847         search_len = search[1];
6848
6849         if (search_len > len || search_kind == TCPOPT_NOP ||
6850             search_kind == TCPOPT_EOL)
6851                 return -EINVAL;
6852
6853         if (search_kind == TCPOPT_EXP || search_kind == 253) {
6854                 /* 16 or 32 bit magic.  +2 for kind and kind length */
6855                 if (search_len != 4 && search_len != 6)
6856                         return -EINVAL;
6857                 magic = &search[2];
6858                 magic_len = search_len - 2;
6859         } else {
6860                 if (search_len)
6861                         return -EINVAL;
6862                 magic = NULL;
6863                 magic_len = 0;
6864         }
6865
6866         if (load_syn) {
6867                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
6868                 if (ret < 0)
6869                         return ret;
6870
6871                 opend = op + ret;
6872                 op += sizeof(struct tcphdr);
6873         } else {
6874                 if (!bpf_sock->skb ||
6875                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6876                         /* This bpf_sock->op cannot call this helper */
6877                         return -EPERM;
6878
6879                 opend = bpf_sock->skb_data_end;
6880                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
6881         }
6882
6883         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
6884                                 &eol);
6885         if (IS_ERR(op))
6886                 return PTR_ERR(op);
6887
6888         copy_len = op[1];
6889         ret = copy_len;
6890         if (copy_len > len) {
6891                 ret = -ENOSPC;
6892                 copy_len = len;
6893         }
6894
6895         memcpy(search_res, op, copy_len);
6896         return ret;
6897 }
6898
6899 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
6900         .func           = bpf_sock_ops_load_hdr_opt,
6901         .gpl_only       = false,
6902         .ret_type       = RET_INTEGER,
6903         .arg1_type      = ARG_PTR_TO_CTX,
6904         .arg2_type      = ARG_PTR_TO_MEM,
6905         .arg3_type      = ARG_CONST_SIZE,
6906         .arg4_type      = ARG_ANYTHING,
6907 };
6908
6909 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6910            const void *, from, u32, len, u64, flags)
6911 {
6912         u8 new_kind, new_kind_len, magic_len = 0, *opend;
6913         const u8 *op, *new_op, *magic = NULL;
6914         struct sk_buff *skb;
6915         bool eol;
6916
6917         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
6918                 return -EPERM;
6919
6920         if (len < 2 || flags)
6921                 return -EINVAL;
6922
6923         new_op = from;
6924         new_kind = new_op[0];
6925         new_kind_len = new_op[1];
6926
6927         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
6928             new_kind == TCPOPT_EOL)
6929                 return -EINVAL;
6930
6931         if (new_kind_len > bpf_sock->remaining_opt_len)
6932                 return -ENOSPC;
6933
6934         /* 253 is another experimental kind */
6935         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
6936                 if (new_kind_len < 4)
6937                         return -EINVAL;
6938                 /* Match for the 2 byte magic also.
6939                  * RFC 6994: the magic could be 2 or 4 bytes.
6940                  * Hence, matching by 2 byte only is on the
6941                  * conservative side but it is the right
6942                  * thing to do for the 'search-for-duplication'
6943                  * purpose.
6944                  */
6945                 magic = &new_op[2];
6946                 magic_len = 2;
6947         }
6948
6949         /* Check for duplication */
6950         skb = bpf_sock->skb;
6951         op = skb->data + sizeof(struct tcphdr);
6952         opend = bpf_sock->skb_data_end;
6953
6954         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
6955                                 &eol);
6956         if (!IS_ERR(op))
6957                 return -EEXIST;
6958
6959         if (PTR_ERR(op) != -ENOMSG)
6960                 return PTR_ERR(op);
6961
6962         if (eol)
6963                 /* The option has been ended.  Treat it as no more
6964                  * header option can be written.
6965                  */
6966                 return -ENOSPC;
6967
6968         /* No duplication found.  Store the header option. */
6969         memcpy(opend, from, new_kind_len);
6970
6971         bpf_sock->remaining_opt_len -= new_kind_len;
6972         bpf_sock->skb_data_end += new_kind_len;
6973
6974         return 0;
6975 }
6976
6977 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
6978         .func           = bpf_sock_ops_store_hdr_opt,
6979         .gpl_only       = false,
6980         .ret_type       = RET_INTEGER,
6981         .arg1_type      = ARG_PTR_TO_CTX,
6982         .arg2_type      = ARG_PTR_TO_MEM,
6983         .arg3_type      = ARG_CONST_SIZE,
6984         .arg4_type      = ARG_ANYTHING,
6985 };
6986
6987 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
6988            u32, len, u64, flags)
6989 {
6990         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
6991                 return -EPERM;
6992
6993         if (flags || len < 2)
6994                 return -EINVAL;
6995
6996         if (len > bpf_sock->remaining_opt_len)
6997                 return -ENOSPC;
6998
6999         bpf_sock->remaining_opt_len -= len;
7000
7001         return 0;
7002 }
7003
7004 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7005         .func           = bpf_sock_ops_reserve_hdr_opt,
7006         .gpl_only       = false,
7007         .ret_type       = RET_INTEGER,
7008         .arg1_type      = ARG_PTR_TO_CTX,
7009         .arg2_type      = ARG_ANYTHING,
7010         .arg3_type      = ARG_ANYTHING,
7011 };
7012
7013 #endif /* CONFIG_INET */
7014
7015 bool bpf_helper_changes_pkt_data(void *func)
7016 {
7017         if (func == bpf_skb_vlan_push ||
7018             func == bpf_skb_vlan_pop ||
7019             func == bpf_skb_store_bytes ||
7020             func == bpf_skb_change_proto ||
7021             func == bpf_skb_change_head ||
7022             func == sk_skb_change_head ||
7023             func == bpf_skb_change_tail ||
7024             func == sk_skb_change_tail ||
7025             func == bpf_skb_adjust_room ||
7026             func == sk_skb_adjust_room ||
7027             func == bpf_skb_pull_data ||
7028             func == sk_skb_pull_data ||
7029             func == bpf_clone_redirect ||
7030             func == bpf_l3_csum_replace ||
7031             func == bpf_l4_csum_replace ||
7032             func == bpf_xdp_adjust_head ||
7033             func == bpf_xdp_adjust_meta ||
7034             func == bpf_msg_pull_data ||
7035             func == bpf_msg_push_data ||
7036             func == bpf_msg_pop_data ||
7037             func == bpf_xdp_adjust_tail ||
7038 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7039             func == bpf_lwt_seg6_store_bytes ||
7040             func == bpf_lwt_seg6_adjust_srh ||
7041             func == bpf_lwt_seg6_action ||
7042 #endif
7043 #ifdef CONFIG_INET
7044             func == bpf_sock_ops_store_hdr_opt ||
7045 #endif
7046             func == bpf_lwt_in_push_encap ||
7047             func == bpf_lwt_xmit_push_encap)
7048                 return true;
7049
7050         return false;
7051 }
7052
7053 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7054 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7055
7056 static const struct bpf_func_proto *
7057 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7058 {
7059         switch (func_id) {
7060         /* inet and inet6 sockets are created in a process
7061          * context so there is always a valid uid/gid
7062          */
7063         case BPF_FUNC_get_current_uid_gid:
7064                 return &bpf_get_current_uid_gid_proto;
7065         case BPF_FUNC_get_local_storage:
7066                 return &bpf_get_local_storage_proto;
7067         case BPF_FUNC_get_socket_cookie:
7068                 return &bpf_get_socket_cookie_sock_proto;
7069         case BPF_FUNC_get_netns_cookie:
7070                 return &bpf_get_netns_cookie_sock_proto;
7071         case BPF_FUNC_perf_event_output:
7072                 return &bpf_event_output_data_proto;
7073         case BPF_FUNC_get_current_pid_tgid:
7074                 return &bpf_get_current_pid_tgid_proto;
7075         case BPF_FUNC_get_current_comm:
7076                 return &bpf_get_current_comm_proto;
7077 #ifdef CONFIG_CGROUPS
7078         case BPF_FUNC_get_current_cgroup_id:
7079                 return &bpf_get_current_cgroup_id_proto;
7080         case BPF_FUNC_get_current_ancestor_cgroup_id:
7081                 return &bpf_get_current_ancestor_cgroup_id_proto;
7082 #endif
7083 #ifdef CONFIG_CGROUP_NET_CLASSID
7084         case BPF_FUNC_get_cgroup_classid:
7085                 return &bpf_get_cgroup_classid_curr_proto;
7086 #endif
7087         case BPF_FUNC_sk_storage_get:
7088                 return &bpf_sk_storage_get_cg_sock_proto;
7089         default:
7090                 return bpf_base_func_proto(func_id);
7091         }
7092 }
7093
7094 static const struct bpf_func_proto *
7095 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7096 {
7097         switch (func_id) {
7098         /* inet and inet6 sockets are created in a process
7099          * context so there is always a valid uid/gid
7100          */
7101         case BPF_FUNC_get_current_uid_gid:
7102                 return &bpf_get_current_uid_gid_proto;
7103         case BPF_FUNC_bind:
7104                 switch (prog->expected_attach_type) {
7105                 case BPF_CGROUP_INET4_CONNECT:
7106                 case BPF_CGROUP_INET6_CONNECT:
7107                         return &bpf_bind_proto;
7108                 default:
7109                         return NULL;
7110                 }
7111         case BPF_FUNC_get_socket_cookie:
7112                 return &bpf_get_socket_cookie_sock_addr_proto;
7113         case BPF_FUNC_get_netns_cookie:
7114                 return &bpf_get_netns_cookie_sock_addr_proto;
7115         case BPF_FUNC_get_local_storage:
7116                 return &bpf_get_local_storage_proto;
7117         case BPF_FUNC_perf_event_output:
7118                 return &bpf_event_output_data_proto;
7119         case BPF_FUNC_get_current_pid_tgid:
7120                 return &bpf_get_current_pid_tgid_proto;
7121         case BPF_FUNC_get_current_comm:
7122                 return &bpf_get_current_comm_proto;
7123 #ifdef CONFIG_CGROUPS
7124         case BPF_FUNC_get_current_cgroup_id:
7125                 return &bpf_get_current_cgroup_id_proto;
7126         case BPF_FUNC_get_current_ancestor_cgroup_id:
7127                 return &bpf_get_current_ancestor_cgroup_id_proto;
7128 #endif
7129 #ifdef CONFIG_CGROUP_NET_CLASSID
7130         case BPF_FUNC_get_cgroup_classid:
7131                 return &bpf_get_cgroup_classid_curr_proto;
7132 #endif
7133 #ifdef CONFIG_INET
7134         case BPF_FUNC_sk_lookup_tcp:
7135                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7136         case BPF_FUNC_sk_lookup_udp:
7137                 return &bpf_sock_addr_sk_lookup_udp_proto;
7138         case BPF_FUNC_sk_release:
7139                 return &bpf_sk_release_proto;
7140         case BPF_FUNC_skc_lookup_tcp:
7141                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7142 #endif /* CONFIG_INET */
7143         case BPF_FUNC_sk_storage_get:
7144                 return &bpf_sk_storage_get_proto;
7145         case BPF_FUNC_sk_storage_delete:
7146                 return &bpf_sk_storage_delete_proto;
7147         case BPF_FUNC_setsockopt:
7148                 switch (prog->expected_attach_type) {
7149                 case BPF_CGROUP_INET4_BIND:
7150                 case BPF_CGROUP_INET6_BIND:
7151                 case BPF_CGROUP_INET4_CONNECT:
7152                 case BPF_CGROUP_INET6_CONNECT:
7153                 case BPF_CGROUP_UDP4_RECVMSG:
7154                 case BPF_CGROUP_UDP6_RECVMSG:
7155                 case BPF_CGROUP_UDP4_SENDMSG:
7156                 case BPF_CGROUP_UDP6_SENDMSG:
7157                 case BPF_CGROUP_INET4_GETPEERNAME:
7158                 case BPF_CGROUP_INET6_GETPEERNAME:
7159                 case BPF_CGROUP_INET4_GETSOCKNAME:
7160                 case BPF_CGROUP_INET6_GETSOCKNAME:
7161                         return &bpf_sock_addr_setsockopt_proto;
7162                 default:
7163                         return NULL;
7164                 }
7165         case BPF_FUNC_getsockopt:
7166                 switch (prog->expected_attach_type) {
7167                 case BPF_CGROUP_INET4_BIND:
7168                 case BPF_CGROUP_INET6_BIND:
7169                 case BPF_CGROUP_INET4_CONNECT:
7170                 case BPF_CGROUP_INET6_CONNECT:
7171                 case BPF_CGROUP_UDP4_RECVMSG:
7172                 case BPF_CGROUP_UDP6_RECVMSG:
7173                 case BPF_CGROUP_UDP4_SENDMSG:
7174                 case BPF_CGROUP_UDP6_SENDMSG:
7175                 case BPF_CGROUP_INET4_GETPEERNAME:
7176                 case BPF_CGROUP_INET6_GETPEERNAME:
7177                 case BPF_CGROUP_INET4_GETSOCKNAME:
7178                 case BPF_CGROUP_INET6_GETSOCKNAME:
7179                         return &bpf_sock_addr_getsockopt_proto;
7180                 default:
7181                         return NULL;
7182                 }
7183         default:
7184                 return bpf_sk_base_func_proto(func_id);
7185         }
7186 }
7187
7188 static const struct bpf_func_proto *
7189 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7190 {
7191         switch (func_id) {
7192         case BPF_FUNC_skb_load_bytes:
7193                 return &bpf_skb_load_bytes_proto;
7194         case BPF_FUNC_skb_load_bytes_relative:
7195                 return &bpf_skb_load_bytes_relative_proto;
7196         case BPF_FUNC_get_socket_cookie:
7197                 return &bpf_get_socket_cookie_proto;
7198         case BPF_FUNC_get_socket_uid:
7199                 return &bpf_get_socket_uid_proto;
7200         case BPF_FUNC_perf_event_output:
7201                 return &bpf_skb_event_output_proto;
7202         default:
7203                 return bpf_sk_base_func_proto(func_id);
7204         }
7205 }
7206
7207 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7208 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7209
7210 static const struct bpf_func_proto *
7211 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7212 {
7213         switch (func_id) {
7214         case BPF_FUNC_get_local_storage:
7215                 return &bpf_get_local_storage_proto;
7216         case BPF_FUNC_sk_fullsock:
7217                 return &bpf_sk_fullsock_proto;
7218         case BPF_FUNC_sk_storage_get:
7219                 return &bpf_sk_storage_get_proto;
7220         case BPF_FUNC_sk_storage_delete:
7221                 return &bpf_sk_storage_delete_proto;
7222         case BPF_FUNC_perf_event_output:
7223                 return &bpf_skb_event_output_proto;
7224 #ifdef CONFIG_SOCK_CGROUP_DATA
7225         case BPF_FUNC_skb_cgroup_id:
7226                 return &bpf_skb_cgroup_id_proto;
7227         case BPF_FUNC_skb_ancestor_cgroup_id:
7228                 return &bpf_skb_ancestor_cgroup_id_proto;
7229         case BPF_FUNC_sk_cgroup_id:
7230                 return &bpf_sk_cgroup_id_proto;
7231         case BPF_FUNC_sk_ancestor_cgroup_id:
7232                 return &bpf_sk_ancestor_cgroup_id_proto;
7233 #endif
7234 #ifdef CONFIG_INET
7235         case BPF_FUNC_sk_lookup_tcp:
7236                 return &bpf_sk_lookup_tcp_proto;
7237         case BPF_FUNC_sk_lookup_udp:
7238                 return &bpf_sk_lookup_udp_proto;
7239         case BPF_FUNC_sk_release:
7240                 return &bpf_sk_release_proto;
7241         case BPF_FUNC_skc_lookup_tcp:
7242                 return &bpf_skc_lookup_tcp_proto;
7243         case BPF_FUNC_tcp_sock:
7244                 return &bpf_tcp_sock_proto;
7245         case BPF_FUNC_get_listener_sock:
7246                 return &bpf_get_listener_sock_proto;
7247         case BPF_FUNC_skb_ecn_set_ce:
7248                 return &bpf_skb_ecn_set_ce_proto;
7249 #endif
7250         default:
7251                 return sk_filter_func_proto(func_id, prog);
7252         }
7253 }
7254
7255 static const struct bpf_func_proto *
7256 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7257 {
7258         switch (func_id) {
7259         case BPF_FUNC_skb_store_bytes:
7260                 return &bpf_skb_store_bytes_proto;
7261         case BPF_FUNC_skb_load_bytes:
7262                 return &bpf_skb_load_bytes_proto;
7263         case BPF_FUNC_skb_load_bytes_relative:
7264                 return &bpf_skb_load_bytes_relative_proto;
7265         case BPF_FUNC_skb_pull_data:
7266                 return &bpf_skb_pull_data_proto;
7267         case BPF_FUNC_csum_diff:
7268                 return &bpf_csum_diff_proto;
7269         case BPF_FUNC_csum_update:
7270                 return &bpf_csum_update_proto;
7271         case BPF_FUNC_csum_level:
7272                 return &bpf_csum_level_proto;
7273         case BPF_FUNC_l3_csum_replace:
7274                 return &bpf_l3_csum_replace_proto;
7275         case BPF_FUNC_l4_csum_replace:
7276                 return &bpf_l4_csum_replace_proto;
7277         case BPF_FUNC_clone_redirect:
7278                 return &bpf_clone_redirect_proto;
7279         case BPF_FUNC_get_cgroup_classid:
7280                 return &bpf_get_cgroup_classid_proto;
7281         case BPF_FUNC_skb_vlan_push:
7282                 return &bpf_skb_vlan_push_proto;
7283         case BPF_FUNC_skb_vlan_pop:
7284                 return &bpf_skb_vlan_pop_proto;
7285         case BPF_FUNC_skb_change_proto:
7286                 return &bpf_skb_change_proto_proto;
7287         case BPF_FUNC_skb_change_type:
7288                 return &bpf_skb_change_type_proto;
7289         case BPF_FUNC_skb_adjust_room:
7290                 return &bpf_skb_adjust_room_proto;
7291         case BPF_FUNC_skb_change_tail:
7292                 return &bpf_skb_change_tail_proto;
7293         case BPF_FUNC_skb_change_head:
7294                 return &bpf_skb_change_head_proto;
7295         case BPF_FUNC_skb_get_tunnel_key:
7296                 return &bpf_skb_get_tunnel_key_proto;
7297         case BPF_FUNC_skb_set_tunnel_key:
7298                 return bpf_get_skb_set_tunnel_proto(func_id);
7299         case BPF_FUNC_skb_get_tunnel_opt:
7300                 return &bpf_skb_get_tunnel_opt_proto;
7301         case BPF_FUNC_skb_set_tunnel_opt:
7302                 return bpf_get_skb_set_tunnel_proto(func_id);
7303         case BPF_FUNC_redirect:
7304                 return &bpf_redirect_proto;
7305         case BPF_FUNC_redirect_neigh:
7306                 return &bpf_redirect_neigh_proto;
7307         case BPF_FUNC_redirect_peer:
7308                 return &bpf_redirect_peer_proto;
7309         case BPF_FUNC_get_route_realm:
7310                 return &bpf_get_route_realm_proto;
7311         case BPF_FUNC_get_hash_recalc:
7312                 return &bpf_get_hash_recalc_proto;
7313         case BPF_FUNC_set_hash_invalid:
7314                 return &bpf_set_hash_invalid_proto;
7315         case BPF_FUNC_set_hash:
7316                 return &bpf_set_hash_proto;
7317         case BPF_FUNC_perf_event_output:
7318                 return &bpf_skb_event_output_proto;
7319         case BPF_FUNC_get_smp_processor_id:
7320                 return &bpf_get_smp_processor_id_proto;
7321         case BPF_FUNC_skb_under_cgroup:
7322                 return &bpf_skb_under_cgroup_proto;
7323         case BPF_FUNC_get_socket_cookie:
7324                 return &bpf_get_socket_cookie_proto;
7325         case BPF_FUNC_get_socket_uid:
7326                 return &bpf_get_socket_uid_proto;
7327         case BPF_FUNC_fib_lookup:
7328                 return &bpf_skb_fib_lookup_proto;
7329         case BPF_FUNC_check_mtu:
7330                 return &bpf_skb_check_mtu_proto;
7331         case BPF_FUNC_sk_fullsock:
7332                 return &bpf_sk_fullsock_proto;
7333         case BPF_FUNC_sk_storage_get:
7334                 return &bpf_sk_storage_get_proto;
7335         case BPF_FUNC_sk_storage_delete:
7336                 return &bpf_sk_storage_delete_proto;
7337 #ifdef CONFIG_XFRM
7338         case BPF_FUNC_skb_get_xfrm_state:
7339                 return &bpf_skb_get_xfrm_state_proto;
7340 #endif
7341 #ifdef CONFIG_CGROUP_NET_CLASSID
7342         case BPF_FUNC_skb_cgroup_classid:
7343                 return &bpf_skb_cgroup_classid_proto;
7344 #endif
7345 #ifdef CONFIG_SOCK_CGROUP_DATA
7346         case BPF_FUNC_skb_cgroup_id:
7347                 return &bpf_skb_cgroup_id_proto;
7348         case BPF_FUNC_skb_ancestor_cgroup_id:
7349                 return &bpf_skb_ancestor_cgroup_id_proto;
7350 #endif
7351 #ifdef CONFIG_INET
7352         case BPF_FUNC_sk_lookup_tcp:
7353                 return &bpf_sk_lookup_tcp_proto;
7354         case BPF_FUNC_sk_lookup_udp:
7355                 return &bpf_sk_lookup_udp_proto;
7356         case BPF_FUNC_sk_release:
7357                 return &bpf_sk_release_proto;
7358         case BPF_FUNC_tcp_sock:
7359                 return &bpf_tcp_sock_proto;
7360         case BPF_FUNC_get_listener_sock:
7361                 return &bpf_get_listener_sock_proto;
7362         case BPF_FUNC_skc_lookup_tcp:
7363                 return &bpf_skc_lookup_tcp_proto;
7364         case BPF_FUNC_tcp_check_syncookie:
7365                 return &bpf_tcp_check_syncookie_proto;
7366         case BPF_FUNC_skb_ecn_set_ce:
7367                 return &bpf_skb_ecn_set_ce_proto;
7368         case BPF_FUNC_tcp_gen_syncookie:
7369                 return &bpf_tcp_gen_syncookie_proto;
7370         case BPF_FUNC_sk_assign:
7371                 return &bpf_sk_assign_proto;
7372 #endif
7373         default:
7374                 return bpf_sk_base_func_proto(func_id);
7375         }
7376 }
7377
7378 static const struct bpf_func_proto *
7379 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7380 {
7381         switch (func_id) {
7382         case BPF_FUNC_perf_event_output:
7383                 return &bpf_xdp_event_output_proto;
7384         case BPF_FUNC_get_smp_processor_id:
7385                 return &bpf_get_smp_processor_id_proto;
7386         case BPF_FUNC_csum_diff:
7387                 return &bpf_csum_diff_proto;
7388         case BPF_FUNC_xdp_adjust_head:
7389                 return &bpf_xdp_adjust_head_proto;
7390         case BPF_FUNC_xdp_adjust_meta:
7391                 return &bpf_xdp_adjust_meta_proto;
7392         case BPF_FUNC_redirect:
7393                 return &bpf_xdp_redirect_proto;
7394         case BPF_FUNC_redirect_map:
7395                 return &bpf_xdp_redirect_map_proto;
7396         case BPF_FUNC_xdp_adjust_tail:
7397                 return &bpf_xdp_adjust_tail_proto;
7398         case BPF_FUNC_fib_lookup:
7399                 return &bpf_xdp_fib_lookup_proto;
7400         case BPF_FUNC_check_mtu:
7401                 return &bpf_xdp_check_mtu_proto;
7402 #ifdef CONFIG_INET
7403         case BPF_FUNC_sk_lookup_udp:
7404                 return &bpf_xdp_sk_lookup_udp_proto;
7405         case BPF_FUNC_sk_lookup_tcp:
7406                 return &bpf_xdp_sk_lookup_tcp_proto;
7407         case BPF_FUNC_sk_release:
7408                 return &bpf_sk_release_proto;
7409         case BPF_FUNC_skc_lookup_tcp:
7410                 return &bpf_xdp_skc_lookup_tcp_proto;
7411         case BPF_FUNC_tcp_check_syncookie:
7412                 return &bpf_tcp_check_syncookie_proto;
7413         case BPF_FUNC_tcp_gen_syncookie:
7414                 return &bpf_tcp_gen_syncookie_proto;
7415 #endif
7416         default:
7417                 return bpf_sk_base_func_proto(func_id);
7418         }
7419 }
7420
7421 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7422 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7423
7424 static const struct bpf_func_proto *
7425 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7426 {
7427         switch (func_id) {
7428         case BPF_FUNC_setsockopt:
7429                 return &bpf_sock_ops_setsockopt_proto;
7430         case BPF_FUNC_getsockopt:
7431                 return &bpf_sock_ops_getsockopt_proto;
7432         case BPF_FUNC_sock_ops_cb_flags_set:
7433                 return &bpf_sock_ops_cb_flags_set_proto;
7434         case BPF_FUNC_sock_map_update:
7435                 return &bpf_sock_map_update_proto;
7436         case BPF_FUNC_sock_hash_update:
7437                 return &bpf_sock_hash_update_proto;
7438         case BPF_FUNC_get_socket_cookie:
7439                 return &bpf_get_socket_cookie_sock_ops_proto;
7440         case BPF_FUNC_get_local_storage:
7441                 return &bpf_get_local_storage_proto;
7442         case BPF_FUNC_perf_event_output:
7443                 return &bpf_event_output_data_proto;
7444         case BPF_FUNC_sk_storage_get:
7445                 return &bpf_sk_storage_get_proto;
7446         case BPF_FUNC_sk_storage_delete:
7447                 return &bpf_sk_storage_delete_proto;
7448 #ifdef CONFIG_INET
7449         case BPF_FUNC_load_hdr_opt:
7450                 return &bpf_sock_ops_load_hdr_opt_proto;
7451         case BPF_FUNC_store_hdr_opt:
7452                 return &bpf_sock_ops_store_hdr_opt_proto;
7453         case BPF_FUNC_reserve_hdr_opt:
7454                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7455         case BPF_FUNC_tcp_sock:
7456                 return &bpf_tcp_sock_proto;
7457 #endif /* CONFIG_INET */
7458         default:
7459                 return bpf_sk_base_func_proto(func_id);
7460         }
7461 }
7462
7463 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7464 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7465
7466 static const struct bpf_func_proto *
7467 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7468 {
7469         switch (func_id) {
7470         case BPF_FUNC_msg_redirect_map:
7471                 return &bpf_msg_redirect_map_proto;
7472         case BPF_FUNC_msg_redirect_hash:
7473                 return &bpf_msg_redirect_hash_proto;
7474         case BPF_FUNC_msg_apply_bytes:
7475                 return &bpf_msg_apply_bytes_proto;
7476         case BPF_FUNC_msg_cork_bytes:
7477                 return &bpf_msg_cork_bytes_proto;
7478         case BPF_FUNC_msg_pull_data:
7479                 return &bpf_msg_pull_data_proto;
7480         case BPF_FUNC_msg_push_data:
7481                 return &bpf_msg_push_data_proto;
7482         case BPF_FUNC_msg_pop_data:
7483                 return &bpf_msg_pop_data_proto;
7484         case BPF_FUNC_perf_event_output:
7485                 return &bpf_event_output_data_proto;
7486         case BPF_FUNC_get_current_uid_gid:
7487                 return &bpf_get_current_uid_gid_proto;
7488         case BPF_FUNC_get_current_pid_tgid:
7489                 return &bpf_get_current_pid_tgid_proto;
7490         case BPF_FUNC_sk_storage_get:
7491                 return &bpf_sk_storage_get_proto;
7492         case BPF_FUNC_sk_storage_delete:
7493                 return &bpf_sk_storage_delete_proto;
7494 #ifdef CONFIG_CGROUPS
7495         case BPF_FUNC_get_current_cgroup_id:
7496                 return &bpf_get_current_cgroup_id_proto;
7497         case BPF_FUNC_get_current_ancestor_cgroup_id:
7498                 return &bpf_get_current_ancestor_cgroup_id_proto;
7499 #endif
7500 #ifdef CONFIG_CGROUP_NET_CLASSID
7501         case BPF_FUNC_get_cgroup_classid:
7502                 return &bpf_get_cgroup_classid_curr_proto;
7503 #endif
7504         default:
7505                 return bpf_sk_base_func_proto(func_id);
7506         }
7507 }
7508
7509 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
7510 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
7511
7512 static const struct bpf_func_proto *
7513 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7514 {
7515         switch (func_id) {
7516         case BPF_FUNC_skb_store_bytes:
7517                 return &bpf_skb_store_bytes_proto;
7518         case BPF_FUNC_skb_load_bytes:
7519                 return &bpf_skb_load_bytes_proto;
7520         case BPF_FUNC_skb_pull_data:
7521                 return &sk_skb_pull_data_proto;
7522         case BPF_FUNC_skb_change_tail:
7523                 return &sk_skb_change_tail_proto;
7524         case BPF_FUNC_skb_change_head:
7525                 return &sk_skb_change_head_proto;
7526         case BPF_FUNC_skb_adjust_room:
7527                 return &sk_skb_adjust_room_proto;
7528         case BPF_FUNC_get_socket_cookie:
7529                 return &bpf_get_socket_cookie_proto;
7530         case BPF_FUNC_get_socket_uid:
7531                 return &bpf_get_socket_uid_proto;
7532         case BPF_FUNC_sk_redirect_map:
7533                 return &bpf_sk_redirect_map_proto;
7534         case BPF_FUNC_sk_redirect_hash:
7535                 return &bpf_sk_redirect_hash_proto;
7536         case BPF_FUNC_perf_event_output:
7537                 return &bpf_skb_event_output_proto;
7538 #ifdef CONFIG_INET
7539         case BPF_FUNC_sk_lookup_tcp:
7540                 return &bpf_sk_lookup_tcp_proto;
7541         case BPF_FUNC_sk_lookup_udp:
7542                 return &bpf_sk_lookup_udp_proto;
7543         case BPF_FUNC_sk_release:
7544                 return &bpf_sk_release_proto;
7545         case BPF_FUNC_skc_lookup_tcp:
7546                 return &bpf_skc_lookup_tcp_proto;
7547 #endif
7548         default:
7549                 return bpf_sk_base_func_proto(func_id);
7550         }
7551 }
7552
7553 static const struct bpf_func_proto *
7554 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7555 {
7556         switch (func_id) {
7557         case BPF_FUNC_skb_load_bytes:
7558                 return &bpf_flow_dissector_load_bytes_proto;
7559         default:
7560                 return bpf_sk_base_func_proto(func_id);
7561         }
7562 }
7563
7564 static const struct bpf_func_proto *
7565 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7566 {
7567         switch (func_id) {
7568         case BPF_FUNC_skb_load_bytes:
7569                 return &bpf_skb_load_bytes_proto;
7570         case BPF_FUNC_skb_pull_data:
7571                 return &bpf_skb_pull_data_proto;
7572         case BPF_FUNC_csum_diff:
7573                 return &bpf_csum_diff_proto;
7574         case BPF_FUNC_get_cgroup_classid:
7575                 return &bpf_get_cgroup_classid_proto;
7576         case BPF_FUNC_get_route_realm:
7577                 return &bpf_get_route_realm_proto;
7578         case BPF_FUNC_get_hash_recalc:
7579                 return &bpf_get_hash_recalc_proto;
7580         case BPF_FUNC_perf_event_output:
7581                 return &bpf_skb_event_output_proto;
7582         case BPF_FUNC_get_smp_processor_id:
7583                 return &bpf_get_smp_processor_id_proto;
7584         case BPF_FUNC_skb_under_cgroup:
7585                 return &bpf_skb_under_cgroup_proto;
7586         default:
7587                 return bpf_sk_base_func_proto(func_id);
7588         }
7589 }
7590
7591 static const struct bpf_func_proto *
7592 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7593 {
7594         switch (func_id) {
7595         case BPF_FUNC_lwt_push_encap:
7596                 return &bpf_lwt_in_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_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7604 {
7605         switch (func_id) {
7606         case BPF_FUNC_skb_get_tunnel_key:
7607                 return &bpf_skb_get_tunnel_key_proto;
7608         case BPF_FUNC_skb_set_tunnel_key:
7609                 return bpf_get_skb_set_tunnel_proto(func_id);
7610         case BPF_FUNC_skb_get_tunnel_opt:
7611                 return &bpf_skb_get_tunnel_opt_proto;
7612         case BPF_FUNC_skb_set_tunnel_opt:
7613                 return bpf_get_skb_set_tunnel_proto(func_id);
7614         case BPF_FUNC_redirect:
7615                 return &bpf_redirect_proto;
7616         case BPF_FUNC_clone_redirect:
7617                 return &bpf_clone_redirect_proto;
7618         case BPF_FUNC_skb_change_tail:
7619                 return &bpf_skb_change_tail_proto;
7620         case BPF_FUNC_skb_change_head:
7621                 return &bpf_skb_change_head_proto;
7622         case BPF_FUNC_skb_store_bytes:
7623                 return &bpf_skb_store_bytes_proto;
7624         case BPF_FUNC_csum_update:
7625                 return &bpf_csum_update_proto;
7626         case BPF_FUNC_csum_level:
7627                 return &bpf_csum_level_proto;
7628         case BPF_FUNC_l3_csum_replace:
7629                 return &bpf_l3_csum_replace_proto;
7630         case BPF_FUNC_l4_csum_replace:
7631                 return &bpf_l4_csum_replace_proto;
7632         case BPF_FUNC_set_hash_invalid:
7633                 return &bpf_set_hash_invalid_proto;
7634         case BPF_FUNC_lwt_push_encap:
7635                 return &bpf_lwt_xmit_push_encap_proto;
7636         default:
7637                 return lwt_out_func_proto(func_id, prog);
7638         }
7639 }
7640
7641 static const struct bpf_func_proto *
7642 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7643 {
7644         switch (func_id) {
7645 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7646         case BPF_FUNC_lwt_seg6_store_bytes:
7647                 return &bpf_lwt_seg6_store_bytes_proto;
7648         case BPF_FUNC_lwt_seg6_action:
7649                 return &bpf_lwt_seg6_action_proto;
7650         case BPF_FUNC_lwt_seg6_adjust_srh:
7651                 return &bpf_lwt_seg6_adjust_srh_proto;
7652 #endif
7653         default:
7654                 return lwt_out_func_proto(func_id, prog);
7655         }
7656 }
7657
7658 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
7659                                     const struct bpf_prog *prog,
7660                                     struct bpf_insn_access_aux *info)
7661 {
7662         const int size_default = sizeof(__u32);
7663
7664         if (off < 0 || off >= sizeof(struct __sk_buff))
7665                 return false;
7666
7667         /* The verifier guarantees that size > 0. */
7668         if (off % size != 0)
7669                 return false;
7670
7671         switch (off) {
7672         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7673                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
7674                         return false;
7675                 break;
7676         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
7677         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
7678         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
7679         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
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                 if (size != size_default)
7684                         return false;
