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 #include <net/xdp.h>
  81 #include <net/mptcp.h>
  82
  83 static const struct bpf_func_proto *
  84 bpf_sk_base_func_proto(enum bpf_func_id func_id);
  85
  86 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len)
  87 {
  88         if (in_compat_syscall()) {
  89                 struct compat_sock_fprog f32;
  90
  91                 if (len != sizeof(f32))
  92                         return -EINVAL;
  93                 if (copy_from_sockptr(&f32, src, sizeof(f32)))
  94                         return -EFAULT;
  95                 memset(dst, 0, sizeof(*dst));
  96                 dst->len = f32.len;
  97                 dst->filter = compat_ptr(f32.filter);
  98         } else {
  99                 if (len != sizeof(*dst))
 100                         return -EINVAL;
 101                 if (copy_from_sockptr(dst, src, sizeof(*dst)))
 102                         return -EFAULT;
 103         }
 104
 105         return 0;
 106 }
 107 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user);
 108
 109 /**
 110  *      sk_filter_trim_cap - run a packet through a socket filter
 111  *      @sk: sock associated with &sk_buff
 112  *      @skb: buffer to filter
 113  *      @cap: limit on how short the eBPF program may trim the packet
 114  *
 115  * Run the eBPF program and then cut skb->data to correct size returned by
 116  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
 117  * than pkt_len we keep whole skb->data. This is the socket level
 118  * wrapper to bpf_prog_run. It returns 0 if the packet should
 119  * be accepted or -EPERM if the packet should be tossed.
 120  *
 121  */
 122 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
 123 {
 124         int err;
 125         struct sk_filter *filter;
 126
 127         /*
 128          * If the skb was allocated from pfmemalloc reserves, only
 129          * allow SOCK_MEMALLOC sockets to use it as this socket is
 130          * helping free memory
 131          */
 132         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 133                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 134                 return -ENOMEM;
 135         }
 136         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 137         if (err)
 138                 return err;
 139
 140         err = security_sock_rcv_skb(sk, skb);
 141         if (err)
 142                 return err;
 143
 144         rcu_read_lock();
 145         filter = rcu_dereference(sk->sk_filter);
 146         if (filter) {
 147                 struct sock *save_sk = skb->sk;
 148                 unsigned int pkt_len;
 149
 150                 skb->sk = sk;
 151                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 152                 skb->sk = save_sk;
 153                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 154         }
 155         rcu_read_unlock();
 156
 157         return err;
 158 }
 159 EXPORT_SYMBOL(sk_filter_trim_cap);
 160
 161 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 162 {
 163         return skb_get_poff(skb);
 164 }
 165
 166 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 167 {
 168         struct nlattr *nla;
 169
 170         if (skb_is_nonlinear(skb))
 171                 return 0;
 172
 173         if (skb->len < sizeof(struct nlattr))
 174                 return 0;
 175
 176         if (a > skb->len - sizeof(struct nlattr))
 177                 return 0;
 178
 179         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 180         if (nla)
 181                 return (void *) nla - (void *) skb->data;
 182
 183         return 0;
 184 }
 185
 186 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 187 {
 188         struct nlattr *nla;
 189
 190         if (skb_is_nonlinear(skb))
 191                 return 0;
 192
 193         if (skb->len < sizeof(struct nlattr))
 194                 return 0;
 195
 196         if (a > skb->len - sizeof(struct nlattr))
 197                 return 0;
 198
 199         nla = (struct nlattr *) &skb->data[a];
 200         if (nla->nla_len > skb->len - a)
 201                 return 0;
 202
 203         nla = nla_find_nested(nla, x);
 204         if (nla)
 205                 return (void *) nla - (void *) skb->data;
 206
 207         return 0;
 208 }
 209
 210 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 211            data, int, headlen, int, offset)
 212 {
 213         u8 tmp, *ptr;
 214         const int len = sizeof(tmp);
 215
 216         if (offset >= 0) {
 217                 if (headlen - offset >= len)
 218                         return *(u8 *)(data + offset);
 219                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 220                         return tmp;
 221         } else {
 222                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 223                 if (likely(ptr))
 224                         return *(u8 *)ptr;
 225         }
 226
 227         return -EFAULT;
 228 }
 229
 230 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 231            int, offset)
 232 {
 233         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 234                                          offset);
 235 }
 236
 237 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 238            data, int, headlen, int, offset)
 239 {
 240         __be16 tmp, *ptr;
 241         const int len = sizeof(tmp);
 242
 243         if (offset >= 0) {
 244                 if (headlen - offset >= len)
 245                         return get_unaligned_be16(data + offset);
 246                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 247                         return be16_to_cpu(tmp);
 248         } else {
 249                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 250                 if (likely(ptr))
 251                         return get_unaligned_be16(ptr);
 252         }
 253
 254         return -EFAULT;
 255 }
 256
 257 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 258            int, offset)
 259 {
 260         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 261                                           offset);
 262 }
 263
 264 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 265            data, int, headlen, int, offset)
 266 {
 267         __be32 tmp, *ptr;
 268         const int len = sizeof(tmp);
 269
 270         if (likely(offset >= 0)) {
 271                 if (headlen - offset >= len)
 272                         return get_unaligned_be32(data + offset);
 273                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 274                         return be32_to_cpu(tmp);
 275         } else {
 276                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 277                 if (likely(ptr))
 278                         return get_unaligned_be32(ptr);
 279         }
 280
 281         return -EFAULT;
 282 }
 283
 284 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 285            int, offset)
 286 {
 287         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 288                                           offset);
 289 }
 290
 291 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 292                               struct bpf_insn *insn_buf)
 293 {
 294         struct bpf_insn *insn = insn_buf;
 295
 296         switch (skb_field) {
 297         case SKF_AD_MARK:
 298                 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4);
 299
 300                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 301                                       offsetof(struct sk_buff, mark));
 302                 break;
 303
 304         case SKF_AD_PKTTYPE:
 305                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET);
 306                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 307 #ifdef __BIG_ENDIAN_BITFIELD
 308                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 309 #endif
 310                 break;
 311
 312         case SKF_AD_QUEUE:
 313                 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2);
 314
 315                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 316                                       offsetof(struct sk_buff, queue_mapping));
 317                 break;
 318
 319         case SKF_AD_VLAN_TAG:
 320                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2);
 321
 322                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 323                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 324                                       offsetof(struct sk_buff, vlan_tci));
 325                 break;
 326         case SKF_AD_VLAN_TAG_PRESENT:
 327                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET);
 328                 if (PKT_VLAN_PRESENT_BIT)
 329                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT);
 330                 if (PKT_VLAN_PRESENT_BIT < 7)
 331                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 332                 break;
 333         }
 334
 335         return insn - insn_buf;
 336 }
 337
 338 static bool convert_bpf_extensions(struct sock_filter *fp,
 339                                    struct bpf_insn **insnp)
 340 {
 341         struct bpf_insn *insn = *insnp;
 342         u32 cnt;
 343
 344         switch (fp->k) {
 345         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 346                 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2);
 347
 348                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 349                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 350                                       offsetof(struct sk_buff, protocol));
 351                 /* A = ntohs(A) [emitting a nop or swap16] */
 352                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 353                 break;
 354
 355         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 356                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 357                 insn += cnt - 1;
 358                 break;
 359
 360         case SKF_AD_OFF + SKF_AD_IFINDEX:
 361         case SKF_AD_OFF + SKF_AD_HATYPE:
 362                 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4);
 363                 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2);
 364
 365                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 366                                       BPF_REG_TMP, BPF_REG_CTX,
 367                                       offsetof(struct sk_buff, dev));
 368                 /* if (tmp != 0) goto pc + 1 */
 369                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 370                 *insn++ = BPF_EXIT_INSN();
 371                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 372                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 373                                             offsetof(struct net_device, ifindex));
 374                 else
 375                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 376                                             offsetof(struct net_device, type));
 377                 break;
 378
 379         case SKF_AD_OFF + SKF_AD_MARK:
 380                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 381                 insn += cnt - 1;
 382                 break;
 383
 384         case SKF_AD_OFF + SKF_AD_RXHASH:
 385                 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4);
 386
 387                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 388                                     offsetof(struct sk_buff, hash));
 389                 break;
 390
 391         case SKF_AD_OFF + SKF_AD_QUEUE:
 392                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 393                 insn += cnt - 1;
 394                 break;
 395
 396         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 397                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 398                                          BPF_REG_A, BPF_REG_CTX, insn);
 399                 insn += cnt - 1;
 400                 break;
 401
 402         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 403                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 404                                          BPF_REG_A, BPF_REG_CTX, insn);
 405                 insn += cnt - 1;
 406                 break;
 407
 408         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 409                 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2);
 410
 411                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 412                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 413                                       offsetof(struct sk_buff, vlan_proto));
 414                 /* A = ntohs(A) [emitting a nop or swap16] */
 415                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 416                 break;
 417
 418         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 419         case SKF_AD_OFF + SKF_AD_NLATTR:
 420         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 421         case SKF_AD_OFF + SKF_AD_CPU:
 422         case SKF_AD_OFF + SKF_AD_RANDOM:
 423                 /* arg1 = CTX */
 424                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 425                 /* arg2 = A */
 426                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 427                 /* arg3 = X */
 428                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 429                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 430                 switch (fp->k) {
 431                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 432                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 433                         break;
 434                 case SKF_AD_OFF + SKF_AD_NLATTR:
 435                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 436                         break;
 437                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 438                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 439                         break;
 440                 case SKF_AD_OFF + SKF_AD_CPU:
 441                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 442                         break;
 443                 case SKF_AD_OFF + SKF_AD_RANDOM:
 444                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 445                         bpf_user_rnd_init_once();
 446                         break;
 447                 }
 448                 break;
 449
 450         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 451                 /* A ^= X */
 452                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 453                 break;
 454
 455         default:
 456                 /* This is just a dummy call to avoid letting the compiler
 457                  * evict __bpf_call_base() as an optimization. Placed here
 458                  * where no-one bothers.