7685                 break;
7686         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
7687                 return false;
7688         case bpf_ctx_range(struct __sk_buff, tstamp):
7689                 if (size != sizeof(__u64))
7690                         return false;
7691                 break;
7692         case offsetof(struct __sk_buff, sk):
7693                 if (type == BPF_WRITE || size != sizeof(__u64))
7694                         return false;
7695                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
7696                 break;
7697         default:
7698                 /* Only narrow read access allowed for now. */
7699                 if (type == BPF_WRITE) {
7700                         if (size != size_default)
7701                                 return false;
7702                 } else {
7703                         bpf_ctx_record_field_size(info, size_default);
7704                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
7705                                 return false;
7706                 }
7707         }
7708
7709         return true;
7710 }
7711
7712 static bool sk_filter_is_valid_access(int off, int size,
7713                                       enum bpf_access_type type,
7714                                       const struct bpf_prog *prog,
7715                                       struct bpf_insn_access_aux *info)
7716 {
7717         switch (off) {
7718         case bpf_ctx_range(struct __sk_buff, tc_classid):
7719         case bpf_ctx_range(struct __sk_buff, data):
7720         case bpf_ctx_range(struct __sk_buff, data_meta):
7721         case bpf_ctx_range(struct __sk_buff, data_end):
7722         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7723         case bpf_ctx_range(struct __sk_buff, tstamp):
7724         case bpf_ctx_range(struct __sk_buff, wire_len):
7725                 return false;
7726         }
7727
7728         if (type == BPF_WRITE) {
7729                 switch (off) {
7730                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7731                         break;
7732                 default:
7733                         return false;
7734                 }
7735         }
7736
7737         return bpf_skb_is_valid_access(off, size, type, prog, info);
7738 }
7739
7740 static bool cg_skb_is_valid_access(int off, int size,
7741                                    enum bpf_access_type type,
7742                                    const struct bpf_prog *prog,
7743                                    struct bpf_insn_access_aux *info)
7744 {
7745         switch (off) {
7746         case bpf_ctx_range(struct __sk_buff, tc_classid):
7747         case bpf_ctx_range(struct __sk_buff, data_meta):
7748         case bpf_ctx_range(struct __sk_buff, wire_len):
7749                 return false;
7750         case bpf_ctx_range(struct __sk_buff, data):
7751         case bpf_ctx_range(struct __sk_buff, data_end):
7752                 if (!bpf_capable())
7753                         return false;
7754                 break;
7755         }
7756
7757         if (type == BPF_WRITE) {
7758                 switch (off) {
7759                 case bpf_ctx_range(struct __sk_buff, mark):
7760                 case bpf_ctx_range(struct __sk_buff, priority):
7761                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7762                         break;
7763                 case bpf_ctx_range(struct __sk_buff, tstamp):
7764                         if (!bpf_capable())
7765                                 return false;
7766                         break;
7767                 default:
7768                         return false;
7769                 }
7770         }
7771
7772         switch (off) {
7773         case bpf_ctx_range(struct __sk_buff, data):
7774                 info->reg_type = PTR_TO_PACKET;
7775                 break;
7776         case bpf_ctx_range(struct __sk_buff, data_end):
7777                 info->reg_type = PTR_TO_PACKET_END;
7778                 break;
7779         }
7780
7781         return bpf_skb_is_valid_access(off, size, type, prog, info);
7782 }
7783
7784 static bool lwt_is_valid_access(int off, int size,
7785                                 enum bpf_access_type type,
7786                                 const struct bpf_prog *prog,
7787                                 struct bpf_insn_access_aux *info)
7788 {
7789         switch (off) {
7790         case bpf_ctx_range(struct __sk_buff, tc_classid):
7791         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
7792         case bpf_ctx_range(struct __sk_buff, data_meta):
7793         case bpf_ctx_range(struct __sk_buff, tstamp):
7794         case bpf_ctx_range(struct __sk_buff, wire_len):
7795                 return false;
7796         }
7797
7798         if (type == BPF_WRITE) {
7799                 switch (off) {
7800                 case bpf_ctx_range(struct __sk_buff, mark):
7801                 case bpf_ctx_range(struct __sk_buff, priority):
7802                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
7803                         break;
7804                 default:
7805                         return false;
7806                 }
7807         }
7808
7809         switch (off) {
7810         case bpf_ctx_range(struct __sk_buff, data):
7811                 info->reg_type = PTR_TO_PACKET;
7812                 break;
7813         case bpf_ctx_range(struct __sk_buff, data_end):
7814                 info->reg_type = PTR_TO_PACKET_END;
7815                 break;
7816         }
7817
7818         return bpf_skb_is_valid_access(off, size, type, prog, info);
7819 }
7820
7821 /* Attach type specific accesses */
7822 static bool __sock_filter_check_attach_type(int off,
7823                                             enum bpf_access_type access_type,
7824                                             enum bpf_attach_type attach_type)
7825 {
7826         switch (off) {
7827         case offsetof(struct bpf_sock, bound_dev_if):
7828         case offsetof(struct bpf_sock, mark):
7829         case offsetof(struct bpf_sock, priority):
7830                 switch (attach_type) {
7831                 case BPF_CGROUP_INET_SOCK_CREATE:
7832                 case BPF_CGROUP_INET_SOCK_RELEASE:
7833                         goto full_access;
7834                 default:
7835                         return false;
7836                 }
7837         case bpf_ctx_range(struct bpf_sock, src_ip4):
7838                 switch (attach_type) {
7839                 case BPF_CGROUP_INET4_POST_BIND:
7840                         goto read_only;
7841                 default:
7842                         return false;
7843                 }
7844         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7845                 switch (attach_type) {
7846                 case BPF_CGROUP_INET6_POST_BIND:
7847                         goto read_only;
7848                 default:
7849                         return false;
7850                 }
7851         case bpf_ctx_range(struct bpf_sock, src_port):
7852                 switch (attach_type) {
7853                 case BPF_CGROUP_INET4_POST_BIND:
7854                 case BPF_CGROUP_INET6_POST_BIND:
7855                         goto read_only;
7856                 default:
7857                         return false;
7858                 }
7859         }
7860 read_only:
7861         return access_type == BPF_READ;
7862 full_access:
7863         return true;
7864 }
7865
7866 bool bpf_sock_common_is_valid_access(int off, int size,
7867                                      enum bpf_access_type type,
7868                                      struct bpf_insn_access_aux *info)
7869 {
7870         switch (off) {
7871         case bpf_ctx_range_till(struct bpf_sock, type, priority):
7872                 return false;
7873         default:
7874                 return bpf_sock_is_valid_access(off, size, type, info);
7875         }
7876 }
7877
7878 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
7879                               struct bpf_insn_access_aux *info)
7880 {
7881         const int size_default = sizeof(__u32);
7882
7883         if (off < 0 || off >= sizeof(struct bpf_sock))
7884                 return false;
7885         if (off % size != 0)
7886                 return false;
7887
7888         switch (off) {
7889         case offsetof(struct bpf_sock, state):
7890         case offsetof(struct bpf_sock, family):
7891         case offsetof(struct bpf_sock, type):
7892         case offsetof(struct bpf_sock, protocol):
7893         case offsetof(struct bpf_sock, dst_port):
7894         case offsetof(struct bpf_sock, src_port):
7895         case offsetof(struct bpf_sock, rx_queue_mapping):
7896         case bpf_ctx_range(struct bpf_sock, src_ip4):
7897         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
7898         case bpf_ctx_range(struct bpf_sock, dst_ip4):
7899         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
7900                 bpf_ctx_record_field_size(info, size_default);
7901                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7902         }
7903
7904         return size == size_default;
7905 }
7906
7907 static bool sock_filter_is_valid_access(int off, int size,
7908                                         enum bpf_access_type type,
7909                                         const struct bpf_prog *prog,
7910                                         struct bpf_insn_access_aux *info)
7911 {
7912         if (!bpf_sock_is_valid_access(off, size, type, info))
7913                 return false;
7914         return __sock_filter_check_attach_type(off, type,
7915                                                prog->expected_attach_type);
7916 }
7917
7918 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
7919                              const struct bpf_prog *prog)
7920 {
7921         /* Neither direct read nor direct write requires any preliminary
7922          * action.
7923          */
7924         return 0;
7925 }
7926
7927 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
7928                                 const struct bpf_prog *prog, int drop_verdict)
7929 {
7930         struct bpf_insn *insn = insn_buf;
7931
7932         if (!direct_write)
7933                 return 0;
7934
7935         /* if (!skb->cloned)
7936          *       goto start;
7937          *
7938          * (Fast-path, otherwise approximation that we might be
7939          *  a clone, do the rest in helper.)
7940          */
7941         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
7942         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
7943         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
7944
7945         /* ret = bpf_skb_pull_data(skb, 0); */
7946         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
7947         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
7948         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
7949                                BPF_FUNC_skb_pull_data);
7950         /* if (!ret)
7951          *      goto restore;
7952          * return TC_ACT_SHOT;
7953          */
7954         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
7955         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
7956         *insn++ = BPF_EXIT_INSN();
7957
7958         /* restore: */
7959         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
7960         /* start: */
7961         *insn++ = prog->insnsi[0];
7962
7963         return insn - insn_buf;
7964 }
7965
7966 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
7967                           struct bpf_insn *insn_buf)
7968 {
7969         bool indirect = BPF_MODE(orig->code) == BPF_IND;
7970         struct bpf_insn *insn = insn_buf;
7971
7972         if (!indirect) {
7973                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
7974         } else {
7975                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
7976                 if (orig->imm)
7977                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
7978         }
7979         /* We're guaranteed here that CTX is in R6. */
7980         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
7981
7982         switch (BPF_SIZE(orig->code)) {
7983         case BPF_B:
7984                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
7985                 break;
7986         case BPF_H:
7987                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
7988                 break;
7989         case BPF_W:
7990                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
7991                 break;
7992         }
7993
7994         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
7995         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
7996         *insn++ = BPF_EXIT_INSN();
7997
7998         return insn - insn_buf;
7999 }
8000
8001 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8002                                const struct bpf_prog *prog)
8003 {
8004         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8005 }
8006
8007 static bool tc_cls_act_is_valid_access(int off, int size,
8008                                        enum bpf_access_type type,
8009                                        const struct bpf_prog *prog,
8010                                        struct bpf_insn_access_aux *info)
8011 {
8012         if (type == BPF_WRITE) {
8013                 switch (off) {
8014                 case bpf_ctx_range(struct __sk_buff, mark):
8015                 case bpf_ctx_range(struct __sk_buff, tc_index):
8016                 case bpf_ctx_range(struct __sk_buff, priority):
8017                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8018                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8019                 case bpf_ctx_range(struct __sk_buff, tstamp):
8020                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8021                         break;
8022                 default:
8023                         return false;
8024                 }
8025         }
8026
8027         switch (off) {
8028         case bpf_ctx_range(struct __sk_buff, data):
8029                 info->reg_type = PTR_TO_PACKET;
8030                 break;
8031         case bpf_ctx_range(struct __sk_buff, data_meta):
8032                 info->reg_type = PTR_TO_PACKET_META;
8033                 break;
8034         case bpf_ctx_range(struct __sk_buff, data_end):
8035                 info->reg_type = PTR_TO_PACKET_END;
8036                 break;
8037         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8038                 return false;
8039         }
8040
8041         return bpf_skb_is_valid_access(off, size, type, prog, info);
8042 }
8043
8044 static bool __is_valid_xdp_access(int off, int size)
8045 {
8046         if (off < 0 || off >= sizeof(struct xdp_md))
8047                 return false;
8048         if (off % size != 0)
8049                 return false;
8050         if (size != sizeof(__u32))
8051                 return false;
8052
8053         return true;
8054 }
8055
8056 static bool xdp_is_valid_access(int off, int size,
8057                                 enum bpf_access_type type,
8058                                 const struct bpf_prog *prog,
8059                                 struct bpf_insn_access_aux *info)
8060 {
8061         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8062                 switch (off) {
8063                 case offsetof(struct xdp_md, egress_ifindex):
8064                         return false;
8065                 }
8066         }
8067
8068         if (type == BPF_WRITE) {
8069                 if (bpf_prog_is_dev_bound(prog->aux)) {
8070                         switch (off) {
8071                         case offsetof(struct xdp_md, rx_queue_index):
8072                                 return __is_valid_xdp_access(off, size);
8073                         }
8074                 }
8075                 return false;
8076         }
8077
8078         switch (off) {
8079         case offsetof(struct xdp_md, data):
8080                 info->reg_type = PTR_TO_PACKET;
8081                 break;
8082         case offsetof(struct xdp_md, data_meta):
8083                 info->reg_type = PTR_TO_PACKET_META;
8084                 break;
8085         case offsetof(struct xdp_md, data_end):
8086                 info->reg_type = PTR_TO_PACKET_END;
8087                 break;
8088         }
8089
8090         return __is_valid_xdp_access(off, size);
8091 }
8092
8093 void bpf_warn_invalid_xdp_action(u32 act)
8094 {
8095         const u32 act_max = XDP_REDIRECT;
8096
8097         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
8098                   act > act_max ? "Illegal" : "Driver unsupported",
8099                   act);
8100 }
8101 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8102
8103 static bool sock_addr_is_valid_access(int off, int size,
8104                                       enum bpf_access_type type,
8105                                       const struct bpf_prog *prog,
8106                                       struct bpf_insn_access_aux *info)
8107 {
8108         const int size_default = sizeof(__u32);
8109
8110         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8111                 return false;
8112         if (off % size != 0)
8113                 return false;
8114
8115         /* Disallow access to IPv6 fields from IPv4 contex and vise
8116          * versa.
8117          */
8118         switch (off) {
8119         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8120                 switch (prog->expected_attach_type) {
8121                 case BPF_CGROUP_INET4_BIND:
8122                 case BPF_CGROUP_INET4_CONNECT:
8123                 case BPF_CGROUP_INET4_GETPEERNAME:
8124                 case BPF_CGROUP_INET4_GETSOCKNAME:
8125                 case BPF_CGROUP_UDP4_SENDMSG:
8126                 case BPF_CGROUP_UDP4_RECVMSG:
8127                         break;
8128                 default:
8129                         return false;
8130                 }
8131                 break;
8132         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8133                 switch (prog->expected_attach_type) {
8134                 case BPF_CGROUP_INET6_BIND:
8135                 case BPF_CGROUP_INET6_CONNECT:
8136                 case BPF_CGROUP_INET6_GETPEERNAME:
8137                 case BPF_CGROUP_INET6_GETSOCKNAME:
8138                 case BPF_CGROUP_UDP6_SENDMSG:
8139                 case BPF_CGROUP_UDP6_RECVMSG:
8140                         break;
8141                 default:
8142                         return false;
8143                 }
8144                 break;
8145         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8146                 switch (prog->expected_attach_type) {
8147                 case BPF_CGROUP_UDP4_SENDMSG:
8148                         break;
8149                 default:
8150                         return false;
8151                 }
8152                 break;
8153         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8154                                 msg_src_ip6[3]):
8155                 switch (prog->expected_attach_type) {
8156                 case BPF_CGROUP_UDP6_SENDMSG:
8157                         break;
8158                 default:
8159                         return false;
8160                 }
8161                 break;
8162         }
8163
8164         switch (off) {
8165         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8166         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8167         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8168         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8169                                 msg_src_ip6[3]):
8170         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8171                 if (type == BPF_READ) {
8172                         bpf_ctx_record_field_size(info, size_default);
8173
8174                         if (bpf_ctx_wide_access_ok(off, size,
8175                                                    struct bpf_sock_addr,
8176                                                    user_ip6))
8177                                 return true;
8178
8179                         if (bpf_ctx_wide_access_ok(off, size,
8180                                                    struct bpf_sock_addr,
8181                                                    msg_src_ip6))
8182                                 return true;
8183
8184                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8185                                 return false;
8186                 } else {
8187                         if (bpf_ctx_wide_access_ok(off, size,
8188                                                    struct bpf_sock_addr,
8189                                                    user_ip6))
8190                                 return true;
8191
8192                         if (bpf_ctx_wide_access_ok(off, size,
8193                                                    struct bpf_sock_addr,
8194                                                    msg_src_ip6))
8195                                 return true;
8196
8197                         if (size != size_default)
8198                                 return false;
8199                 }
8200                 break;
8201         case offsetof(struct bpf_sock_addr, sk):
8202                 if (type != BPF_READ)
8203                         return false;
8204                 if (size != sizeof(__u64))
8205                         return false;
8206                 info->reg_type = PTR_TO_SOCKET;
8207                 break;
8208         default:
8209                 if (type == BPF_READ) {
8210                         if (size != size_default)
8211                                 return false;
8212                 } else {
8213                         return false;
8214                 }
8215         }
8216
8217         return true;
8218 }
8219
8220 static bool sock_ops_is_valid_access(int off, int size,
8221                                      enum bpf_access_type type,
8222                                      const struct bpf_prog *prog,
8223                                      struct bpf_insn_access_aux *info)
8224 {
8225         const int size_default = sizeof(__u32);
8226
8227         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8228                 return false;
8229
8230         /* The verifier guarantees that size > 0. */
8231         if (off % size != 0)
8232                 return false;
8233
8234         if (type == BPF_WRITE) {
8235                 switch (off) {
8236                 case offsetof(struct bpf_sock_ops, reply):
8237                 case offsetof(struct bpf_sock_ops, sk_txhash):
8238                         if (size != size_default)
8239                                 return false;
8240                         break;
8241                 default:
8242                         return false;
8243                 }
8244         } else {
8245                 switch (off) {
8246                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8247                                         bytes_acked):
8248                         if (size != sizeof(__u64))
8249                                 return false;
8250                         break;
8251                 case offsetof(struct bpf_sock_ops, sk):
8252                         if (size != sizeof(__u64))
8253                                 return false;
8254                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8255                         break;
8256                 case offsetof(struct bpf_sock_ops, skb_data):
8257                         if (size != sizeof(__u64))
8258                                 return false;
8259                         info->reg_type = PTR_TO_PACKET;
8260                         break;
8261                 case offsetof(struct bpf_sock_ops, skb_data_end):
8262                         if (size != sizeof(__u64))
8263                                 return false;
8264                         info->reg_type = PTR_TO_PACKET_END;
8265                         break;
8266                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8267                         bpf_ctx_record_field_size(info, size_default);
8268                         return bpf_ctx_narrow_access_ok(off, size,
8269                                                         size_default);
8270                 default:
8271                         if (size != size_default)
8272                                 return false;
8273                         break;
8274                 }
8275         }
8276
8277         return true;
8278 }
8279
8280 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8281                            const struct bpf_prog *prog)
8282 {
8283         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8284 }
8285
8286 static bool sk_skb_is_valid_access(int off, int size,
8287                                    enum bpf_access_type type,
8288                                    const struct bpf_prog *prog,
8289                                    struct bpf_insn_access_aux *info)
8290 {
8291         switch (off) {
8292         case bpf_ctx_range(struct __sk_buff, tc_classid):
8293         case bpf_ctx_range(struct __sk_buff, data_meta):
8294         case bpf_ctx_range(struct __sk_buff, tstamp):
8295         case bpf_ctx_range(struct __sk_buff, wire_len):
8296                 return false;
8297         }
8298
8299         if (type == BPF_WRITE) {
8300                 switch (off) {
8301                 case bpf_ctx_range(struct __sk_buff, tc_index):
8302                 case bpf_ctx_range(struct __sk_buff, priority):
8303                         break;
8304                 default:
8305                         return false;
8306                 }
8307         }
8308
8309         switch (off) {
8310         case bpf_ctx_range(struct __sk_buff, mark):
8311                 return false;
8312         case bpf_ctx_range(struct __sk_buff, data):
8313                 info->reg_type = PTR_TO_PACKET;
8314                 break;
8315         case bpf_ctx_range(struct __sk_buff, data_end):
8316                 info->reg_type = PTR_TO_PACKET_END;
8317                 break;
8318         }
8319
8320         return bpf_skb_is_valid_access(off, size, type, prog, info);
8321 }
8322
8323 static bool sk_msg_is_valid_access(int off, int size,
8324                                    enum bpf_access_type type,
8325                                    const struct bpf_prog *prog,
8326                                    struct bpf_insn_access_aux *info)
8327 {
8328         if (type == BPF_WRITE)
8329                 return false;
8330
8331         if (off % size != 0)
8332                 return false;
8333
8334         switch (off) {
8335         case offsetof(struct sk_msg_md, data):
8336                 info->reg_type = PTR_TO_PACKET;
8337                 if (size != sizeof(__u64))
8338                         return false;
8339                 break;
8340         case offsetof(struct sk_msg_md, data_end):
8341                 info->reg_type = PTR_TO_PACKET_END;
8342                 if (size != sizeof(__u64))
8343                         return false;
8344                 break;
8345         case offsetof(struct sk_msg_md, sk):
8346                 if (size != sizeof(__u64))
8347                         return false;
8348                 info->reg_type = PTR_TO_SOCKET;
8349                 break;
8350         case bpf_ctx_range(struct sk_msg_md, family):
8351         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8352         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8353         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8354         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8355         case bpf_ctx_range(struct sk_msg_md, remote_port):
8356         case bpf_ctx_range(struct sk_msg_md, local_port):
8357         case bpf_ctx_range(struct sk_msg_md, size):
8358                 if (size != sizeof(__u32))
8359                         return false;
8360                 break;
8361         default:
8362                 return false;
8363         }
8364         return true;
8365 }
8366
8367 static bool flow_dissector_is_valid_access(int off, int size,
8368                                            enum bpf_access_type type,
8369                                            const struct bpf_prog *prog,
8370                                            struct bpf_insn_access_aux *info)
8371 {
8372         const int size_default = sizeof(__u32);
8373
8374         if (off < 0 || off >= sizeof(struct __sk_buff))
8375                 return false;
8376
8377         if (type == BPF_WRITE)
8378                 return false;
8379
8380         switch (off) {
8381         case bpf_ctx_range(struct __sk_buff, data):
8382                 if (size != size_default)
8383                         return false;
8384                 info->reg_type = PTR_TO_PACKET;
8385                 return true;
8386         case bpf_ctx_range(struct __sk_buff, data_end):
8387                 if (size != size_default)
8388                         return false;
8389                 info->reg_type = PTR_TO_PACKET_END;
8390                 return true;
8391         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8392                 if (size != sizeof(__u64))
8393                         return false;
8394                 info->reg_type = PTR_TO_FLOW_KEYS;
8395                 return true;
8396         default:
8397                 return false;
8398         }
8399 }
8400
8401 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8402                                              const struct bpf_insn *si,
8403                                              struct bpf_insn *insn_buf,
8404                                              struct bpf_prog *prog,
8405                                              u32 *target_size)
8406
8407 {
8408         struct bpf_insn *insn = insn_buf;
8409
8410         switch (si->off) {
8411         case offsetof(struct __sk_buff, data):
8412                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8413                                       si->dst_reg, si->src_reg,
8414                                       offsetof(struct bpf_flow_dissector, data));
8415                 break;
8416
8417         case offsetof(struct __sk_buff, data_end):
8418                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8419                                       si->dst_reg, si->src_reg,
8420                                       offsetof(struct bpf_flow_dissector, data_end));
8421                 break;
8422
8423         case offsetof(struct __sk_buff, flow_keys):
8424                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8425                                       si->dst_reg, si->src_reg,
8426                                       offsetof(struct bpf_flow_dissector, flow_keys));
8427                 break;
8428         }
8429
8430         return insn - insn_buf;
8431 }
8432
8433 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
8434                                                   struct bpf_insn *insn)
8435 {
8436         /* si->dst_reg = skb_shinfo(SKB); */
8437 #ifdef NET_SKBUFF_DATA_USES_OFFSET
8438         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8439                               BPF_REG_AX, si->src_reg,
8440                               offsetof(struct sk_buff, end));
8441         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
8442                               si->dst_reg, si->src_reg,
8443                               offsetof(struct sk_buff, head));
8444         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
8445 #else
8446         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
8447                               si->dst_reg, si->src_reg,
8448                               offsetof(struct sk_buff, end));
8449 #endif
8450
8451         return insn;
8452 }
8453
8454 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
8455                                   const struct bpf_insn *si,
8456                                   struct bpf_insn *insn_buf,
8457                                   struct bpf_prog *prog, u32 *target_size)
8458 {
8459         struct bpf_insn *insn = insn_buf;
8460         int off;
8461
8462         switch (si->off) {
8463         case offsetof(struct __sk_buff, len):
8464                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8465                                       bpf_target_off(struct sk_buff, len, 4,
8466                                                      target_size));
8467                 break;
8468
8469         case offsetof(struct __sk_buff, protocol):
8470                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8471                                       bpf_target_off(struct sk_buff, protocol, 2,
8472                                                      target_size));
8473                 break;
8474
8475         case offsetof(struct __sk_buff, vlan_proto):
8476                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8477                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
8478                                                      target_size));
8479                 break;
8480
8481         case offsetof(struct __sk_buff, priority):
8482                 if (type == BPF_WRITE)
8483                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8484                                               bpf_target_off(struct sk_buff, priority, 4,
8485                                                              target_size));
8486                 else
8487                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8488                                               bpf_target_off(struct sk_buff, priority, 4,
8489                                                              target_size));
8490                 break;
8491
8492         case offsetof(struct __sk_buff, ingress_ifindex):
8493                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8494                                       bpf_target_off(struct sk_buff, skb_iif, 4,
8495                                                      target_size));
8496                 break;
8497
8498         case offsetof(struct __sk_buff, ifindex):
8499                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8500                                       si->dst_reg, si->src_reg,
8501                                       offsetof(struct sk_buff, dev));
8502                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
8503                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8504                                       bpf_target_off(struct net_device, ifindex, 4,
8505                                                      target_size));
8506                 break;
8507
8508         case offsetof(struct __sk_buff, hash):
8509                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8510                                       bpf_target_off(struct sk_buff, hash, 4,
8511                                                      target_size));
8512                 break;
8513
8514         case offsetof(struct __sk_buff, mark):
8515                 if (type == BPF_WRITE)
8516                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8517                                               bpf_target_off(struct sk_buff, mark, 4,
8518                                                              target_size));
8519                 else
8520                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8521                                               bpf_target_off(struct sk_buff, mark, 4,
8522                                                              target_size));
8523                 break;
8524
8525         case offsetof(struct __sk_buff, pkt_type):
8526                 *target_size = 1;
8527                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8528                                       PKT_TYPE_OFFSET());
8529                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
8530 #ifdef __BIG_ENDIAN_BITFIELD
8531                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
8532 #endif
8533                 break;
8534
8535         case offsetof(struct __sk_buff, queue_mapping):
8536                 if (type == BPF_WRITE) {
8537                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
8538                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8539                                               bpf_target_off(struct sk_buff,
8540                                                              queue_mapping,
8541                                                              2, target_size));
8542                 } else {
8543                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8544                                               bpf_target_off(struct sk_buff,
8545                                                              queue_mapping,
8546                                                              2, target_size));
8547                 }
8548                 break;
8549
8550         case offsetof(struct __sk_buff, vlan_present):
8551                 *target_size = 1;
8552                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
8553                                       PKT_VLAN_PRESENT_OFFSET());
8554                 if (PKT_VLAN_PRESENT_BIT)
8555                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
8556                 if (PKT_VLAN_PRESENT_BIT < 7)
8557                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
8558                 break;
8559
8560         case offsetof(struct __sk_buff, vlan_tci):
8561                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8562                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
8563                                                      target_size));
8564                 break;
8565
8566         case offsetof(struct __sk_buff, cb[0]) ...
8567              offsetofend(struct __sk_buff, cb[4]) - 1:
8568                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
8569                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
8570                               offsetof(struct qdisc_skb_cb, data)) %
8571                              sizeof(__u64));
8572
8573                 prog->cb_access = 1;
8574                 off  = si->off;
8575                 off -= offsetof(struct __sk_buff, cb[0]);
8576                 off += offsetof(struct sk_buff, cb);
8577                 off += offsetof(struct qdisc_skb_cb, data);
8578                 if (type == BPF_WRITE)
8579                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
8580                                               si->src_reg, off);
8581                 else
8582                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
8583                                               si->src_reg, off);
8584                 break;
8585
8586         case offsetof(struct __sk_buff, tc_classid):
8587                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
8588
8589                 off  = si->off;
8590                 off -= offsetof(struct __sk_buff, tc_classid);
8591                 off += offsetof(struct sk_buff, cb);
8592                 off += offsetof(struct qdisc_skb_cb, tc_classid);
8593                 *target_size = 2;
8594                 if (type == BPF_WRITE)
8595                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
8596                                               si->src_reg, off);
8597                 else
8598                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
8599                                               si->src_reg, off);
8600                 break;
8601
8602         case offsetof(struct __sk_buff, data):
8603                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
8604                                       si->dst_reg, si->src_reg,
8605                                       offsetof(struct sk_buff, data));
8606                 break;
8607
8608         case offsetof(struct __sk_buff, data_meta):
8609                 off  = si->off;
8610                 off -= offsetof(struct __sk_buff, data_meta);
8611                 off += offsetof(struct sk_buff, cb);
8612                 off += offsetof(struct bpf_skb_data_end, data_meta);
8613                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8614                                       si->src_reg, off);
8615                 break;
8616
8617         case offsetof(struct __sk_buff, data_end):
8618                 off  = si->off;
8619                 off -= offsetof(struct __sk_buff, data_end);
8620                 off += offsetof(struct sk_buff, cb);
8621                 off += offsetof(struct bpf_skb_data_end, data_end);
8622                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
8623                                       si->src_reg, off);
8624                 break;
8625
8626         case offsetof(struct __sk_buff, tc_index):
8627 #ifdef CONFIG_NET_SCHED
8628                 if (type == BPF_WRITE)
8629                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
8630                                               bpf_target_off(struct sk_buff, tc_index, 2,
8631                                                              target_size));
8632                 else
8633                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
8634                                               bpf_target_off(struct sk_buff, tc_index, 2,
8635                                                              target_size));
8636 #else
8637                 *target_size = 2;
8638                 if (type == BPF_WRITE)
8639                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
8640                 else
8641                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8642 #endif
8643                 break;
8644
8645         case offsetof(struct __sk_buff, napi_id):
8646 #if defined(CONFIG_NET_RX_BUSY_POLL)
8647                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8648                                       bpf_target_off(struct sk_buff, napi_id, 4,
8649                                                      target_size));
8650                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
8651                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8652 #else
8653                 *target_size = 4;
8654                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
8655 #endif
8656                 break;
8657         case offsetof(struct __sk_buff, family):
8658                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
8659
8660                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8661                                       si->dst_reg, si->src_reg,
8662                                       offsetof(struct sk_buff, sk));
8663                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8664                                       bpf_target_off(struct sock_common,
8665                                                      skc_family,
8666                                                      2, target_size));
8667                 break;
8668         case offsetof(struct __sk_buff, remote_ip4):
8669                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
8670
8671                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8672                                       si->dst_reg, si->src_reg,
8673                                       offsetof(struct sk_buff, sk));
8674                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8675                                       bpf_target_off(struct sock_common,
8676                                                      skc_daddr,
8677                                                      4, target_size));
8678                 break;
8679         case offsetof(struct __sk_buff, local_ip4):
8680                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8681                                           skc_rcv_saddr) != 4);
8682
8683                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8684                                       si->dst_reg, si->src_reg,
8685                                       offsetof(struct sk_buff, sk));
8686                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8687                                       bpf_target_off(struct sock_common,
8688                                                      skc_rcv_saddr,
8689                                                      4, target_size));
8690                 break;
8691         case offsetof(struct __sk_buff, remote_ip6[0]) ...