 459                  */
 460                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 461                 return false;
 462         }
 463
 464         *insnp = insn;
 465         return true;
 466 }
 467
 468 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 469 {
 470         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 471         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 472         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 473                       BPF_SIZE(fp->code) == BPF_W;
 474         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 475         const int ip_align = NET_IP_ALIGN;
 476         struct bpf_insn *insn = *insnp;
 477         int offset = fp->k;
 478
 479         if (!indirect &&
 480             ((unaligned_ok && offset >= 0) ||
 481              (!unaligned_ok && offset >= 0 &&
 482               offset + ip_align >= 0 &&
 483               offset + ip_align % size == 0))) {
 484                 bool ldx_off_ok = offset <= S16_MAX;
 485
 486                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 487                 if (offset)
 488                         *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 489                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 490                                       size, 2 + endian + (!ldx_off_ok * 2));
 491                 if (ldx_off_ok) {
 492                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 493                                               BPF_REG_D, offset);
 494                 } else {
 495                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 496                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 497                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 498                                               BPF_REG_TMP, 0);
 499                 }
 500                 if (endian)
 501                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 502                 *insn++ = BPF_JMP_A(8);
 503         }
 504
 505         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 506         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 507         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 508         if (!indirect) {
 509                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 510         } else {
 511                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 512                 if (fp->k)
 513                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 514         }
 515
 516         switch (BPF_SIZE(fp->code)) {
 517         case BPF_B:
 518                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 519                 break;
 520         case BPF_H:
 521                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 522                 break;
 523         case BPF_W:
 524                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 525                 break;
 526         default:
 527                 return false;
 528         }
 529
 530         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 531         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 532         *insn   = BPF_EXIT_INSN();
 533
 534         *insnp = insn;
 535         return true;
 536 }
 537
 538 /**
 539  *      bpf_convert_filter - convert filter program
 540  *      @prog: the user passed filter program
 541  *      @len: the length of the user passed filter program
 542  *      @new_prog: allocated 'struct bpf_prog' or NULL
 543  *      @new_len: pointer to store length of converted program
 544  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 545  *
 546  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 547  * style extended BPF (eBPF).
 548  * Conversion workflow:
 549  *
 550  * 1) First pass for calculating the new program length:
 551  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 552  *
 553  * 2) 2nd pass to remap in two passes: 1st pass finds new
 554  *    jump offsets, 2nd pass remapping:
 555  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 556  */
 557 static int bpf_convert_filter(struct sock_filter *prog, int len,
 558                               struct bpf_prog *new_prog, int *new_len,
 559                               bool *seen_ld_abs)
 560 {
 561         int new_flen = 0, pass = 0, target, i, stack_off;
 562         struct bpf_insn *new_insn, *first_insn = NULL;
 563         struct sock_filter *fp;
 564         int *addrs = NULL;
 565         u8 bpf_src;
 566
 567         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 568         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 569
 570         if (len <= 0 || len > BPF_MAXINSNS)
 571                 return -EINVAL;
 572
 573         if (new_prog) {
 574                 first_insn = new_prog->insnsi;
 575                 addrs = kcalloc(len, sizeof(*addrs),
 576                                 GFP_KERNEL | __GFP_NOWARN);
 577                 if (!addrs)
 578                         return -ENOMEM;
 579         }
 580
 581 do_pass:
 582         new_insn = first_insn;
 583         fp = prog;
 584
 585         /* Classic BPF related prologue emission. */
 586         if (new_prog) {
 587                 /* Classic BPF expects A and X to be reset first. These need
 588                  * to be guaranteed to be the first two instructions.
 589                  */
 590                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 591                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 592
 593                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 594                  * In eBPF case it's done by the compiler, here we need to
 595                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 596                  */
 597                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 598                 if (*seen_ld_abs) {
 599                         /* For packet access in classic BPF, cache skb->data
 600                          * in callee-saved BPF R8 and skb->len - skb->data_len
 601                          * (headlen) in BPF R9. Since classic BPF is read-only
 602                          * on CTX, we only need to cache it once.
 603                          */
 604                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 605                                                   BPF_REG_D, BPF_REG_CTX,
 606                                                   offsetof(struct sk_buff, data));
 607                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 608                                                   offsetof(struct sk_buff, len));
 609                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 610                                                   offsetof(struct sk_buff, data_len));
 611                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 612                 }
 613         } else {
 614                 new_insn += 3;
 615         }
 616
 617         for (i = 0; i < len; fp++, i++) {
 618                 struct bpf_insn tmp_insns[32] = { };
 619                 struct bpf_insn *insn = tmp_insns;
 620
 621                 if (addrs)
 622                         addrs[i] = new_insn - first_insn;
 623
 624                 switch (fp->code) {
 625                 /* All arithmetic insns and skb loads map as-is. */
 626                 case BPF_ALU | BPF_ADD | BPF_X:
 627                 case BPF_ALU | BPF_ADD | BPF_K:
 628                 case BPF_ALU | BPF_SUB | BPF_X:
 629                 case BPF_ALU | BPF_SUB | BPF_K:
 630                 case BPF_ALU | BPF_AND | BPF_X:
 631                 case BPF_ALU | BPF_AND | BPF_K:
 632                 case BPF_ALU | BPF_OR | BPF_X:
 633                 case BPF_ALU | BPF_OR | BPF_K:
 634                 case BPF_ALU | BPF_LSH | BPF_X:
 635                 case BPF_ALU | BPF_LSH | BPF_K:
 636                 case BPF_ALU | BPF_RSH | BPF_X:
 637                 case BPF_ALU | BPF_RSH | BPF_K:
 638                 case BPF_ALU | BPF_XOR | BPF_X:
 639                 case BPF_ALU | BPF_XOR | BPF_K:
 640                 case BPF_ALU | BPF_MUL | BPF_X:
 641                 case BPF_ALU | BPF_MUL | BPF_K:
 642                 case BPF_ALU | BPF_DIV | BPF_X:
 643                 case BPF_ALU | BPF_DIV | BPF_K:
 644                 case BPF_ALU | BPF_MOD | BPF_X:
 645                 case BPF_ALU | BPF_MOD | BPF_K:
 646                 case BPF_ALU | BPF_NEG:
 647                 case BPF_LD | BPF_ABS | BPF_W:
 648                 case BPF_LD | BPF_ABS | BPF_H:
 649                 case BPF_LD | BPF_ABS | BPF_B:
 650                 case BPF_LD | BPF_IND | BPF_W:
 651                 case BPF_LD | BPF_IND | BPF_H:
 652                 case BPF_LD | BPF_IND | BPF_B:
 653                         /* Check for overloaded BPF extension and
 654                          * directly convert it if found, otherwise
 655                          * just move on with mapping.
 656                          */
 657                         if (BPF_CLASS(fp->code) == BPF_LD &&
 658                             BPF_MODE(fp->code) == BPF_ABS &&
 659                             convert_bpf_extensions(fp, &insn))
 660                                 break;
 661                         if (BPF_CLASS(fp->code) == BPF_LD &&
 662                             convert_bpf_ld_abs(fp, &insn)) {
 663                                 *seen_ld_abs = true;
 664                                 break;
 665                         }
 666
 667                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 668                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 669                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 670                                 /* Error with exception code on div/mod by 0.
 671                                  * For cBPF programs, this was always return 0.
 672                                  */
 673                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 674                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 675                                 *insn++ = BPF_EXIT_INSN();
 676                         }
 677
 678                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 679                         break;
 680
 681                 /* Jump transformation cannot use BPF block macros
 682                  * everywhere as offset calculation and target updates
 683                  * require a bit more work than the rest, i.e. jump
 684                  * opcodes map as-is, but offsets need adjustment.