8692              offsetof(struct __sk_buff, remote_ip6[3]):
8693 #if IS_ENABLED(CONFIG_IPV6)
8694                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8695                                           skc_v6_daddr.s6_addr32[0]) != 4);
8696
8697                 off = si->off;
8698                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
8699
8700                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8701                                       si->dst_reg, si->src_reg,
8702                                       offsetof(struct sk_buff, sk));
8703                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8704                                       offsetof(struct sock_common,
8705                                                skc_v6_daddr.s6_addr32[0]) +
8706                                       off);
8707 #else
8708                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8709 #endif
8710                 break;
8711         case offsetof(struct __sk_buff, local_ip6[0]) ...
8712              offsetof(struct __sk_buff, local_ip6[3]):
8713 #if IS_ENABLED(CONFIG_IPV6)
8714                 BUILD_BUG_ON(sizeof_field(struct sock_common,
8715                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
8716
8717                 off = si->off;
8718                 off -= offsetof(struct __sk_buff, local_ip6[0]);
8719
8720                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8721                                       si->dst_reg, si->src_reg,
8722                                       offsetof(struct sk_buff, sk));
8723                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
8724                                       offsetof(struct sock_common,
8725                                                skc_v6_rcv_saddr.s6_addr32[0]) +
8726                                       off);
8727 #else
8728                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8729 #endif
8730                 break;
8731
8732         case offsetof(struct __sk_buff, remote_port):
8733                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
8734
8735                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8736                                       si->dst_reg, si->src_reg,
8737                                       offsetof(struct sk_buff, sk));
8738                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8739                                       bpf_target_off(struct sock_common,
8740                                                      skc_dport,
8741                                                      2, target_size));
8742 #ifndef __BIG_ENDIAN_BITFIELD
8743                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
8744 #endif
8745                 break;
8746
8747         case offsetof(struct __sk_buff, local_port):
8748                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
8749
8750                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8751                                       si->dst_reg, si->src_reg,
8752                                       offsetof(struct sk_buff, sk));
8753                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
8754                                       bpf_target_off(struct sock_common,
8755                                                      skc_num, 2, target_size));
8756                 break;
8757
8758         case offsetof(struct __sk_buff, tstamp):
8759                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
8760
8761                 if (type == BPF_WRITE)
8762                         *insn++ = BPF_STX_MEM(BPF_DW,
8763                                               si->dst_reg, si->src_reg,
8764                                               bpf_target_off(struct sk_buff,
8765                                                              tstamp, 8,
8766                                                              target_size));
8767                 else
8768                         *insn++ = BPF_LDX_MEM(BPF_DW,
8769                                               si->dst_reg, si->src_reg,
8770                                               bpf_target_off(struct sk_buff,
8771                                                              tstamp, 8,
8772                                                              target_size));
8773                 break;
8774
8775         case offsetof(struct __sk_buff, gso_segs):
8776                 insn = bpf_convert_shinfo_access(si, insn);
8777                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
8778                                       si->dst_reg, si->dst_reg,
8779                                       bpf_target_off(struct skb_shared_info,
8780                                                      gso_segs, 2,
8781                                                      target_size));
8782                 break;
8783         case offsetof(struct __sk_buff, gso_size):
8784                 insn = bpf_convert_shinfo_access(si, insn);
8785                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
8786                                       si->dst_reg, si->dst_reg,
8787                                       bpf_target_off(struct skb_shared_info,
8788                                                      gso_size, 2,
8789                                                      target_size));
8790                 break;
8791         case offsetof(struct __sk_buff, wire_len):
8792                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
8793
8794                 off = si->off;
8795                 off -= offsetof(struct __sk_buff, wire_len);
8796                 off += offsetof(struct sk_buff, cb);
8797                 off += offsetof(struct qdisc_skb_cb, pkt_len);
8798                 *target_size = 4;
8799                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
8800                 break;
8801
8802         case offsetof(struct __sk_buff, sk):
8803                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
8804                                       si->dst_reg, si->src_reg,
8805                                       offsetof(struct sk_buff, sk));
8806                 break;
8807         }
8808
8809         return insn - insn_buf;
8810 }
8811
8812 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
8813                                 const struct bpf_insn *si,
8814                                 struct bpf_insn *insn_buf,
8815                                 struct bpf_prog *prog, u32 *target_size)
8816 {
8817         struct bpf_insn *insn = insn_buf;
8818         int off;
8819
8820         switch (si->off) {
8821         case offsetof(struct bpf_sock, bound_dev_if):
8822                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
8823
8824                 if (type == BPF_WRITE)
8825                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8826                                         offsetof(struct sock, sk_bound_dev_if));
8827                 else
8828                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8829                                       offsetof(struct sock, sk_bound_dev_if));
8830                 break;
8831
8832         case offsetof(struct bpf_sock, mark):
8833                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
8834
8835                 if (type == BPF_WRITE)
8836                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8837                                         offsetof(struct sock, sk_mark));
8838                 else
8839                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8840                                       offsetof(struct sock, sk_mark));
8841                 break;
8842
8843         case offsetof(struct bpf_sock, priority):
8844                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
8845
8846                 if (type == BPF_WRITE)
8847                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
8848                                         offsetof(struct sock, sk_priority));
8849                 else
8850                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
8851                                       offsetof(struct sock, sk_priority));
8852                 break;
8853
8854         case offsetof(struct bpf_sock, family):
8855                 *insn++ = BPF_LDX_MEM(
8856                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
8857                         si->dst_reg, si->src_reg,
8858                         bpf_target_off(struct sock_common,
8859                                        skc_family,
8860                                        sizeof_field(struct sock_common,
8861                                                     skc_family),
8862                                        target_size));
8863                 break;
8864
8865         case offsetof(struct bpf_sock, type):
8866                 *insn++ = BPF_LDX_MEM(
8867                         BPF_FIELD_SIZEOF(struct sock, sk_type),
8868                         si->dst_reg, si->src_reg,
8869                         bpf_target_off(struct sock, sk_type,
8870                                        sizeof_field(struct sock, sk_type),
8871                                        target_size));
8872                 break;
8873
8874         case offsetof(struct bpf_sock, protocol):
8875                 *insn++ = BPF_LDX_MEM(
8876                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
8877                         si->dst_reg, si->src_reg,
8878                         bpf_target_off(struct sock, sk_protocol,
8879                                        sizeof_field(struct sock, sk_protocol),
8880                                        target_size));
8881                 break;
8882
8883         case offsetof(struct bpf_sock, src_ip4):
8884                 *insn++ = BPF_LDX_MEM(
8885                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8886                         bpf_target_off(struct sock_common, skc_rcv_saddr,
8887                                        sizeof_field(struct sock_common,
8888                                                     skc_rcv_saddr),
8889                                        target_size));
8890                 break;
8891
8892         case offsetof(struct bpf_sock, dst_ip4):
8893                 *insn++ = BPF_LDX_MEM(
8894                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8895                         bpf_target_off(struct sock_common, skc_daddr,
8896                                        sizeof_field(struct sock_common,
8897                                                     skc_daddr),
8898                                        target_size));
8899                 break;
8900
8901         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8902 #if IS_ENABLED(CONFIG_IPV6)
8903                 off = si->off;
8904                 off -= offsetof(struct bpf_sock, src_ip6[0]);
8905                 *insn++ = BPF_LDX_MEM(
8906                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8907                         bpf_target_off(
8908                                 struct sock_common,
8909                                 skc_v6_rcv_saddr.s6_addr32[0],
8910                                 sizeof_field(struct sock_common,
8911                                              skc_v6_rcv_saddr.s6_addr32[0]),
8912                                 target_size) + off);
8913 #else
8914                 (void)off;
8915                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8916 #endif
8917                 break;
8918
8919         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8920 #if IS_ENABLED(CONFIG_IPV6)
8921                 off = si->off;
8922                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
8923                 *insn++ = BPF_LDX_MEM(
8924                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
8925                         bpf_target_off(struct sock_common,
8926                                        skc_v6_daddr.s6_addr32[0],
8927                                        sizeof_field(struct sock_common,
8928                                                     skc_v6_daddr.s6_addr32[0]),
8929                                        target_size) + off);
8930 #else
8931                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
8932                 *target_size = 4;
8933 #endif
8934                 break;
8935
8936         case offsetof(struct bpf_sock, src_port):
8937                 *insn++ = BPF_LDX_MEM(
8938                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
8939                         si->dst_reg, si->src_reg,
8940                         bpf_target_off(struct sock_common, skc_num,
8941                                        sizeof_field(struct sock_common,
8942                                                     skc_num),
8943                                        target_size));
8944                 break;
8945
8946         case offsetof(struct bpf_sock, dst_port):
8947                 *insn++ = BPF_LDX_MEM(
8948                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
8949                         si->dst_reg, si->src_reg,
8950                         bpf_target_off(struct sock_common, skc_dport,
8951                                        sizeof_field(struct sock_common,
8952                                                     skc_dport),
8953                                        target_size));
8954                 break;
8955
8956         case offsetof(struct bpf_sock, state):
8957                 *insn++ = BPF_LDX_MEM(
8958                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
8959                         si->dst_reg, si->src_reg,
8960                         bpf_target_off(struct sock_common, skc_state,
8961                                        sizeof_field(struct sock_common,
8962                                                     skc_state),
8963                                        target_size));
8964                 break;
8965         case offsetof(struct bpf_sock, rx_queue_mapping):
8966 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
8967                 *insn++ = BPF_LDX_MEM(
8968                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
8969                         si->dst_reg, si->src_reg,
8970                         bpf_target_off(struct sock, sk_rx_queue_mapping,
8971                                        sizeof_field(struct sock,
8972                                                     sk_rx_queue_mapping),
8973                                        target_size));
8974                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
8975                                       1);
8976                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8977 #else
8978                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
8979                 *target_size = 2;
8980 #endif
8981                 break;
8982         }
8983
8984         return insn - insn_buf;
8985 }
8986
8987 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
8988                                          const struct bpf_insn *si,
8989                                          struct bpf_insn *insn_buf,
8990                                          struct bpf_prog *prog, u32 *target_size)
8991 {
8992         struct bpf_insn *insn = insn_buf;
8993
8994         switch (si->off) {
8995         case offsetof(struct __sk_buff, ifindex):
8996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
8997                                       si->dst_reg, si->src_reg,
8998                                       offsetof(struct sk_buff, dev));
8999                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9000                                       bpf_target_off(struct net_device, ifindex, 4,
9001                                                      target_size));
9002                 break;
9003         default:
9004                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9005                                               target_size);
9006         }
9007
9008         return insn - insn_buf;
9009 }
9010
9011 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9012                                   const struct bpf_insn *si,
9013                                   struct bpf_insn *insn_buf,
9014                                   struct bpf_prog *prog, u32 *target_size)
9015 {
9016         struct bpf_insn *insn = insn_buf;
9017
9018         switch (si->off) {
9019         case offsetof(struct xdp_md, data):
9020                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9021                                       si->dst_reg, si->src_reg,
9022                                       offsetof(struct xdp_buff, data));
9023                 break;
9024         case offsetof(struct xdp_md, data_meta):
9025                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9026                                       si->dst_reg, si->src_reg,
9027                                       offsetof(struct xdp_buff, data_meta));
9028                 break;
9029         case offsetof(struct xdp_md, data_end):
9030                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9031                                       si->dst_reg, si->src_reg,
9032                                       offsetof(struct xdp_buff, data_end));
9033                 break;
9034         case offsetof(struct xdp_md, ingress_ifindex):
9035                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9036                                       si->dst_reg, si->src_reg,
9037                                       offsetof(struct xdp_buff, rxq));
9038                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9039                                       si->dst_reg, si->dst_reg,
9040                                       offsetof(struct xdp_rxq_info, dev));
9041                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9042                                       offsetof(struct net_device, ifindex));
9043                 break;
9044         case offsetof(struct xdp_md, rx_queue_index):
9045                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9046                                       si->dst_reg, si->src_reg,
9047                                       offsetof(struct xdp_buff, rxq));
9048                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9049                                       offsetof(struct xdp_rxq_info,
9050                                                queue_index));
9051                 break;
9052         case offsetof(struct xdp_md, egress_ifindex):
9053                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9054                                       si->dst_reg, si->src_reg,
9055                                       offsetof(struct xdp_buff, txq));
9056                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9057                                       si->dst_reg, si->dst_reg,
9058                                       offsetof(struct xdp_txq_info, dev));
9059                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9060                                       offsetof(struct net_device, ifindex));
9061                 break;
9062         }
9063
9064         return insn - insn_buf;
9065 }
9066
9067 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9068  * context Structure, F is Field in context structure that contains a pointer
9069  * to Nested Structure of type NS that has the field NF.
9070  *
9071  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9072  * sure that SIZE is not greater than actual size of S.F.NF.
9073  *
9074  * If offset OFF is provided, the load happens from that offset relative to
9075  * offset of NF.
9076  */
9077 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9078         do {                                                                   \
9079                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9080                                       si->src_reg, offsetof(S, F));            \
9081                 *insn++ = BPF_LDX_MEM(                                         \
9082                         SIZE, si->dst_reg, si->dst_reg,                        \
9083                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9084                                        target_size)                            \
9085                                 + OFF);                                        \
9086         } while (0)
9087
9088 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9089         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9090                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9091
9092 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9093  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9094  *
9095  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9096  * "register" since two registers available in convert_ctx_access are not
9097  * enough: we can't override neither SRC, since it contains value to store, nor
9098  * DST since it contains pointer to context that may be used by later
9099  * instructions. But we need a temporary place to save pointer to nested
9100  * structure whose field we want to store to.