 685                  */
 686
 687 #define BPF_EMIT_JMP                                                    \
 688         do {                                                            \
 689                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 690                 s32 off;                                                \
 691                                                                         \
 692                 if (target >= len || target < 0)                        \
 693                         goto err;                                       \
 694                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 695                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 696                 off -= insn - tmp_insns;                                \
 697                 /* Reject anything not fitting into insn->off. */       \
 698                 if (off < off_min || off > off_max)                     \
 699                         goto err;                                       \
 700                 insn->off = off;                                        \
 701         } while (0)
 702
 703                 case BPF_JMP | BPF_JA:
 704                         target = i + fp->k + 1;
 705                         insn->code = fp->code;
 706                         BPF_EMIT_JMP;
 707                         break;
 708
 709                 case BPF_JMP | BPF_JEQ | BPF_K:
 710                 case BPF_JMP | BPF_JEQ | BPF_X:
 711                 case BPF_JMP | BPF_JSET | BPF_K:
 712                 case BPF_JMP | BPF_JSET | BPF_X:
 713                 case BPF_JMP | BPF_JGT | BPF_K:
 714                 case BPF_JMP | BPF_JGT | BPF_X:
 715                 case BPF_JMP | BPF_JGE | BPF_K:
 716                 case BPF_JMP | BPF_JGE | BPF_X:
 717                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 718                                 /* BPF immediates are signed, zero extend
 719                                  * immediate into tmp register and use it
 720                                  * in compare insn.
 721                                  */
 722                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 723
 724                                 insn->dst_reg = BPF_REG_A;
 725                                 insn->src_reg = BPF_REG_TMP;
 726                                 bpf_src = BPF_X;
 727                         } else {
 728                                 insn->dst_reg = BPF_REG_A;
 729                                 insn->imm = fp->k;
 730                                 bpf_src = BPF_SRC(fp->code);
 731                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 732                         }
 733
 734                         /* Common case where 'jump_false' is next insn. */
 735                         if (fp->jf == 0) {
 736                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 737                                 target = i + fp->jt + 1;
 738                                 BPF_EMIT_JMP;
 739                                 break;
 740                         }
 741
 742                         /* Convert some jumps when 'jump_true' is next insn. */
 743                         if (fp->jt == 0) {
 744                                 switch (BPF_OP(fp->code)) {
 745                                 case BPF_JEQ:
 746                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 747                                         break;
 748                                 case BPF_JGT:
 749                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 750                                         break;
 751                                 case BPF_JGE:
 752                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 753                                         break;
 754                                 default:
 755                                         goto jmp_rest;
 756                                 }
 757
 758                                 target = i + fp->jf + 1;
 759                                 BPF_EMIT_JMP;
 760                                 break;
 761                         }
 762 jmp_rest:
 763                         /* Other jumps are mapped into two insns: Jxx and JA. */
 764                         target = i + fp->jt + 1;
 765                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 766                         BPF_EMIT_JMP;
 767                         insn++;
 768
 769                         insn->code = BPF_JMP | BPF_JA;
 770                         target = i + fp->jf + 1;
 771                         BPF_EMIT_JMP;
 772                         break;
 773
 774                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 775                 case BPF_LDX | BPF_MSH | BPF_B: {
 776                         struct sock_filter tmp = {
 777                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 778                                 .k      = fp->k,
 779                         };
 780
 781                         *seen_ld_abs = true;
 782
 783                         /* X = A */
 784                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 785                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 786                         convert_bpf_ld_abs(&tmp, &insn);
 787                         insn++;
 788                         /* A &= 0xf */
 789                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 790                         /* A <<= 2 */
 791                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 792                         /* tmp = X */
 793                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 794                         /* X = A */
 795                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 796                         /* A = tmp */
 797                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 798                         break;
 799                 }
 800                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 801                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 802                  */
 803                 case BPF_RET | BPF_A:
 804                 case BPF_RET | BPF_K:
 805                         if (BPF_RVAL(fp->code) == BPF_K)
 806                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 807                                                         0, fp->k);
 808                         *insn = BPF_EXIT_INSN();
 809                         break;
 810
 811                 /* Store to stack. */
 812                 case BPF_ST:
 813                 case BPF_STX:
 814                         stack_off = fp->k * 4  + 4;
 815                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 816                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 817                                             -stack_off);
 818                         /* check_load_and_stores() verifies that classic BPF can
 819                          * load from stack only after write, so tracking
 820                          * stack_depth for ST|STX insns is enough
 821                          */
 822                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 823                                 new_prog->aux->stack_depth = stack_off;
 824                         break;
 825
 826                 /* Load from stack. */
 827                 case BPF_LD | BPF_MEM:
 828                 case BPF_LDX | BPF_MEM:
 829                         stack_off = fp->k * 4  + 4;
 830                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 831                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 832                                             -stack_off);
 833                         break;
 834
 835                 /* A = K or X = K */
 836                 case BPF_LD | BPF_IMM:
 837                 case BPF_LDX | BPF_IMM:
 838                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 839                                               BPF_REG_A : BPF_REG_X, fp->k);
 840                         break;
 841
 842                 /* X = A */
 843                 case BPF_MISC | BPF_TAX:
 844                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 845                         break;
 846
 847                 /* A = X */
 848                 case BPF_MISC | BPF_TXA:
 849                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 850                         break;
 851
 852                 /* A = skb->len or X = skb->len */
 853                 case BPF_LD | BPF_W | BPF_LEN:
 854                 case BPF_LDX | BPF_W | BPF_LEN:
 855                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 856                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 857                                             offsetof(struct sk_buff, len));
 858                         break;
 859
 860                 /* Access seccomp_data fields. */
 861                 case BPF_LDX | BPF_ABS | BPF_W:
 862                         /* A = *(u32 *) (ctx + K) */
 863                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 864                         break;
 865
 866                 /* Unknown instruction. */
 867                 default:
 868                         goto err;
 869                 }
 870
 871                 insn++;
 872                 if (new_prog)
 873                         memcpy(new_insn, tmp_insns,
 874                                sizeof(*insn) * (insn - tmp_insns));
 875                 new_insn += insn - tmp_insns;
 876         }
 877
 878         if (!new_prog) {
 879                 /* Only calculating new length. */
 880                 *new_len = new_insn - first_insn;
 881                 if (*seen_ld_abs)
 882                         *new_len += 4; /* Prologue bits. */
 883                 return 0;
 884         }
 885
 886         pass++;
 887         if (new_flen != new_insn - first_insn) {
 888                 new_flen = new_insn - first_insn;
 889                 if (pass > 2)
 890                         goto err;
 891                 goto do_pass;
 892         }
 893
 894         kfree(addrs);
 895         BUG_ON(*new_len != new_flen);
 896         return 0;
 897 err:
 898         kfree(addrs);
 899         return -EINVAL;
 900 }
 901
 902 /* Security:
 903  *
 904  * As we dont want to clear mem[] array for each packet going through
 905  * __bpf_prog_run(), we check that filter loaded by user never try to read
 906  * a cell if not previously written, and we check all branches to be sure
 907  * a malicious user doesn't try to abuse us.
 908  */
 909 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 910 {
 911         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 912         int pc, ret = 0;
 913
 914         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 915
 916         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 917         if (!masks)
 918                 return -ENOMEM;
 919
 920         memset(masks, 0xff, flen * sizeof(*masks));
 921
 922         for (pc = 0; pc < flen; pc++) {
 923                 memvalid &= masks[pc];
 924
 925                 switch (filter[pc].code) {
 926                 case BPF_ST:
 927                 case BPF_STX:
 928                         memvalid |= (1 << filter[pc].k);
 929                         break;
 930                 case BPF_LD | BPF_MEM:
 931                 case BPF_LDX | BPF_MEM:
 932                         if (!(memvalid & (1 << filter[pc].k))) {
 933                                 ret = -EINVAL;
 934                                 goto error;
 935                         }
 936                         break;
 937                 case BPF_JMP | BPF_JA:
 938                         /* A jump must set masks on target */
 939                         masks[pc + 1 + filter[pc].k] &= memvalid;
 940                         memvalid = ~0;
 941                         break;
 942                 case BPF_JMP | BPF_JEQ | BPF_K:
 943                 case BPF_JMP | BPF_JEQ | BPF_X:
 944                 case BPF_JMP | BPF_JGE | BPF_K:
 945                 case BPF_JMP | BPF_JGE | BPF_X:
 946                 case BPF_JMP | BPF_JGT | BPF_K:
 947                 case BPF_JMP | BPF_JGT | BPF_X:
 948                 case BPF_JMP | BPF_JSET | BPF_K:
 949                 case BPF_JMP | BPF_JSET | BPF_X:
 950                         /* A jump must set masks on targets */
 951                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 952                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 953                         memvalid = ~0;
 954                         break;
 955                 }
 956         }
 957 error:
 958         kfree(masks);
 959         return ret;
 960 }
 961
 962 static bool chk_code_allowed(u16 code_to_probe)
 963 {
 964         static const bool codes[] = {
 965                 /* 32 bit ALU operations */
 966                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 967                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 968                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 969                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 970                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 971                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 972                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 973                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 974                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 975                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 976                 [BPF_ALU | BPF_AND | BPF_K] = true,
 977                 [BPF_ALU | BPF_AND | BPF_X] = true,
 978                 [BPF_ALU | BPF_OR | BPF_K] = true,
 979                 [BPF_ALU | BPF_OR | BPF_X] = true,
 980                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 981                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 982                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 983                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 984                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 985                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 986                 [BPF_ALU | BPF_NEG] = true,
 987                 /* Load instructions */
 988                 [BPF_LD | BPF_W | BPF_ABS] = true,
 989                 [BPF_LD | BPF_H | BPF_ABS] = true,
 990                 [BPF_LD | BPF_B | BPF_ABS] = true,
 991                 [BPF_LD | BPF_W | BPF_LEN] = true,
 992                 [BPF_LD | BPF_W | BPF_IND] = true,
 993                 [BPF_LD | BPF_H | BPF_IND] = true,
 994                 [BPF_LD | BPF_B | BPF_IND] = true,
 995                 [BPF_LD | BPF_IMM] = true,
 996                 [BPF_LD | BPF_MEM] = true,
 997                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 998                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 999                 [BPF_LDX | BPF_IMM] = true,
1000                 [BPF_LDX | BPF_MEM] = true,
1001                 /* Store instructions */
1002                 [BPF_ST] = true,
1003                 [BPF_STX] = true,
1004                 /* Misc instructions */
1005                 [BPF_MISC | BPF_TAX] = true,
1006                 [BPF_MISC | BPF_TXA] = true,
1007                 /* Return instructions */
1008                 [BPF_RET | BPF_K] = true,
1009                 [BPF_RET | BPF_A] = true,
1010                 /* Jump instructions */
1011                 [BPF_JMP | BPF_JA] = true,
1012                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
1013                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
1014                 [BPF_JMP | BPF_JGE | BPF_K] = true,
1015                 [BPF_JMP | BPF_JGE | BPF_X] = true,
1016                 [BPF_JMP | BPF_JGT | BPF_K] = true,
1017                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1018                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1019                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1020         };
1021
1022         if (code_to_probe >= ARRAY_SIZE(codes))
1023                 return false;
1024
1025         return codes[code_to_probe];
1026 }
1027
1028 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1029                                 unsigned int flen)
1030 {
1031         if (filter == NULL)
1032                 return false;
1033         if (flen == 0 || flen > BPF_MAXINSNS)
1034                 return false;
1035
1036         return true;
1037 }
1038
1039 /**
1040  *      bpf_check_classic - verify socket filter code
1041  *      @filter: filter to verify
1042  *      @flen: length of filter
1043  *
1044  * Check the user's filter code. If we let some ugly
1045  * filter code slip through kaboom! The filter must contain
1046  * no references or jumps that are out of range, no illegal
1047  * instructions, and must end with a RET instruction.
1048  *
1049  * All jumps are forward as they are not signed.
1050  *
1051  * Returns 0 if the rule set is legal or -EINVAL if not.
1052  */
1053 static int bpf_check_classic(const struct sock_filter *filter,
1054                              unsigned int flen)
1055 {
1056         bool anc_found;
1057         int pc;
1058
1059         /* Check the filter code now */
1060         for (pc = 0; pc < flen; pc++) {
1061                 const struct sock_filter *ftest = &filter[pc];
1062
1063                 /* May we actually operate on this code? */
1064                 if (!chk_code_allowed(ftest->code))
1065                         return -EINVAL;
1066
1067                 /* Some instructions need special checks */
1068                 switch (ftest->code) {
1069                 case BPF_ALU | BPF_DIV | BPF_K:
1070                 case BPF_ALU | BPF_MOD | BPF_K:
1071                         /* Check for division by zero */
1072                         if (ftest->k == 0)
1073                                 return -EINVAL;
1074                         break;
1075                 case BPF_ALU | BPF_LSH | BPF_K:
1076                 case BPF_ALU | BPF_RSH | BPF_K:
1077                         if (ftest->k >= 32)
1078                                 return -EINVAL;
1079                         break;
1080                 case BPF_LD | BPF_MEM:
1081                 case BPF_LDX | BPF_MEM:
1082                 case BPF_ST:
1083                 case BPF_STX:
1084                         /* Check for invalid memory addresses */
1085                         if (ftest->k >= BPF_MEMWORDS)
1086                                 return -EINVAL;
1087                         break;
1088                 case BPF_JMP | BPF_JA:
1089                         /* Note, the large ftest->k might cause loops.
1090                          * Compare this with conditional jumps below,
1091                          * where offsets are limited. --ANK (981016)
1092                          */
1093                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1094                                 return -EINVAL;
1095                         break;
1096                 case BPF_JMP | BPF_JEQ | BPF_K:
1097                 case BPF_JMP | BPF_JEQ | BPF_X:
1098                 case BPF_JMP | BPF_JGE | BPF_K:
1099                 case BPF_JMP | BPF_JGE | BPF_X:
1100                 case BPF_JMP | BPF_JGT | BPF_K:
1101                 case BPF_JMP | BPF_JGT | BPF_X:
1102                 case BPF_JMP | BPF_JSET | BPF_K:
1103                 case BPF_JMP | BPF_JSET | BPF_X:
1104                         /* Both conditionals must be safe */
1105                         if (pc + ftest->jt + 1 >= flen ||
1106                             pc + ftest->jf + 1 >= flen)
1107                                 return -EINVAL;
1108                         break;
1109                 case BPF_LD | BPF_W | BPF_ABS:
1110                 case BPF_LD | BPF_H | BPF_ABS:
1111                 case BPF_LD | BPF_B | BPF_ABS:
1112                         anc_found = false;
1113                         if (bpf_anc_helper(ftest) & BPF_ANC)
1114                                 anc_found = true;
1115                         /* Ancillary operation unknown or unsupported */
1116                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1117                                 return -EINVAL;
1118                 }
1119         }
1120
1121         /* Last instruction must be a RET code */
1122         switch (filter[flen - 1].code) {
1123         case BPF_RET | BPF_K:
1124         case BPF_RET | BPF_A:
1125                 return check_load_and_stores(filter, flen);
1126         }
1127
1128         return -EINVAL;
1129 }
1130
1131 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1132                                       const struct sock_fprog *fprog)
1133 {
1134         unsigned int fsize = bpf_classic_proglen(fprog);
1135         struct sock_fprog_kern *fkprog;
1136
1137         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1138         if (!fp->orig_prog)
1139                 return -ENOMEM;
1140
1141         fkprog = fp->orig_prog;
1142         fkprog->len = fprog->len;
1143
1144         fkprog->filter = kmemdup(fp->insns, fsize,
1145                                  GFP_KERNEL | __GFP_NOWARN);
1146         if (!fkprog->filter) {
1147                 kfree(fp->orig_prog);
1148                 return -ENOMEM;
1149         }
1150
1151         return 0;
1152 }
1153
1154 static void bpf_release_orig_filter(struct bpf_prog *fp)
1155 {
1156         struct sock_fprog_kern *fprog = fp->orig_prog;
1157
1158         if (fprog) {
1159                 kfree(fprog->filter);
1160                 kfree(fprog);
1161         }
1162 }
1163
1164 static void __bpf_prog_release(struct bpf_prog *prog)
1165 {
1166         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1167                 bpf_prog_put(prog);
1168         } else {
1169                 bpf_release_orig_filter(prog);
1170                 bpf_prog_free(prog);
1171         }
1172 }
1173
1174 static void __sk_filter_release(struct sk_filter *fp)
1175 {
1176         __bpf_prog_release(fp->prog);
1177         kfree(fp);
1178 }
1179
1180 /**
1181  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1182  *      @rcu: rcu_head that contains the sk_filter to free
1183  */
1184 static void sk_filter_release_rcu(struct rcu_head *rcu)
1185 {
1186         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1187
1188         __sk_filter_release(fp);
1189 }
1190
1191 /**
1192  *      sk_filter_release - release a socket filter
1193  *      @fp: filter to remove
1194  *
1195  *      Remove a filter from a socket and release its resources.
1196  */
1197 static void sk_filter_release(struct sk_filter *fp)
1198 {
1199         if (refcount_dec_and_test(&fp->refcnt))
1200                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1201 }
1202
1203 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1204 {
1205         u32 filter_size = bpf_prog_size(fp->prog->len);
1206
1207         atomic_sub(filter_size, &sk->sk_omem_alloc);
1208         sk_filter_release(fp);
1209 }
1210
1211 /* try to charge the socket memory if there is space available
1212  * return true on success
1213  */
1214 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1215 {
1216         u32 filter_size = bpf_prog_size(fp->prog->len);
1217         int optmem_max = READ_ONCE(sysctl_optmem_max);
1218
1219         /* same check as in sock_kmalloc() */
1220         if (filter_size <= optmem_max &&
1221             atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) {
1222                 atomic_add(filter_size, &sk->sk_omem_alloc);
1223                 return true;
1224         }
1225         return false;
1226 }
1227
1228 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1229 {
1230         if (!refcount_inc_not_zero(&fp->refcnt))
1231                 return false;
1232
1233         if (!__sk_filter_charge(sk, fp)) {
1234                 sk_filter_release(fp);
1235                 return false;
1236         }
1237         return true;
1238 }
1239
1240 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1241 {
1242         struct sock_filter *old_prog;
1243         struct bpf_prog *old_fp;
1244         int err, new_len, old_len = fp->len;
1245         bool seen_ld_abs = false;
1246
1247         /* We are free to overwrite insns et al right here as it won't be used at
1248          * this point in time anymore internally after the migration to the eBPF
1249          * instruction representation.