9101  */
9102 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9103         do {                                                                   \
9104                 int tmp_reg = BPF_REG_9;                                       \
9105                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9106                         --tmp_reg;                                             \
9107                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9108                         --tmp_reg;                                             \
9109                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9110                                       offsetof(S, TF));                        \
9111                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9112                                       si->dst_reg, offsetof(S, F));            \
9113                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9114                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9115                                        target_size)                            \
9116                                 + OFF);                                        \
9117                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9118                                       offsetof(S, TF));                        \
9119         } while (0)
9120
9121 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9122                                                       TF)                      \
9123         do {                                                                   \
9124                 if (type == BPF_WRITE) {                                       \
9125                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9126                                                          OFF, TF);             \
9127                 } else {                                                       \
9128                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9129                                 S, NS, F, NF, SIZE, OFF);  \
9130                 }                                                              \
9131         } while (0)
9132
9133 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9134         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9135                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9136
9137 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9138                                         const struct bpf_insn *si,
9139                                         struct bpf_insn *insn_buf,
9140                                         struct bpf_prog *prog, u32 *target_size)
9141 {
9142         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9143         struct bpf_insn *insn = insn_buf;
9144
9145         switch (si->off) {
9146         case offsetof(struct bpf_sock_addr, user_family):
9147                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9148                                             struct sockaddr, uaddr, sa_family);
9149                 break;
9150
9151         case offsetof(struct bpf_sock_addr, user_ip4):
9152                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9153                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9154                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9155                 break;
9156
9157         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9158                 off = si->off;
9159                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9160                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9161                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9162                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9163                         tmp_reg);
9164                 break;
9165
9166         case offsetof(struct bpf_sock_addr, user_port):
9167                 /* To get port we need to know sa_family first and then treat
9168                  * sockaddr as either sockaddr_in or sockaddr_in6.
9169                  * Though we can simplify since port field has same offset and
9170                  * size in both structures.
9171                  * Here we check this invariant and use just one of the
9172                  * structures if it's true.
9173                  */
9174                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9175                              offsetof(struct sockaddr_in6, sin6_port));
9176                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9177                              sizeof_field(struct sockaddr_in6, sin6_port));
9178                 /* Account for sin6_port being smaller than user_port. */
9179                 port_size = min(port_size, BPF_LDST_BYTES(si));
9180                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9181                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9182                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9183                 break;
9184
9185         case offsetof(struct bpf_sock_addr, family):
9186                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9187                                             struct sock, sk, sk_family);
9188                 break;
9189
9190         case offsetof(struct bpf_sock_addr, type):
9191                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9192                                             struct sock, sk, sk_type);
9193                 break;
9194
9195         case offsetof(struct bpf_sock_addr, protocol):
9196                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9197                                             struct sock, sk, sk_protocol);
9198                 break;
9199
9200         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9201                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9202                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9203                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9204                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9205                 break;
9206
9207         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9208                                 msg_src_ip6[3]):
9209                 off = si->off;
9210                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9211                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9212                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9213                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9214                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9215                 break;
9216         case offsetof(struct bpf_sock_addr, sk):
9217                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9218                                       si->dst_reg, si->src_reg,
9219                                       offsetof(struct bpf_sock_addr_kern, sk));
9220                 break;
9221         }
9222
9223         return insn - insn_buf;
9224 }
9225
9226 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9227                                        const struct bpf_insn *si,
9228                                        struct bpf_insn *insn_buf,
9229                                        struct bpf_prog *prog,
9230                                        u32 *target_size)
9231 {
9232         struct bpf_insn *insn = insn_buf;
9233         int off;
9234
9235 /* Helper macro for adding read access to tcp_sock or sock fields. */
9236 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9237         do {                                                                  \
9238                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9239                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9240                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9241                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9242                         reg--;                                                \
9243                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9244                         reg--;                                                \
9245                 if (si->dst_reg == si->src_reg) {                             \
9246                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9247                                           offsetof(struct bpf_sock_ops_kern,  \
9248                                           temp));                             \
9249                         fullsock_reg = reg;                                   \
9250                         jmp += 2;                                             \
9251                 }                                                             \
9252                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9253                                                 struct bpf_sock_ops_kern,     \
9254                                                 is_fullsock),                 \
9255                                       fullsock_reg, si->src_reg,              \
9256                                       offsetof(struct bpf_sock_ops_kern,      \
9257                                                is_fullsock));                 \
9258                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9259                 if (si->dst_reg == si->src_reg)                               \
9260                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9261                                       offsetof(struct bpf_sock_ops_kern,      \
9262                                       temp));                                 \
9263                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9264                                                 struct bpf_sock_ops_kern, sk),\
9265                                       si->dst_reg, si->src_reg,               \
9266                                       offsetof(struct bpf_sock_ops_kern, sk));\
9267                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9268                                                        OBJ_FIELD),            \
9269                                       si->dst_reg, si->dst_reg,               \
9270                                       offsetof(OBJ, OBJ_FIELD));              \
9271                 if (si->dst_reg == si->src_reg) {                             \
9272                         *insn++ = BPF_JMP_A(1);                               \
9273                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9274                                       offsetof(struct bpf_sock_ops_kern,      \
9275                                       temp));                                 \
9276                 }                                                             \
9277         } while (0)
9278
9279 #define SOCK_OPS_GET_SK()                                                             \
9280         do {                                                                  \
9281                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9282                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9283                         reg--;                                                \
9284                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9285                         reg--;                                                \
9286                 if (si->dst_reg == si->src_reg) {                             \
9287                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9288                                           offsetof(struct bpf_sock_ops_kern,  \
9289                                           temp));                             \
9290                         fullsock_reg = reg;                                   \
9291                         jmp += 2;                                             \
9292                 }                                                             \
9293                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9294                                                 struct bpf_sock_ops_kern,     \
9295                                                 is_fullsock),                 \
9296                                       fullsock_reg, si->src_reg,              \
9297                                       offsetof(struct bpf_sock_ops_kern,      \
9298                                                is_fullsock));                 \
9299                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9300                 if (si->dst_reg == si->src_reg)                               \
9301                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9302                                       offsetof(struct bpf_sock_ops_kern,      \
9303                                       temp));                                 \
9304                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9305                                                 struct bpf_sock_ops_kern, sk),\
9306                                       si->dst_reg, si->src_reg,               \
9307                                       offsetof(struct bpf_sock_ops_kern, sk));\
9308                 if (si->dst_reg == si->src_reg) {                             \
9309                         *insn++ = BPF_JMP_A(1);                               \
9310                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9311                                       offsetof(struct bpf_sock_ops_kern,      \
9312                                       temp));                                 \
9313                 }                                                             \
9314         } while (0)
9315
9316 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9317                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9318
9319 /* Helper macro for adding write access to tcp_sock or sock fields.
9320  * The macro is called with two registers, dst_reg which contains a pointer
9321  * to ctx (context) and src_reg which contains the value that should be
9322  * stored. However, we need an additional register since we cannot overwrite
9323  * dst_reg because it may be used later in the program.
9324  * Instead we "borrow" one of the other register. We first save its value
9325  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9326  * it at the end of the macro.
9327  */
9328 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9329         do {                                                                  \
9330                 int reg = BPF_REG_9;                                          \
9331                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9332                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9333                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9334                         reg--;                                                \
9335                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9336                         reg--;                                                \
9337                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9338                                       offsetof(struct bpf_sock_ops_kern,      \
9339                                                temp));                        \
9340                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9341                                                 struct bpf_sock_ops_kern,     \
9342                                                 is_fullsock),                 \
9343                                       reg, si->dst_reg,                       \
9344                                       offsetof(struct bpf_sock_ops_kern,      \
9345                                                is_fullsock));                 \
9346                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
9347                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9348                                                 struct bpf_sock_ops_kern, sk),\
9349                                       reg, si->dst_reg,                       \
9350                                       offsetof(struct bpf_sock_ops_kern, sk));\
9351                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
9352                                       reg, si->src_reg,                       \
9353                                       offsetof(OBJ, OBJ_FIELD));              \
9354                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
9355                                       offsetof(struct bpf_sock_ops_kern,      \
9356                                                temp));                        \
9357         } while (0)
9358
9359 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
9360         do {                                                                  \
9361                 if (TYPE == BPF_WRITE)                                        \
9362                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9363                 else                                                          \
9364                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
9365         } while (0)
9366
9367         if (insn > insn_buf)
9368                 return insn - insn_buf;
9369
9370         switch (si->off) {
9371         case offsetof(struct bpf_sock_ops, op):
9372                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9373                                                        op),
9374                                       si->dst_reg, si->src_reg,
9375                                       offsetof(struct bpf_sock_ops_kern, op));
9376                 break;
9377
9378         case offsetof(struct bpf_sock_ops, replylong[0]) ...
9379              offsetof(struct bpf_sock_ops, replylong[3]):
9380                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
9381                              sizeof_field(struct bpf_sock_ops_kern, reply));
9382                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
9383                              sizeof_field(struct bpf_sock_ops_kern, replylong));
9384                 off = si->off;
9385                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
9386                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
9387                 if (type == BPF_WRITE)
9388                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9389                                               off);
9390                 else
9391                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9392                                               off);
9393                 break;
9394
9395         case offsetof(struct bpf_sock_ops, family):
9396                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9397
9398                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9399                                               struct bpf_sock_ops_kern, sk),
9400                                       si->dst_reg, si->src_reg,
9401                                       offsetof(struct bpf_sock_ops_kern, sk));
9402                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9403                                       offsetof(struct sock_common, skc_family));
9404                 break;
9405
9406         case offsetof(struct bpf_sock_ops, remote_ip4):
9407                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9408
9409                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9410                                                 struct bpf_sock_ops_kern, sk),
9411                                       si->dst_reg, si->src_reg,
9412                                       offsetof(struct bpf_sock_ops_kern, sk));
9413                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9414                                       offsetof(struct sock_common, skc_daddr));
9415                 break;
9416
9417         case offsetof(struct bpf_sock_ops, local_ip4):
9418                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9419                                           skc_rcv_saddr) != 4);
9420
9421                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9422                                               struct bpf_sock_ops_kern, sk),
9423                                       si->dst_reg, si->src_reg,
9424                                       offsetof(struct bpf_sock_ops_kern, sk));
9425                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9426                                       offsetof(struct sock_common,
9427                                                skc_rcv_saddr));
9428                 break;
9429
9430         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
9431              offsetof(struct bpf_sock_ops, remote_ip6[3]):
9432 #if IS_ENABLED(CONFIG_IPV6)
9433                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9434                                           skc_v6_daddr.s6_addr32[0]) != 4);
9435
9436                 off = si->off;
9437                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
9438                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9439                                                 struct bpf_sock_ops_kern, sk),
9440                                       si->dst_reg, si->src_reg,
9441                                       offsetof(struct bpf_sock_ops_kern, sk));
9442                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9443                                       offsetof(struct sock_common,
9444                                                skc_v6_daddr.s6_addr32[0]) +
9445                                       off);
9446 #else
9447                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9448 #endif
9449                 break;
9450
9451         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
9452              offsetof(struct bpf_sock_ops, local_ip6[3]):
9453 #if IS_ENABLED(CONFIG_IPV6)
9454                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9455                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9456
9457                 off = si->off;
9458                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
9459                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9460                                                 struct bpf_sock_ops_kern, sk),
9461                                       si->dst_reg, si->src_reg,
9462                                       offsetof(struct bpf_sock_ops_kern, sk));
9463                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9464                                       offsetof(struct sock_common,
9465                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9466                                       off);
9467 #else
9468                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9469 #endif
9470                 break;
9471
9472         case offsetof(struct bpf_sock_ops, remote_port):
9473                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9474
9475                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9476                                                 struct bpf_sock_ops_kern, sk),
9477                                       si->dst_reg, si->src_reg,
9478                                       offsetof(struct bpf_sock_ops_kern, sk));
9479                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9480                                       offsetof(struct sock_common, skc_dport));
9481 #ifndef __BIG_ENDIAN_BITFIELD
9482                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9483 #endif
9484                 break;
9485
9486         case offsetof(struct bpf_sock_ops, local_port):
9487                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9488
9489                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9490                                                 struct bpf_sock_ops_kern, sk),
9491                                       si->dst_reg, si->src_reg,
9492                                       offsetof(struct bpf_sock_ops_kern, sk));
9493                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9494                                       offsetof(struct sock_common, skc_num));
9495                 break;
9496
9497         case offsetof(struct bpf_sock_ops, is_fullsock):
9498                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9499                                                 struct bpf_sock_ops_kern,
9500                                                 is_fullsock),
9501                                       si->dst_reg, si->src_reg,
9502                                       offsetof(struct bpf_sock_ops_kern,
9503                                                is_fullsock));
9504                 break;
9505
9506         case offsetof(struct bpf_sock_ops, state):
9507                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
9508
9509                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9510                                                 struct bpf_sock_ops_kern, sk),
9511                                       si->dst_reg, si->src_reg,
9512                                       offsetof(struct bpf_sock_ops_kern, sk));
9513                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
9514                                       offsetof(struct sock_common, skc_state));
9515                 break;
9516
9517         case offsetof(struct bpf_sock_ops, rtt_min):
9518                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
9519                              sizeof(struct minmax));
9520                 BUILD_BUG_ON(sizeof(struct minmax) <
9521                              sizeof(struct minmax_sample));
9522
9523                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9524                                                 struct bpf_sock_ops_kern, sk),
9525                                       si->dst_reg, si->src_reg,
9526                                       offsetof(struct bpf_sock_ops_kern, sk));
9527                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9528                                       offsetof(struct tcp_sock, rtt_min) +
9529                                       sizeof_field(struct minmax_sample, t));
9530                 break;
9531
9532         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
9533                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
9534                                    struct tcp_sock);
9535                 break;
9536
9537         case offsetof(struct bpf_sock_ops, sk_txhash):
9538                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
9539                                           struct sock, type);
9540                 break;
9541         case offsetof(struct bpf_sock_ops, snd_cwnd):
9542                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
9543                 break;
9544         case offsetof(struct bpf_sock_ops, srtt_us):
9545                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
9546                 break;
9547         case offsetof(struct bpf_sock_ops, snd_ssthresh):
9548                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
9549                 break;
9550         case offsetof(struct bpf_sock_ops, rcv_nxt):
9551                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
9552                 break;
9553         case offsetof(struct bpf_sock_ops, snd_nxt):
9554                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
9555                 break;
9556         case offsetof(struct bpf_sock_ops, snd_una):
9557                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
9558                 break;
9559         case offsetof(struct bpf_sock_ops, mss_cache):
9560                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
9561                 break;
9562         case offsetof(struct bpf_sock_ops, ecn_flags):
9563                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
9564                 break;
9565         case offsetof(struct bpf_sock_ops, rate_delivered):
9566                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
9567                 break;
9568         case offsetof(struct bpf_sock_ops, rate_interval_us):
9569                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
9570                 break;
9571         case offsetof(struct bpf_sock_ops, packets_out):
9572                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
9573                 break;
9574         case offsetof(struct bpf_sock_ops, retrans_out):
9575                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
9576                 break;
9577         case offsetof(struct bpf_sock_ops, total_retrans):
9578                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
9579                 break;
9580         case offsetof(struct bpf_sock_ops, segs_in):
9581                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
9582                 break;
9583         case offsetof(struct bpf_sock_ops, data_segs_in):
9584                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
9585                 break;
9586         case offsetof(struct bpf_sock_ops, segs_out):
9587                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
9588                 break;
9589         case offsetof(struct bpf_sock_ops, data_segs_out):
9590                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
9591                 break;
9592         case offsetof(struct bpf_sock_ops, lost_out):
9593                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
9594                 break;
9595         case offsetof(struct bpf_sock_ops, sacked_out):
9596                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
9597                 break;
9598         case offsetof(struct bpf_sock_ops, bytes_received):
9599                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
9600                 break;
9601         case offsetof(struct bpf_sock_ops, bytes_acked):
9602                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
9603                 break;
9604         case offsetof(struct bpf_sock_ops, sk):
9605                 SOCK_OPS_GET_SK();
9606                 break;
9607         case offsetof(struct bpf_sock_ops, skb_data_end):
9608                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9609                                                        skb_data_end),
9610                                       si->dst_reg, si->src_reg,
9611                                       offsetof(struct bpf_sock_ops_kern,
9612                                                skb_data_end));
9613                 break;
9614         case offsetof(struct bpf_sock_ops, skb_data):
9615                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9616                                                        skb),
9617                                       si->dst_reg, si->src_reg,
9618                                       offsetof(struct bpf_sock_ops_kern,
9619                                                skb));
9620                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9621                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9622                                       si->dst_reg, si->dst_reg,
9623                                       offsetof(struct sk_buff, data));
9624                 break;
9625         case offsetof(struct bpf_sock_ops, skb_len):
9626                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9627                                                        skb),
9628                                       si->dst_reg, si->src_reg,
9629                                       offsetof(struct bpf_sock_ops_kern,
9630                                                skb));
9631                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9632                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9633                                       si->dst_reg, si->dst_reg,
9634                                       offsetof(struct sk_buff, len));
9635                 break;
9636         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
9637                 off = offsetof(struct sk_buff, cb);
9638                 off += offsetof(struct tcp_skb_cb, tcp_flags);
9639                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
9640                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
9641                                                        skb),
9642                                       si->dst_reg, si->src_reg,
9643                                       offsetof(struct bpf_sock_ops_kern,
9644                                                skb));
9645                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9646                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
9647                                                        tcp_flags),
9648                                       si->dst_reg, si->dst_reg, off);
9649                 break;
9650         }
9651         return insn - insn_buf;
9652 }
9653
9654 /* data_end = skb->data + skb_headlen() */
9655 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
9656                                                     struct bpf_insn *insn)
9657 {
9658         /* si->dst_reg = skb->data */
9659         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9660                               si->dst_reg, si->src_reg,
9661                               offsetof(struct sk_buff, data));
9662         /* AX = skb->len */
9663         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
9664                               BPF_REG_AX, si->src_reg,
9665                               offsetof(struct sk_buff, len));
9666         /* si->dst_reg = skb->data + skb->len */
9667         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9668         /* AX = skb->data_len */
9669         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
9670                               BPF_REG_AX, si->src_reg,
9671                               offsetof(struct sk_buff, data_len));
9672         /* si->dst_reg = skb->data + skb->len - skb->data_len */
9673         *insn++ = BPF_ALU64_REG(BPF_SUB, si->dst_reg, BPF_REG_AX);
9674
9675         return insn;
9676 }
9677
9678 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
9679                                      const struct bpf_insn *si,
9680                                      struct bpf_insn *insn_buf,
9681                                      struct bpf_prog *prog, u32 *target_size)
9682 {
9683         struct bpf_insn *insn = insn_buf;
9684
9685         switch (si->off) {
9686         case offsetof(struct __sk_buff, data_end):
9687                 insn = bpf_convert_data_end_access(si, insn);
9688                 break;
9689         default:
9690                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9691                                               target_size);
9692         }
9693
9694         return insn - insn_buf;
9695 }
9696
9697 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
9698                                      const struct bpf_insn *si,
9699                                      struct bpf_insn *insn_buf,
9700                                      struct bpf_prog *prog, u32 *target_size)
9701 {
9702         struct bpf_insn *insn = insn_buf;
9703 #if IS_ENABLED(CONFIG_IPV6)
9704         int off;
9705 #endif
9706
9707         /* convert ctx uses the fact sg element is first in struct */
9708         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
9709
9710         switch (si->off) {
9711         case offsetof(struct sk_msg_md, data):
9712                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
9713                                       si->dst_reg, si->src_reg,
9714                                       offsetof(struct sk_msg, data));
9715                 break;
9716         case offsetof(struct sk_msg_md, data_end):
9717                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
9718                                       si->dst_reg, si->src_reg,
9719                                       offsetof(struct sk_msg, data_end));
9720                 break;
9721         case offsetof(struct sk_msg_md, family):
9722                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9723
9724                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9725                                               struct sk_msg, sk),
9726                                       si->dst_reg, si->src_reg,
9727                                       offsetof(struct sk_msg, sk));
9728                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9729                                       offsetof(struct sock_common, skc_family));
9730                 break;
9731
9732         case offsetof(struct sk_msg_md, remote_ip4):
9733                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9734
9735                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9736                                                 struct sk_msg, sk),
9737                                       si->dst_reg, si->src_reg,
9738                                       offsetof(struct sk_msg, sk));
9739                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9740                                       offsetof(struct sock_common, skc_daddr));
9741                 break;
9742
9743         case offsetof(struct sk_msg_md, local_ip4):
9744                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9745                                           skc_rcv_saddr) != 4);
9746
9747                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9748                                               struct sk_msg, sk),
9749                                       si->dst_reg, si->src_reg,
9750                                       offsetof(struct sk_msg, sk));
9751                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9752                                       offsetof(struct sock_common,
9753                                                skc_rcv_saddr));
9754                 break;
9755
9756         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
9757              offsetof(struct sk_msg_md, remote_ip6[3]):
9758 #if IS_ENABLED(CONFIG_IPV6)
9759                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9760                                           skc_v6_daddr.s6_addr32[0]) != 4);
9761
9762                 off = si->off;
9763                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
9764                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9765                                                 struct sk_msg, sk),
9766                                       si->dst_reg, si->src_reg,
9767                                       offsetof(struct sk_msg, sk));
9768                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9769                                       offsetof(struct sock_common,
9770                                                skc_v6_daddr.s6_addr32[0]) +
9771                                       off);
9772 #else
9773                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9774 #endif
9775                 break;
9776
9777         case offsetof(struct sk_msg_md, local_ip6[0]) ...