1250          */
1251         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1252                      sizeof(struct bpf_insn));
1253
1254         /* Conversion cannot happen on overlapping memory areas,
1255          * so we need to keep the user BPF around until the 2nd
1256          * pass. At this time, the user BPF is stored in fp->insns.
1257          */
1258         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1259                            GFP_KERNEL | __GFP_NOWARN);
1260         if (!old_prog) {
1261                 err = -ENOMEM;
1262                 goto out_err;
1263         }
1264
1265         /* 1st pass: calculate the new program length. */
1266         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1267                                  &seen_ld_abs);
1268         if (err)
1269                 goto out_err_free;
1270
1271         /* Expand fp for appending the new filter representation. */
1272         old_fp = fp;
1273         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1274         if (!fp) {
1275                 /* The old_fp is still around in case we couldn't
1276                  * allocate new memory, so uncharge on that one.
1277                  */
1278                 fp = old_fp;
1279                 err = -ENOMEM;
1280                 goto out_err_free;
1281         }
1282
1283         fp->len = new_len;
1284
1285         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1286         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1287                                  &seen_ld_abs);
1288         if (err)
1289                 /* 2nd bpf_convert_filter() can fail only if it fails
1290                  * to allocate memory, remapping must succeed. Note,
1291                  * that at this time old_fp has already been released
1292                  * by krealloc().
1293                  */
1294                 goto out_err_free;
1295
1296         fp = bpf_prog_select_runtime(fp, &err);
1297         if (err)
1298                 goto out_err_free;
1299
1300         kfree(old_prog);
1301         return fp;
1302
1303 out_err_free:
1304         kfree(old_prog);
1305 out_err:
1306         __bpf_prog_release(fp);
1307         return ERR_PTR(err);
1308 }
1309
1310 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1311                                            bpf_aux_classic_check_t trans)
1312 {
1313         int err;
1314
1315         fp->bpf_func = NULL;
1316         fp->jited = 0;
1317
1318         err = bpf_check_classic(fp->insns, fp->len);
1319         if (err) {
1320                 __bpf_prog_release(fp);
1321                 return ERR_PTR(err);
1322         }
1323
1324         /* There might be additional checks and transformations
1325          * needed on classic filters, f.e. in case of seccomp.
1326          */
1327         if (trans) {
1328                 err = trans(fp->insns, fp->len);
1329                 if (err) {
1330                         __bpf_prog_release(fp);
1331                         return ERR_PTR(err);
1332                 }
1333         }
1334
1335         /* Probe if we can JIT compile the filter and if so, do
1336          * the compilation of the filter.
1337          */
1338         bpf_jit_compile(fp);
1339
1340         /* JIT compiler couldn't process this filter, so do the eBPF translation
1341          * for the optimized interpreter.
1342          */
1343         if (!fp->jited)
1344                 fp = bpf_migrate_filter(fp);
1345
1346         return fp;
1347 }
1348
1349 /**
1350  *      bpf_prog_create - create an unattached filter
1351  *      @pfp: the unattached filter that is created
1352  *      @fprog: the filter program
1353  *
1354  * Create a filter independent of any socket. We first run some
1355  * sanity checks on it to make sure it does not explode on us later.
1356  * If an error occurs or there is insufficient memory for the filter
1357  * a negative errno code is returned. On success the return is zero.
1358  */
1359 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1360 {
1361         unsigned int fsize = bpf_classic_proglen(fprog);
1362         struct bpf_prog *fp;
1363
1364         /* Make sure new filter is there and in the right amounts. */
1365         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1366                 return -EINVAL;
1367
1368         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1369         if (!fp)
1370                 return -ENOMEM;
1371
1372         memcpy(fp->insns, fprog->filter, fsize);
1373
1374         fp->len = fprog->len;
1375         /* Since unattached filters are not copied back to user
1376          * space through sk_get_filter(), we do not need to hold
1377          * a copy here, and can spare us the work.
1378          */
1379         fp->orig_prog = NULL;
1380
1381         /* bpf_prepare_filter() already takes care of freeing
1382          * memory in case something goes wrong.
1383          */
1384         fp = bpf_prepare_filter(fp, NULL);
1385         if (IS_ERR(fp))
1386                 return PTR_ERR(fp);
1387
1388         *pfp = fp;
1389         return 0;
1390 }
1391 EXPORT_SYMBOL_GPL(bpf_prog_create);
1392
1393 /**
1394  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1395  *      @pfp: the unattached filter that is created
1396  *      @fprog: the filter program
1397  *      @trans: post-classic verifier transformation handler
1398  *      @save_orig: save classic BPF program
1399  *
1400  * This function effectively does the same as bpf_prog_create(), only
1401  * that it builds up its insns buffer from user space provided buffer.
1402  * It also allows for passing a bpf_aux_classic_check_t handler.
1403  */
1404 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1405                               bpf_aux_classic_check_t trans, bool save_orig)
1406 {
1407         unsigned int fsize = bpf_classic_proglen(fprog);
1408         struct bpf_prog *fp;
1409         int err;
1410
1411         /* Make sure new filter is there and in the right amounts. */
1412         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1413                 return -EINVAL;
1414
1415         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1416         if (!fp)
1417                 return -ENOMEM;
1418
1419         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1420                 __bpf_prog_free(fp);
1421                 return -EFAULT;
1422         }
1423
1424         fp->len = fprog->len;
1425         fp->orig_prog = NULL;
1426
1427         if (save_orig) {
1428                 err = bpf_prog_store_orig_filter(fp, fprog);
1429                 if (err) {
1430                         __bpf_prog_free(fp);
1431                         return -ENOMEM;
1432                 }
1433         }
1434
1435         /* bpf_prepare_filter() already takes care of freeing
1436          * memory in case something goes wrong.
1437          */
1438         fp = bpf_prepare_filter(fp, trans);
1439         if (IS_ERR(fp))
1440                 return PTR_ERR(fp);
1441
1442         *pfp = fp;
1443         return 0;
1444 }
1445 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1446
1447 void bpf_prog_destroy(struct bpf_prog *fp)
1448 {
1449         __bpf_prog_release(fp);
1450 }
1451 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1452
1453 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1454 {
1455         struct sk_filter *fp, *old_fp;
1456
1457         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1458         if (!fp)
1459                 return -ENOMEM;
1460
1461         fp->prog = prog;
1462
1463         if (!__sk_filter_charge(sk, fp)) {
1464                 kfree(fp);
1465                 return -ENOMEM;
1466         }
1467         refcount_set(&fp->refcnt, 1);
1468
1469         old_fp = rcu_dereference_protected(sk->sk_filter,
1470                                            lockdep_sock_is_held(sk));
1471         rcu_assign_pointer(sk->sk_filter, fp);
1472
1473         if (old_fp)
1474                 sk_filter_uncharge(sk, old_fp);
1475
1476         return 0;
1477 }
1478
1479 static
1480 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1481 {
1482         unsigned int fsize = bpf_classic_proglen(fprog);
1483         struct bpf_prog *prog;
1484         int err;
1485
1486         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1487                 return ERR_PTR(-EPERM);
1488
1489         /* Make sure new filter is there and in the right amounts. */
1490         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1491                 return ERR_PTR(-EINVAL);
1492
1493         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1494         if (!prog)
1495                 return ERR_PTR(-ENOMEM);
1496
1497         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1498                 __bpf_prog_free(prog);
1499                 return ERR_PTR(-EFAULT);
1500         }
1501
1502         prog->len = fprog->len;
1503
1504         err = bpf_prog_store_orig_filter(prog, fprog);
1505         if (err) {
1506                 __bpf_prog_free(prog);
1507                 return ERR_PTR(-ENOMEM);
1508         }
1509
1510         /* bpf_prepare_filter() already takes care of freeing
1511          * memory in case something goes wrong.
1512          */
1513         return bpf_prepare_filter(prog, NULL);
1514 }
1515
1516 /**
1517  *      sk_attach_filter - attach a socket filter
1518  *      @fprog: the filter program
1519  *      @sk: the socket to use
1520  *
1521  * Attach the user's filter code. We first run some sanity checks on
1522  * it to make sure it does not explode on us later. If an error
1523  * occurs or there is insufficient memory for the filter a negative
1524  * errno code is returned. On success the return is zero.
1525  */
1526 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1527 {
1528         struct bpf_prog *prog = __get_filter(fprog, sk);
1529         int err;
1530
1531         if (IS_ERR(prog))
1532                 return PTR_ERR(prog);
1533
1534         err = __sk_attach_prog(prog, sk);
1535         if (err < 0) {
1536                 __bpf_prog_release(prog);
1537                 return err;
1538         }
1539
1540         return 0;
1541 }
1542 EXPORT_SYMBOL_GPL(sk_attach_filter);
1543
1544 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1545 {
1546         struct bpf_prog *prog = __get_filter(fprog, sk);
1547         int err;
1548
1549         if (IS_ERR(prog))
1550                 return PTR_ERR(prog);
1551
1552         if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max))
1553                 err = -ENOMEM;
1554         else
1555                 err = reuseport_attach_prog(sk, prog);
1556
1557         if (err)
1558                 __bpf_prog_release(prog);
1559
1560         return err;
1561 }
1562
1563 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1564 {
1565         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1566                 return ERR_PTR(-EPERM);
1567
1568         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 }
1570
1571 int sk_attach_bpf(u32 ufd, struct sock *sk)
1572 {
1573         struct bpf_prog *prog = __get_bpf(ufd, sk);
1574         int err;
1575
1576         if (IS_ERR(prog))
1577                 return PTR_ERR(prog);
1578
1579         err = __sk_attach_prog(prog, sk);
1580         if (err < 0) {
1581                 bpf_prog_put(prog);
1582                 return err;
1583         }
1584
1585         return 0;
1586 }
1587
1588 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1589 {
1590         struct bpf_prog *prog;
1591         int err;
1592
1593         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1594                 return -EPERM;
1595
1596         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1597         if (PTR_ERR(prog) == -EINVAL)
1598                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1599         if (IS_ERR(prog))
1600                 return PTR_ERR(prog);
1601
1602         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1603                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1604                  * bpf prog (e.g. sockmap).  It depends on the
1605                  * limitation imposed by bpf_prog_load().
1606                  * Hence, sysctl_optmem_max is not checked.
1607                  */
1608                 if ((sk->sk_type != SOCK_STREAM &&
1609                      sk->sk_type != SOCK_DGRAM) ||
1610                     (sk->sk_protocol != IPPROTO_UDP &&
1611                      sk->sk_protocol != IPPROTO_TCP) ||
1612                     (sk->sk_family != AF_INET &&
1613                      sk->sk_family != AF_INET6)) {
1614                         err = -ENOTSUPP;
1615                         goto err_prog_put;
1616                 }
1617         } else {
1618                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1619                 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) {
1620                         err = -ENOMEM;
1621                         goto err_prog_put;
1622                 }
1623         }
1624
1625         err = reuseport_attach_prog(sk, prog);
1626 err_prog_put:
1627         if (err)
1628                 bpf_prog_put(prog);
1629
1630         return err;
1631 }
1632
1633 void sk_reuseport_prog_free(struct bpf_prog *prog)
1634 {
1635         if (!prog)
1636                 return;
1637
1638         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1639                 bpf_prog_put(prog);
1640         else
1641                 bpf_prog_destroy(prog);
1642 }
1643
1644 struct bpf_scratchpad {
1645         union {
1646                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1647                 u8     buff[MAX_BPF_STACK];
1648         };
1649 };
1650
1651 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1652
1653 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1654                                           unsigned int write_len)
1655 {
1656         return skb_ensure_writable(skb, write_len);
1657 }
1658
1659 static inline int bpf_try_make_writable(struct sk_buff *skb,
1660                                         unsigned int write_len)
1661 {
1662         int err = __bpf_try_make_writable(skb, write_len);
1663
1664         bpf_compute_data_pointers(skb);
1665         return err;
1666 }
1667
1668 static int bpf_try_make_head_writable(struct sk_buff *skb)
1669 {
1670         return bpf_try_make_writable(skb, skb_headlen(skb));
1671 }
1672
1673 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1674 {
1675         if (skb_at_tc_ingress(skb))
1676                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1677 }
1678
1679 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1680 {
1681         if (skb_at_tc_ingress(skb))
1682                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1683 }
1684
1685 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1686            const void *, from, u32, len, u64, flags)
1687 {
1688         void *ptr;
1689
1690         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1691                 return -EINVAL;
1692         if (unlikely(offset > INT_MAX))
1693                 return -EFAULT;
1694         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1695                 return -EFAULT;
1696
1697         ptr = skb->data + offset;
1698         if (flags & BPF_F_RECOMPUTE_CSUM)
1699                 __skb_postpull_rcsum(skb, ptr, len, offset);
1700
1701         memcpy(ptr, from, len);
1702
1703         if (flags & BPF_F_RECOMPUTE_CSUM)
1704                 __skb_postpush_rcsum(skb, ptr, len, offset);
1705         if (flags & BPF_F_INVALIDATE_HASH)
1706                 skb_clear_hash(skb);
1707
1708         return 0;
1709 }
1710
1711 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1712         .func           = bpf_skb_store_bytes,
1713         .gpl_only       = false,
1714         .ret_type       = RET_INTEGER,
1715         .arg1_type      = ARG_PTR_TO_CTX,
1716         .arg2_type      = ARG_ANYTHING,
1717         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
1718         .arg4_type      = ARG_CONST_SIZE,
1719         .arg5_type      = ARG_ANYTHING,
1720 };
1721
1722 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1723            void *, to, u32, len)
1724 {
1725         void *ptr;
1726
1727         if (unlikely(offset > INT_MAX))
1728                 goto err_clear;
1729
1730         ptr = skb_header_pointer(skb, offset, len, to);
1731         if (unlikely(!ptr))
1732                 goto err_clear;
1733         if (ptr != to)
1734                 memcpy(to, ptr, len);
1735
1736         return 0;
1737 err_clear:
1738         memset(to, 0, len);
1739         return -EFAULT;
1740 }
1741
1742 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1743         .func           = bpf_skb_load_bytes,
1744         .gpl_only       = false,
1745         .ret_type       = RET_INTEGER,
1746         .arg1_type      = ARG_PTR_TO_CTX,
1747         .arg2_type      = ARG_ANYTHING,
1748         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1749         .arg4_type      = ARG_CONST_SIZE,
1750 };
1751
1752 BPF_CALL_4(bpf_flow_dissector_load_bytes,
1753            const struct bpf_flow_dissector *, ctx, u32, offset,
1754            void *, to, u32, len)
1755 {
1756         void *ptr;
1757
1758         if (unlikely(offset > 0xffff))
1759                 goto err_clear;
1760
1761         if (unlikely(!ctx->skb))
1762                 goto err_clear;
1763
1764         ptr = skb_header_pointer(ctx->skb, offset, len, to);
1765         if (unlikely(!ptr))
1766                 goto err_clear;
1767         if (ptr != to)
1768                 memcpy(to, ptr, len);
1769
1770         return 0;
1771 err_clear:
1772         memset(to, 0, len);
1773         return -EFAULT;
1774 }
1775
1776 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = {
1777         .func           = bpf_flow_dissector_load_bytes,
1778         .gpl_only       = false,
1779         .ret_type       = RET_INTEGER,
1780         .arg1_type      = ARG_PTR_TO_CTX,
1781         .arg2_type      = ARG_ANYTHING,
1782         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1783         .arg4_type      = ARG_CONST_SIZE,
1784 };
1785
1786 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1787            u32, offset, void *, to, u32, len, u32, start_header)
1788 {
1789         u8 *end = skb_tail_pointer(skb);
1790         u8 *start, *ptr;
1791
1792         if (unlikely(offset > 0xffff))
1793                 goto err_clear;
1794
1795         switch (start_header) {
1796         case BPF_HDR_START_MAC:
1797                 if (unlikely(!skb_mac_header_was_set(skb)))
1798                         goto err_clear;
1799                 start = skb_mac_header(skb);
1800                 break;
1801         case BPF_HDR_START_NET:
1802                 start = skb_network_header(skb);
1803                 break;
1804         default:
1805                 goto err_clear;
1806         }
1807
1808         ptr = start + offset;
1809
1810         if (likely(ptr + len <= end)) {
1811                 memcpy(to, ptr, len);
1812                 return 0;
1813         }
1814
1815 err_clear:
1816         memset(to, 0, len);
1817         return -EFAULT;
1818 }
1819
1820 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1821         .func           = bpf_skb_load_bytes_relative,
1822         .gpl_only       = false,
1823         .ret_type       = RET_INTEGER,
1824         .arg1_type      = ARG_PTR_TO_CTX,
1825         .arg2_type      = ARG_ANYTHING,
1826         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1827         .arg4_type      = ARG_CONST_SIZE,
1828         .arg5_type      = ARG_ANYTHING,
1829 };
1830
1831 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1832 {
1833         /* Idea is the following: should the needed direct read/write
1834          * test fail during runtime, we can pull in more data and redo
1835          * again, since implicitly, we invalidate previous checks here.