9778              offsetof(struct sk_msg_md, local_ip6[3]):
9779 #if IS_ENABLED(CONFIG_IPV6)
9780                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9781                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9782
9783                 off = si->off;
9784                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
9785                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9786                                                 struct sk_msg, sk),
9787                                       si->dst_reg, si->src_reg,
9788                                       offsetof(struct sk_msg, sk));
9789                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9790                                       offsetof(struct sock_common,
9791                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9792                                       off);
9793 #else
9794                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9795 #endif
9796                 break;
9797
9798         case offsetof(struct sk_msg_md, remote_port):
9799                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9800
9801                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9802                                                 struct sk_msg, sk),
9803                                       si->dst_reg, si->src_reg,
9804                                       offsetof(struct sk_msg, sk));
9805                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9806                                       offsetof(struct sock_common, skc_dport));
9807 #ifndef __BIG_ENDIAN_BITFIELD
9808                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9809 #endif
9810                 break;
9811
9812         case offsetof(struct sk_msg_md, local_port):
9813                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9814
9815                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
9816                                                 struct sk_msg, sk),
9817                                       si->dst_reg, si->src_reg,
9818                                       offsetof(struct sk_msg, sk));
9819                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9820                                       offsetof(struct sock_common, skc_num));
9821                 break;
9822
9823         case offsetof(struct sk_msg_md, size):
9824                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
9825                                       si->dst_reg, si->src_reg,
9826                                       offsetof(struct sk_msg_sg, size));
9827                 break;
9828
9829         case offsetof(struct sk_msg_md, sk):
9830                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
9831                                       si->dst_reg, si->src_reg,
9832                                       offsetof(struct sk_msg, sk));
9833                 break;
9834         }
9835
9836         return insn - insn_buf;
9837 }
9838
9839 const struct bpf_verifier_ops sk_filter_verifier_ops = {
9840         .get_func_proto         = sk_filter_func_proto,
9841         .is_valid_access        = sk_filter_is_valid_access,
9842         .convert_ctx_access     = bpf_convert_ctx_access,
9843         .gen_ld_abs             = bpf_gen_ld_abs,
9844 };
9845
9846 const struct bpf_prog_ops sk_filter_prog_ops = {
9847         .test_run               = bpf_prog_test_run_skb,
9848 };
9849
9850 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
9851         .get_func_proto         = tc_cls_act_func_proto,
9852         .is_valid_access        = tc_cls_act_is_valid_access,
9853         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
9854         .gen_prologue           = tc_cls_act_prologue,
9855         .gen_ld_abs             = bpf_gen_ld_abs,
9856         .check_kfunc_call       = bpf_prog_test_check_kfunc_call,
9857 };
9858
9859 const struct bpf_prog_ops tc_cls_act_prog_ops = {
9860         .test_run               = bpf_prog_test_run_skb,
9861 };
9862
9863 const struct bpf_verifier_ops xdp_verifier_ops = {
9864         .get_func_proto         = xdp_func_proto,
9865         .is_valid_access        = xdp_is_valid_access,
9866         .convert_ctx_access     = xdp_convert_ctx_access,
9867         .gen_prologue           = bpf_noop_prologue,
9868 };
9869
9870 const struct bpf_prog_ops xdp_prog_ops = {
9871         .test_run               = bpf_prog_test_run_xdp,
9872 };
9873
9874 const struct bpf_verifier_ops cg_skb_verifier_ops = {
9875         .get_func_proto         = cg_skb_func_proto,
9876         .is_valid_access        = cg_skb_is_valid_access,
9877         .convert_ctx_access     = bpf_convert_ctx_access,
9878 };
9879
9880 const struct bpf_prog_ops cg_skb_prog_ops = {
9881         .test_run               = bpf_prog_test_run_skb,
9882 };
9883
9884 const struct bpf_verifier_ops lwt_in_verifier_ops = {
9885         .get_func_proto         = lwt_in_func_proto,
9886         .is_valid_access        = lwt_is_valid_access,
9887         .convert_ctx_access     = bpf_convert_ctx_access,
9888 };
9889
9890 const struct bpf_prog_ops lwt_in_prog_ops = {
9891         .test_run               = bpf_prog_test_run_skb,
9892 };
9893
9894 const struct bpf_verifier_ops lwt_out_verifier_ops = {
9895         .get_func_proto         = lwt_out_func_proto,
9896         .is_valid_access        = lwt_is_valid_access,
9897         .convert_ctx_access     = bpf_convert_ctx_access,
9898 };
9899
9900 const struct bpf_prog_ops lwt_out_prog_ops = {
9901         .test_run               = bpf_prog_test_run_skb,
9902 };
9903
9904 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
9905         .get_func_proto         = lwt_xmit_func_proto,
9906         .is_valid_access        = lwt_is_valid_access,
9907         .convert_ctx_access     = bpf_convert_ctx_access,
9908         .gen_prologue           = tc_cls_act_prologue,
9909 };
9910
9911 const struct bpf_prog_ops lwt_xmit_prog_ops = {
9912         .test_run               = bpf_prog_test_run_skb,
9913 };
9914
9915 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
9916         .get_func_proto         = lwt_seg6local_func_proto,
9917         .is_valid_access        = lwt_is_valid_access,
9918         .convert_ctx_access     = bpf_convert_ctx_access,
9919 };
9920
9921 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
9922         .test_run               = bpf_prog_test_run_skb,
9923 };
9924
9925 const struct bpf_verifier_ops cg_sock_verifier_ops = {
9926         .get_func_proto         = sock_filter_func_proto,
9927         .is_valid_access        = sock_filter_is_valid_access,
9928         .convert_ctx_access     = bpf_sock_convert_ctx_access,
9929 };
9930
9931 const struct bpf_prog_ops cg_sock_prog_ops = {
9932 };
9933
9934 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
9935         .get_func_proto         = sock_addr_func_proto,
9936         .is_valid_access        = sock_addr_is_valid_access,
9937         .convert_ctx_access     = sock_addr_convert_ctx_access,
9938 };
9939
9940 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
9941 };
9942
9943 const struct bpf_verifier_ops sock_ops_verifier_ops = {
9944         .get_func_proto         = sock_ops_func_proto,
9945         .is_valid_access        = sock_ops_is_valid_access,
9946         .convert_ctx_access     = sock_ops_convert_ctx_access,
9947 };
9948
9949 const struct bpf_prog_ops sock_ops_prog_ops = {
9950 };
9951
9952 const struct bpf_verifier_ops sk_skb_verifier_ops = {
9953         .get_func_proto         = sk_skb_func_proto,
9954         .is_valid_access        = sk_skb_is_valid_access,
9955         .convert_ctx_access     = sk_skb_convert_ctx_access,
9956         .gen_prologue           = sk_skb_prologue,
9957 };
9958
9959 const struct bpf_prog_ops sk_skb_prog_ops = {
9960 };
9961
9962 const struct bpf_verifier_ops sk_msg_verifier_ops = {
9963         .get_func_proto         = sk_msg_func_proto,
9964         .is_valid_access        = sk_msg_is_valid_access,
9965         .convert_ctx_access     = sk_msg_convert_ctx_access,
9966         .gen_prologue           = bpf_noop_prologue,
9967 };
9968
9969 const struct bpf_prog_ops sk_msg_prog_ops = {
9970 };
9971
9972 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
9973         .get_func_proto         = flow_dissector_func_proto,
9974         .is_valid_access        = flow_dissector_is_valid_access,
9975         .convert_ctx_access     = flow_dissector_convert_ctx_access,
9976 };
9977
9978 const struct bpf_prog_ops flow_dissector_prog_ops = {
9979         .test_run               = bpf_prog_test_run_flow_dissector,
9980 };
9981
9982 int sk_detach_filter(struct sock *sk)
9983 {
9984         int ret = -ENOENT;
9985         struct sk_filter *filter;
9986
9987         if (sock_flag(sk, SOCK_FILTER_LOCKED))
9988                 return -EPERM;
9989
9990         filter = rcu_dereference_protected(sk->sk_filter,
9991                                            lockdep_sock_is_held(sk));
9992         if (filter) {
9993                 RCU_INIT_POINTER(sk->sk_filter, NULL);
9994                 sk_filter_uncharge(sk, filter);
9995                 ret = 0;
9996         }
9997
9998         return ret;
9999 }
10000 EXPORT_SYMBOL_GPL(sk_detach_filter);
10001
10002 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
10003                   unsigned int len)
10004 {
10005         struct sock_fprog_kern *fprog;
10006         struct sk_filter *filter;
10007         int ret = 0;
10008
10009         lock_sock(sk);
10010         filter = rcu_dereference_protected(sk->sk_filter,
10011                                            lockdep_sock_is_held(sk));
10012         if (!filter)
10013                 goto out;
10014
10015         /* We're copying the filter that has been originally attached,
10016          * so no conversion/decode needed anymore. eBPF programs that
10017          * have no original program cannot be dumped through this.
10018          */
10019         ret = -EACCES;
10020         fprog = filter->prog->orig_prog;
10021         if (!fprog)
10022                 goto out;
10023
10024         ret = fprog->len;
10025         if (!len)
10026                 /* User space only enquires number of filter blocks. */
10027                 goto out;
10028
10029         ret = -EINVAL;
10030         if (len < fprog->len)
10031                 goto out;
10032
10033         ret = -EFAULT;
10034         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
10035                 goto out;
10036
10037         /* Instead of bytes, the API requests to return the number
10038          * of filter blocks.
10039          */
10040         ret = fprog->len;
10041 out:
10042         release_sock(sk);
10043         return ret;
10044 }
10045
10046 #ifdef CONFIG_INET
10047 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10048                                     struct sock_reuseport *reuse,
10049                                     struct sock *sk, struct sk_buff *skb,
10050                                     struct sock *migrating_sk,
10051                                     u32 hash)
10052 {
10053         reuse_kern->skb = skb;
10054         reuse_kern->sk = sk;
10055         reuse_kern->selected_sk = NULL;
10056         reuse_kern->migrating_sk = migrating_sk;
10057         reuse_kern->data_end = skb->data + skb_headlen(skb);
10058         reuse_kern->hash = hash;
10059         reuse_kern->reuseport_id = reuse->reuseport_id;
10060         reuse_kern->bind_inany = reuse->bind_inany;
10061 }
10062
10063 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10064                                   struct bpf_prog *prog, struct sk_buff *skb,
10065                                   struct sock *migrating_sk,
10066                                   u32 hash)
10067 {
10068         struct sk_reuseport_kern reuse_kern;
10069         enum sk_action action;
10070
10071         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10072         action = BPF_PROG_RUN(prog, &reuse_kern);
10073
10074         if (action == SK_PASS)
10075                 return reuse_kern.selected_sk;
10076         else
10077                 return ERR_PTR(-ECONNREFUSED);
10078 }
10079
10080 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10081            struct bpf_map *, map, void *, key, u32, flags)
10082 {
10083         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10084         struct sock_reuseport *reuse;
10085         struct sock *selected_sk;
10086
10087         selected_sk = map->ops->map_lookup_elem(map, key);
10088         if (!selected_sk)
10089                 return -ENOENT;
10090
10091         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10092         if (!reuse) {
10093                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10094                 if (sk_is_refcounted(selected_sk))
10095                         sock_put(selected_sk);
10096
10097                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10098                  * The only (!reuse) case here is - the sk has already been
10099                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10100                  *
10101                  * Other maps (e.g. sock_map) do not provide this guarantee and
10102                  * the sk may never be in the reuseport group to begin with.