1836          *
1837          * Or, since we know how much we need to make read/writeable,
1838          * this can be done once at the program beginning for direct
1839          * access case. By this we overcome limitations of only current
1840          * headroom being accessible.
1841          */
1842         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1843 }
1844
1845 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1846         .func           = bpf_skb_pull_data,
1847         .gpl_only       = false,
1848         .ret_type       = RET_INTEGER,
1849         .arg1_type      = ARG_PTR_TO_CTX,
1850         .arg2_type      = ARG_ANYTHING,
1851 };
1852
1853 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk)
1854 {
1855         return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL;
1856 }
1857
1858 static const struct bpf_func_proto bpf_sk_fullsock_proto = {
1859         .func           = bpf_sk_fullsock,
1860         .gpl_only       = false,
1861         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
1862         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
1863 };
1864
1865 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1866                                            unsigned int write_len)
1867 {
1868         return __bpf_try_make_writable(skb, write_len);
1869 }
1870
1871 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1872 {
1873         /* Idea is the following: should the needed direct read/write
1874          * test fail during runtime, we can pull in more data and redo
1875          * again, since implicitly, we invalidate previous checks here.
1876          *
1877          * Or, since we know how much we need to make read/writeable,
1878          * this can be done once at the program beginning for direct
1879          * access case. By this we overcome limitations of only current
1880          * headroom being accessible.
1881          */
1882         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1883 }
1884
1885 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1886         .func           = sk_skb_pull_data,
1887         .gpl_only       = false,
1888         .ret_type       = RET_INTEGER,
1889         .arg1_type      = ARG_PTR_TO_CTX,
1890         .arg2_type      = ARG_ANYTHING,
1891 };
1892
1893 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1894            u64, from, u64, to, u64, flags)
1895 {
1896         __sum16 *ptr;
1897
1898         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1899                 return -EINVAL;
1900         if (unlikely(offset > 0xffff || offset & 1))
1901                 return -EFAULT;
1902         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1903                 return -EFAULT;
1904
1905         ptr = (__sum16 *)(skb->data + offset);
1906         switch (flags & BPF_F_HDR_FIELD_MASK) {
1907         case 0:
1908                 if (unlikely(from != 0))
1909                         return -EINVAL;
1910
1911                 csum_replace_by_diff(ptr, to);
1912                 break;
1913         case 2:
1914                 csum_replace2(ptr, from, to);
1915                 break;
1916         case 4:
1917                 csum_replace4(ptr, from, to);
1918                 break;
1919         default:
1920                 return -EINVAL;
1921         }
1922
1923         return 0;
1924 }
1925
1926 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1927         .func           = bpf_l3_csum_replace,
1928         .gpl_only       = false,
1929         .ret_type       = RET_INTEGER,
1930         .arg1_type      = ARG_PTR_TO_CTX,
1931         .arg2_type      = ARG_ANYTHING,
1932         .arg3_type      = ARG_ANYTHING,
1933         .arg4_type      = ARG_ANYTHING,
1934         .arg5_type      = ARG_ANYTHING,
1935 };
1936
1937 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1938            u64, from, u64, to, u64, flags)
1939 {
1940         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1941         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1942         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1943         __sum16 *ptr;
1944
1945         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1946                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1947                 return -EINVAL;
1948         if (unlikely(offset > 0xffff || offset & 1))
1949                 return -EFAULT;
1950         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1951                 return -EFAULT;
1952
1953         ptr = (__sum16 *)(skb->data + offset);
1954         if (is_mmzero && !do_mforce && !*ptr)
1955                 return 0;
1956
1957         switch (flags & BPF_F_HDR_FIELD_MASK) {
1958         case 0:
1959                 if (unlikely(from != 0))
1960                         return -EINVAL;
1961
1962                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1963                 break;
1964         case 2:
1965                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1966                 break;
1967         case 4:
1968                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1969                 break;
1970         default:
1971                 return -EINVAL;
1972         }
1973
1974         if (is_mmzero && !*ptr)
1975                 *ptr = CSUM_MANGLED_0;
1976         return 0;
1977 }
1978
1979 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1980         .func           = bpf_l4_csum_replace,
1981         .gpl_only       = false,
1982         .ret_type       = RET_INTEGER,
1983         .arg1_type      = ARG_PTR_TO_CTX,
1984         .arg2_type      = ARG_ANYTHING,
1985         .arg3_type      = ARG_ANYTHING,
1986         .arg4_type      = ARG_ANYTHING,
1987         .arg5_type      = ARG_ANYTHING,
1988 };
1989
1990 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1991            __be32 *, to, u32, to_size, __wsum, seed)
1992 {
1993         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1994         u32 diff_size = from_size + to_size;
1995         int i, j = 0;
1996
1997         /* This is quite flexible, some examples:
1998          *
1999          * from_size == 0, to_size > 0,  seed := csum --> pushing data
2000          * from_size > 0,  to_size == 0, seed := csum --> pulling data
2001          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
2002          *
2003          * Even for diffing, from_size and to_size don't need to be equal.
2004          */
2005         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
2006                      diff_size > sizeof(sp->diff)))
2007                 return -EINVAL;
2008
2009         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
2010                 sp->diff[j] = ~from[i];
2011         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
2012                 sp->diff[j] = to[i];
2013
2014         return csum_partial(sp->diff, diff_size, seed);
2015 }
2016
2017 static const struct bpf_func_proto bpf_csum_diff_proto = {
2018         .func           = bpf_csum_diff,
2019         .gpl_only       = false,
2020         .pkt_access     = true,
2021         .ret_type       = RET_INTEGER,
2022         .arg1_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2023         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
2024         .arg3_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2025         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
2026         .arg5_type      = ARG_ANYTHING,
2027 };
2028
2029 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
2030 {
2031         /* The interface is to be used in combination with bpf_csum_diff()
2032          * for direct packet writes. csum rotation for alignment as well
2033          * as emulating csum_sub() can be done from the eBPF program.
2034          */
2035         if (skb->ip_summed == CHECKSUM_COMPLETE)
2036                 return (skb->csum = csum_add(skb->csum, csum));
2037
2038         return -ENOTSUPP;
2039 }
2040
2041 static const struct bpf_func_proto bpf_csum_update_proto = {
2042         .func           = bpf_csum_update,
2043         .gpl_only       = false,
2044         .ret_type       = RET_INTEGER,
2045         .arg1_type      = ARG_PTR_TO_CTX,
2046         .arg2_type      = ARG_ANYTHING,
2047 };
2048
2049 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level)
2050 {
2051         /* The interface is to be used in combination with bpf_skb_adjust_room()
2052          * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET
2053          * is passed as flags, for example.
2054          */
2055         switch (level) {
2056         case BPF_CSUM_LEVEL_INC:
2057                 __skb_incr_checksum_unnecessary(skb);
2058                 break;
2059         case BPF_CSUM_LEVEL_DEC:
2060                 __skb_decr_checksum_unnecessary(skb);
2061                 break;
2062         case BPF_CSUM_LEVEL_RESET:
2063                 __skb_reset_checksum_unnecessary(skb);
2064                 break;
2065         case BPF_CSUM_LEVEL_QUERY:
2066                 return skb->ip_summed == CHECKSUM_UNNECESSARY ?
2067                        skb->csum_level : -EACCES;
2068         default:
2069                 return -EINVAL;
2070         }
2071
2072         return 0;
2073 }
2074
2075 static const struct bpf_func_proto bpf_csum_level_proto = {
2076         .func           = bpf_csum_level,
2077         .gpl_only       = false,
2078         .ret_type       = RET_INTEGER,
2079         .arg1_type      = ARG_PTR_TO_CTX,
2080         .arg2_type      = ARG_ANYTHING,
2081 };
2082
2083 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
2084 {
2085         return dev_forward_skb_nomtu(dev, skb);
2086 }
2087
2088 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
2089                                       struct sk_buff *skb)
2090 {
2091         int ret = ____dev_forward_skb(dev, skb, false);
2092
2093         if (likely(!ret)) {
2094                 skb->dev = dev;
2095                 ret = netif_rx(skb);
2096         }
2097
2098         return ret;
2099 }
2100
2101 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2102 {
2103         int ret;
2104
2105         if (dev_xmit_recursion()) {
2106                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2107                 kfree_skb(skb);
2108                 return -ENETDOWN;
2109         }
2110
2111         skb->dev = dev;
2112         skb_clear_tstamp(skb);
2113
2114         dev_xmit_recursion_inc();
2115         ret = dev_queue_xmit(skb);
2116         dev_xmit_recursion_dec();
2117
2118         return ret;
2119 }
2120
2121 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2122                                  u32 flags)
2123 {
2124         unsigned int mlen = skb_network_offset(skb);
2125
2126         if (mlen) {
2127                 __skb_pull(skb, mlen);
2128
2129                 /* At ingress, the mac header has already been pulled once.