10103                  */
10104                 return is_sockarray ? -ENOENT : -EINVAL;
10105         }
10106
10107         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10108                 struct sock *sk = reuse_kern->sk;
10109
10110                 if (sk->sk_protocol != selected_sk->sk_protocol)
10111                         return -EPROTOTYPE;
10112                 else if (sk->sk_family != selected_sk->sk_family)
10113                         return -EAFNOSUPPORT;
10114
10115                 /* Catch all. Likely bound to a different sockaddr. */
10116                 return -EBADFD;
10117         }
10118
10119         reuse_kern->selected_sk = selected_sk;
10120
10121         return 0;
10122 }
10123
10124 static const struct bpf_func_proto sk_select_reuseport_proto = {
10125         .func           = sk_select_reuseport,
10126         .gpl_only       = false,
10127         .ret_type       = RET_INTEGER,
10128         .arg1_type      = ARG_PTR_TO_CTX,
10129         .arg2_type      = ARG_CONST_MAP_PTR,
10130         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10131         .arg4_type      = ARG_ANYTHING,
10132 };
10133
10134 BPF_CALL_4(sk_reuseport_load_bytes,
10135            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10136            void *, to, u32, len)
10137 {
10138         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10139 }
10140
10141 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10142         .func           = sk_reuseport_load_bytes,
10143         .gpl_only       = false,
10144         .ret_type       = RET_INTEGER,
10145         .arg1_type      = ARG_PTR_TO_CTX,
10146         .arg2_type      = ARG_ANYTHING,
10147         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10148         .arg4_type      = ARG_CONST_SIZE,
10149 };
10150
10151 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10152            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10153            void *, to, u32, len, u32, start_header)
10154 {
10155         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10156                                                len, start_header);
10157 }
10158
10159 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10160         .func           = sk_reuseport_load_bytes_relative,
10161         .gpl_only       = false,
10162         .ret_type       = RET_INTEGER,
10163         .arg1_type      = ARG_PTR_TO_CTX,
10164         .arg2_type      = ARG_ANYTHING,
10165         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10166         .arg4_type      = ARG_CONST_SIZE,
10167         .arg5_type      = ARG_ANYTHING,
10168 };
10169
10170 static const struct bpf_func_proto *
10171 sk_reuseport_func_proto(enum bpf_func_id func_id,
10172                         const struct bpf_prog *prog)
10173 {
10174         switch (func_id) {
10175         case BPF_FUNC_sk_select_reuseport:
10176                 return &sk_select_reuseport_proto;
10177         case BPF_FUNC_skb_load_bytes:
10178                 return &sk_reuseport_load_bytes_proto;
10179         case BPF_FUNC_skb_load_bytes_relative:
10180                 return &sk_reuseport_load_bytes_relative_proto;
10181         case BPF_FUNC_get_socket_cookie:
10182                 return &bpf_get_socket_ptr_cookie_proto;
10183         default:
10184                 return bpf_base_func_proto(func_id);
10185         }
10186 }
10187
10188 static bool
10189 sk_reuseport_is_valid_access(int off, int size,
10190                              enum bpf_access_type type,
10191                              const struct bpf_prog *prog,
10192                              struct bpf_insn_access_aux *info)
10193 {
10194         const u32 size_default = sizeof(__u32);
10195
10196         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10197             off % size || type != BPF_READ)
10198                 return false;
10199
10200         switch (off) {
10201         case offsetof(struct sk_reuseport_md, data):
10202                 info->reg_type = PTR_TO_PACKET;
10203                 return size == sizeof(__u64);
10204
10205         case offsetof(struct sk_reuseport_md, data_end):
10206                 info->reg_type = PTR_TO_PACKET_END;
10207                 return size == sizeof(__u64);
10208
10209         case offsetof(struct sk_reuseport_md, hash):
10210                 return size == size_default;
10211
10212         case offsetof(struct sk_reuseport_md, sk):
10213                 info->reg_type = PTR_TO_SOCKET;
10214                 return size == sizeof(__u64);
10215
10216         case offsetof(struct sk_reuseport_md, migrating_sk):
10217                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10218                 return size == sizeof(__u64);
10219
10220         /* Fields that allow narrowing */
10221         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10222                 if (size < sizeof_field(struct sk_buff, protocol))
10223                         return false;
10224                 fallthrough;
10225         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10226         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10227         case bpf_ctx_range(struct sk_reuseport_md, len):
10228                 bpf_ctx_record_field_size(info, size_default);
10229                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10230
10231         default:
10232                 return false;
10233         }
10234 }
10235
10236 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10237         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10238                               si->dst_reg, si->src_reg,                 \
10239                               bpf_target_off(struct sk_reuseport_kern, F, \
10240                                              sizeof_field(struct sk_reuseport_kern, F), \
10241                                              target_size));             \
10242         })
10243
10244 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10245         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10246                                     struct sk_buff,                     \
10247                                     skb,                                \
10248                                     SKB_FIELD)
10249
10250 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10251         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10252                                     struct sock,                        \
10253                                     sk,                                 \
10254                                     SK_FIELD)
10255
10256 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10257                                            const struct bpf_insn *si,
10258                                            struct bpf_insn *insn_buf,
10259                                            struct bpf_prog *prog,
10260                                            u32 *target_size)
10261 {
10262         struct bpf_insn *insn = insn_buf;
10263
10264         switch (si->off) {
10265         case offsetof(struct sk_reuseport_md, data):
10266                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10267                 break;
10268
10269         case offsetof(struct sk_reuseport_md, len):
10270                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10271                 break;
10272
10273         case offsetof(struct sk_reuseport_md, eth_protocol):
10274                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10275                 break;
10276
10277         case offsetof(struct sk_reuseport_md, ip_protocol):
10278                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10279                 break;
10280
10281         case offsetof(struct sk_reuseport_md, data_end):
10282                 SK_REUSEPORT_LOAD_FIELD(data_end);
10283                 break;
10284
10285         case offsetof(struct sk_reuseport_md, hash):
10286                 SK_REUSEPORT_LOAD_FIELD(hash);
10287                 break;
10288
10289         case offsetof(struct sk_reuseport_md, bind_inany):
10290                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10291                 break;
10292
10293         case offsetof(struct sk_reuseport_md, sk):
10294                 SK_REUSEPORT_LOAD_FIELD(sk);
10295                 break;
10296
10297         case offsetof(struct sk_reuseport_md, migrating_sk):
10298                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
10299                 break;
10300         }
10301
10302         return insn - insn_buf;
10303 }
10304
10305 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
10306         .get_func_proto         = sk_reuseport_func_proto,
10307         .is_valid_access        = sk_reuseport_is_valid_access,
10308         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
10309 };
10310
10311 const struct bpf_prog_ops sk_reuseport_prog_ops = {
10312 };
10313
10314 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
10315 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
10316
10317 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
10318            struct sock *, sk, u64, flags)
10319 {
10320         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
10321                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
10322                 return -EINVAL;
10323         if (unlikely(sk && sk_is_refcounted(sk)))
10324                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
10325         if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED))
10326                 return -ESOCKTNOSUPPORT; /* reject connected sockets */
10327
10328         /* Check if socket is suitable for packet L3/L4 protocol */
10329         if (sk && sk->sk_protocol != ctx->protocol)
10330                 return -EPROTOTYPE;
10331         if (sk && sk->sk_family != ctx->family &&
10332             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
10333                 return -EAFNOSUPPORT;
10334
10335         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
10336                 return -EEXIST;
10337
10338         /* Select socket as lookup result */
10339         ctx->selected_sk = sk;
10340         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
10341         return 0;
10342 }
10343
10344 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
10345         .func           = bpf_sk_lookup_assign,
10346         .gpl_only       = false,
10347         .ret_type       = RET_INTEGER,
10348         .arg1_type      = ARG_PTR_TO_CTX,
10349         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
10350         .arg3_type      = ARG_ANYTHING,
10351 };
10352
10353 static const struct bpf_func_proto *
10354 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
10355 {
10356         switch (func_id) {
10357         case BPF_FUNC_perf_event_output:
10358                 return &bpf_event_output_data_proto;
10359         case BPF_FUNC_sk_assign:
10360                 return &bpf_sk_lookup_assign_proto;
10361         case BPF_FUNC_sk_release:
10362                 return &bpf_sk_release_proto;
10363         default:
10364                 return bpf_sk_base_func_proto(func_id);
10365         }
10366 }
10367
10368 static bool sk_lookup_is_valid_access(int off, int size,
10369                                       enum bpf_access_type type,
10370                                       const struct bpf_prog *prog,
10371                                       struct bpf_insn_access_aux *info)
10372 {
10373         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
10374                 return false;
10375         if (off % size != 0)
10376                 return false;
10377         if (type != BPF_READ)
10378                 return false;
10379
10380         switch (off) {
10381         case offsetof(struct bpf_sk_lookup, sk):
10382                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
10383                 return size == sizeof(__u64);
10384
10385         case bpf_ctx_range(struct bpf_sk_lookup, family):
10386         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
10387         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
10388         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
10389         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
10390         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
10391         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
10392         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
10393                 bpf_ctx_record_field_size(info, sizeof(__u32));
10394                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
10395
10396         default:
10397                 return false;
10398         }
10399 }
10400
10401 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
10402                                         const struct bpf_insn *si,
10403                                         struct bpf_insn *insn_buf,
10404                                         struct bpf_prog *prog,
10405                                         u32 *target_size)
10406 {
10407         struct bpf_insn *insn = insn_buf;
10408
10409         switch (si->off) {
10410         case offsetof(struct bpf_sk_lookup, sk):
10411                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10412                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
10413                 break;
10414
10415         case offsetof(struct bpf_sk_lookup, family):
10416                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10417                                       bpf_target_off(struct bpf_sk_lookup_kern,
10418                                                      family, 2, target_size));
10419                 break;
10420
10421         case offsetof(struct bpf_sk_lookup, protocol):
10422                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10423                                       bpf_target_off(struct bpf_sk_lookup_kern,
10424                                                      protocol, 2, target_size));
10425                 break;
10426
10427         case offsetof(struct bpf_sk_lookup, remote_ip4):
10428                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10429                                       bpf_target_off(struct bpf_sk_lookup_kern,
10430                                                      v4.saddr, 4, target_size));
10431                 break;
10432
10433         case offsetof(struct bpf_sk_lookup, local_ip4):
10434                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10435                                       bpf_target_off(struct bpf_sk_lookup_kern,
10436                                                      v4.daddr, 4, target_size));
10437                 break;
10438
10439         case bpf_ctx_range_till(struct bpf_sk_lookup,
10440                                 remote_ip6[0], remote_ip6[3]): {
10441 #if IS_ENABLED(CONFIG_IPV6)
10442                 int off = si->off;
10443
10444                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
10445                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10446                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10447                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
10448                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10449                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10450 #else
10451                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10452 #endif
10453                 break;
10454         }
10455         case bpf_ctx_range_till(struct bpf_sk_lookup,
10456                                 local_ip6[0], local_ip6[3]): {
10457 #if IS_ENABLED(CONFIG_IPV6)
10458                 int off = si->off;
10459
10460                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
10461                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
10462                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
10463                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
10464                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10465                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
10466 #else
10467                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10468 #endif
10469                 break;
10470         }
10471         case offsetof(struct bpf_sk_lookup, remote_port):
10472                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10473                                       bpf_target_off(struct bpf_sk_lookup_kern,
10474                                                      sport, 2, target_size));
10475                 break;
10476
10477         case offsetof(struct bpf_sk_lookup, local_port):
10478                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
10479                                       bpf_target_off(struct bpf_sk_lookup_kern,
10480                                                      dport, 2, target_size));
10481                 break;
10482         }
10483
10484         return insn - insn_buf;
10485 }
10486
10487 const struct bpf_prog_ops sk_lookup_prog_ops = {
10488         .test_run = bpf_prog_test_run_sk_lookup,
10489 };
10490
10491 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
10492         .get_func_proto         = sk_lookup_func_proto,
10493         .is_valid_access        = sk_lookup_is_valid_access,
10494         .convert_ctx_access     = sk_lookup_convert_ctx_access,
10495 };
10496
10497 #endif /* CONFIG_INET */
10498
10499 DEFINE_BPF_DISPATCHER(xdp)
10500
10501 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
10502 {
10503         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
10504 }
10505
10506 #ifdef CONFIG_DEBUG_INFO_BTF
10507 BTF_ID_LIST_GLOBAL(btf_sock_ids)
10508 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
10509 BTF_SOCK_TYPE_xxx
10510 #undef BTF_SOCK_TYPE
10511 #else
10512 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE];
10513 #endif
10514
10515 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
10516 {
10517         /* tcp6_sock type is not generated in dwarf and hence btf,
10518          * trigger an explicit type generation here.
10519          */
10520         BTF_TYPE_EMIT(struct tcp6_sock);
10521         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
10522             sk->sk_family == AF_INET6)
10523                 return (unsigned long)sk;
10524
10525         return (unsigned long)NULL;
10526 }
10527
10528 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
10529         .func                   = bpf_skc_to_tcp6_sock,
10530         .gpl_only               = false,
10531         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10532         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10533         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
10534 };
10535
10536 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
10537 {
10538         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
10539                 return (unsigned long)sk;
10540
10541         return (unsigned long)NULL;
10542 }
10543
10544 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
10545         .func                   = bpf_skc_to_tcp_sock,
10546         .gpl_only               = false,
10547         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10548         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10549         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
10550 };
10551
10552 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
10553 {
10554         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
10555          * generated if CONFIG_INET=n. Trigger an explicit generation here.
10556          */
10557         BTF_TYPE_EMIT(struct inet_timewait_sock);
10558         BTF_TYPE_EMIT(struct tcp_timewait_sock);
10559
10560 #ifdef CONFIG_INET
10561         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
10562                 return (unsigned long)sk;
10563 #endif
10564
10565 #if IS_BUILTIN(CONFIG_IPV6)
10566         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
10567                 return (unsigned long)sk;
10568 #endif
10569
10570         return (unsigned long)NULL;
10571 }
10572
10573 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
10574         .func                   = bpf_skc_to_tcp_timewait_sock,
10575         .gpl_only               = false,
10576         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10577         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10578         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
10579 };
10580
10581 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
10582 {
10583 #ifdef CONFIG_INET
10584         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10585                 return (unsigned long)sk;
10586 #endif
10587
10588 #if IS_BUILTIN(CONFIG_IPV6)
10589         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
10590                 return (unsigned long)sk;
10591 #endif
10592
10593         return (unsigned long)NULL;
10594 }
10595
10596 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
10597         .func                   = bpf_skc_to_tcp_request_sock,
10598         .gpl_only               = false,
10599         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10600         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10601         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
10602 };
10603
10604 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
10605 {
10606         /* udp6_sock type is not generated in dwarf and hence btf,
10607          * trigger an explicit type generation here.
10608          */
10609         BTF_TYPE_EMIT(struct udp6_sock);
10610         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
10611             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
10612                 return (unsigned long)sk;
10613
10614         return (unsigned long)NULL;
10615 }
10616
10617 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
10618         .func                   = bpf_skc_to_udp6_sock,
10619         .gpl_only               = false,
10620         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
10621         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
10622         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
10623 };
10624
10625 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
10626 {
10627         return (unsigned long)sock_from_file(file);
10628 }
10629
10630 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
10631 BTF_ID(struct, socket)
10632 BTF_ID(struct, file)
10633
10634 const struct bpf_func_proto bpf_sock_from_file_proto = {
10635         .func           = bpf_sock_from_file,
10636         .gpl_only       = false,
10637         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
10638         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
10639         .arg1_type      = ARG_PTR_TO_BTF_ID,
10640         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
10641 };
10642
10643 static const struct bpf_func_proto *
10644 bpf_sk_base_func_proto(enum bpf_func_id func_id)
10645 {
10646         const struct bpf_func_proto *func;
10647
10648         switch (func_id) {
10649         case BPF_FUNC_skc_to_tcp6_sock:
10650                 func = &bpf_skc_to_tcp6_sock_proto;
10651                 break;
10652         case BPF_FUNC_skc_to_tcp_sock:
10653                 func = &bpf_skc_to_tcp_sock_proto;
10654                 break;
10655         case BPF_FUNC_skc_to_tcp_timewait_sock:
10656                 func = &bpf_skc_to_tcp_timewait_sock_proto;
10657                 break;
10658         case BPF_FUNC_skc_to_tcp_request_sock:
10659                 func = &bpf_skc_to_tcp_request_sock_proto;
10660                 break;
10661         case BPF_FUNC_skc_to_udp6_sock:
10662                 func = &bpf_skc_to_udp6_sock_proto;
10663                 break;
10664         default:
10665                 return bpf_base_func_proto(func_id);
10666         }
10667
10668         if (!perfmon_capable())
10669                 return NULL;
10670
10671         return func;
10672 }