2130                  * At egress, skb_pospull_rcsum has to be done in case that
2131                  * the skb is originated from ingress (i.e. a forwarded skb)
2132                  * to ensure that rcsum starts at net header.
2133                  */
2134                 if (!skb_at_tc_ingress(skb))
2135                         skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2136         }
2137         skb_pop_mac_header(skb);
2138         skb_reset_mac_len(skb);
2139         return flags & BPF_F_INGRESS ?
2140                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2141 }
2142
2143 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2144                                  u32 flags)
2145 {
2146         /* Verify that a link layer header is carried */
2147         if (unlikely(skb->mac_header >= skb->network_header)) {
2148                 kfree_skb(skb);
2149                 return -ERANGE;
2150         }
2151
2152         bpf_push_mac_rcsum(skb);
2153         return flags & BPF_F_INGRESS ?
2154                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2155 }
2156
2157 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2158                           u32 flags)
2159 {
2160         if (dev_is_mac_header_xmit(dev))
2161                 return __bpf_redirect_common(skb, dev, flags);
2162         else
2163                 return __bpf_redirect_no_mac(skb, dev, flags);
2164 }
2165
2166 #if IS_ENABLED(CONFIG_IPV6)
2167 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb,
2168                             struct net_device *dev, struct bpf_nh_params *nh)
2169 {
2170         u32 hh_len = LL_RESERVED_SPACE(dev);
2171         const struct in6_addr *nexthop;
2172         struct dst_entry *dst = NULL;
2173         struct neighbour *neigh;
2174
2175         if (dev_xmit_recursion()) {
2176                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2177                 goto out_drop;
2178         }
2179
2180         skb->dev = dev;
2181         skb_clear_tstamp(skb);
2182
2183         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2184                 skb = skb_expand_head(skb, hh_len);
2185                 if (!skb)
2186                         return -ENOMEM;
2187         }
2188
2189         rcu_read_lock_bh();
2190         if (!nh) {
2191                 dst = skb_dst(skb);
2192                 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst),
2193                                       &ipv6_hdr(skb)->daddr);
2194         } else {
2195                 nexthop = &nh->ipv6_nh;
2196         }
2197         neigh = ip_neigh_gw6(dev, nexthop);
2198         if (likely(!IS_ERR(neigh))) {
2199                 int ret;
2200
2201                 sock_confirm_neigh(skb, neigh);
2202                 dev_xmit_recursion_inc();
2203                 ret = neigh_output(neigh, skb, false);
2204                 dev_xmit_recursion_dec();
2205                 rcu_read_unlock_bh();
2206                 return ret;
2207         }
2208         rcu_read_unlock_bh();
2209         if (dst)
2210                 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES);
2211 out_drop:
2212         kfree_skb(skb);
2213         return -ENETDOWN;
2214 }
2215
2216 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2217                                    struct bpf_nh_params *nh)
2218 {
2219         const struct ipv6hdr *ip6h = ipv6_hdr(skb);
2220         struct net *net = dev_net(dev);
2221         int err, ret = NET_XMIT_DROP;
2222
2223         if (!nh) {
2224                 struct dst_entry *dst;
2225                 struct flowi6 fl6 = {
2226                         .flowi6_flags = FLOWI_FLAG_ANYSRC,
2227                         .flowi6_mark  = skb->mark,
2228                         .flowlabel    = ip6_flowinfo(ip6h),
2229                         .flowi6_oif   = dev->ifindex,
2230                         .flowi6_proto = ip6h->nexthdr,
2231                         .daddr        = ip6h->daddr,
2232                         .saddr        = ip6h->saddr,
2233                 };
2234
2235                 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL);
2236                 if (IS_ERR(dst))
2237                         goto out_drop;
2238
2239                 skb_dst_set(skb, dst);
2240         } else if (nh->nh_family != AF_INET6) {
2241                 goto out_drop;
2242         }
2243
2244         err = bpf_out_neigh_v6(net, skb, dev, nh);
2245         if (unlikely(net_xmit_eval(err)))
2246                 dev->stats.tx_errors++;
2247         else
2248                 ret = NET_XMIT_SUCCESS;
2249         goto out_xmit;
2250 out_drop:
2251         dev->stats.tx_errors++;
2252         kfree_skb(skb);
2253 out_xmit:
2254         return ret;
2255 }
2256 #else
2257 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev,
2258                                    struct bpf_nh_params *nh)
2259 {
2260         kfree_skb(skb);
2261         return NET_XMIT_DROP;
2262 }
2263 #endif /* CONFIG_IPV6 */
2264
2265 #if IS_ENABLED(CONFIG_INET)
2266 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb,
2267                             struct net_device *dev, struct bpf_nh_params *nh)
2268 {
2269         u32 hh_len = LL_RESERVED_SPACE(dev);
2270         struct neighbour *neigh;
2271         bool is_v6gw = false;
2272
2273         if (dev_xmit_recursion()) {
2274                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2275                 goto out_drop;
2276         }
2277
2278         skb->dev = dev;
2279         skb_clear_tstamp(skb);
2280
2281         if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) {
2282                 skb = skb_expand_head(skb, hh_len);
2283                 if (!skb)
2284                         return -ENOMEM;
2285         }
2286
2287         rcu_read_lock_bh();
2288         if (!nh) {
2289                 struct dst_entry *dst = skb_dst(skb);
2290                 struct rtable *rt = container_of(dst, struct rtable, dst);
2291
2292                 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw);
2293         } else if (nh->nh_family == AF_INET6) {
2294                 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh);
2295                 is_v6gw = true;
2296         } else if (nh->nh_family == AF_INET) {
2297                 neigh = ip_neigh_gw4(dev, nh->ipv4_nh);
2298         } else {
2299                 rcu_read_unlock_bh();
2300                 goto out_drop;
2301         }
2302
2303         if (likely(!IS_ERR(neigh))) {
2304                 int ret;
2305
2306                 sock_confirm_neigh(skb, neigh);
2307                 dev_xmit_recursion_inc();
2308                 ret = neigh_output(neigh, skb, is_v6gw);
2309                 dev_xmit_recursion_dec();
2310                 rcu_read_unlock_bh();
2311                 return ret;
2312         }
2313         rcu_read_unlock_bh();
2314 out_drop:
2315         kfree_skb(skb);
2316         return -ENETDOWN;
2317 }
2318
2319 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2320                                    struct bpf_nh_params *nh)
2321 {
2322         const struct iphdr *ip4h = ip_hdr(skb);
2323         struct net *net = dev_net(dev);
2324         int err, ret = NET_XMIT_DROP;
2325
2326         if (!nh) {
2327                 struct flowi4 fl4 = {
2328                         .flowi4_flags = FLOWI_FLAG_ANYSRC,
2329                         .flowi4_mark  = skb->mark,
2330                         .flowi4_tos   = RT_TOS(ip4h->tos),
2331                         .flowi4_oif   = dev->ifindex,
2332                         .flowi4_proto = ip4h->protocol,
2333                         .daddr        = ip4h->daddr,
2334                         .saddr        = ip4h->saddr,
2335                 };
2336                 struct rtable *rt;
2337
2338                 rt = ip_route_output_flow(net, &fl4, NULL);
2339                 if (IS_ERR(rt))
2340                         goto out_drop;
2341                 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) {
2342                         ip_rt_put(rt);
2343                         goto out_drop;
2344                 }
2345
2346                 skb_dst_set(skb, &rt->dst);
2347         }
2348
2349         err = bpf_out_neigh_v4(net, skb, dev, nh);
2350         if (unlikely(net_xmit_eval(err)))
2351                 dev->stats.tx_errors++;
2352         else
2353                 ret = NET_XMIT_SUCCESS;
2354         goto out_xmit;
2355 out_drop:
2356         dev->stats.tx_errors++;
2357         kfree_skb(skb);
2358 out_xmit:
2359         return ret;
2360 }
2361 #else
2362 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev,
2363                                    struct bpf_nh_params *nh)
2364 {
2365         kfree_skb(skb);
2366         return NET_XMIT_DROP;
2367 }
2368 #endif /* CONFIG_INET */
2369
2370 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev,
2371                                 struct bpf_nh_params *nh)
2372 {
2373         struct ethhdr *ethh = eth_hdr(skb);
2374
2375         if (unlikely(skb->mac_header >= skb->network_header))
2376                 goto out;
2377         bpf_push_mac_rcsum(skb);
2378         if (is_multicast_ether_addr(ethh->h_dest))
2379                 goto out;
2380
2381         skb_pull(skb, sizeof(*ethh));
2382         skb_unset_mac_header(skb);
2383         skb_reset_network_header(skb);
2384
2385         if (skb->protocol == htons(ETH_P_IP))
2386                 return __bpf_redirect_neigh_v4(skb, dev, nh);
2387         else if (skb->protocol == htons(ETH_P_IPV6))
2388                 return __bpf_redirect_neigh_v6(skb, dev, nh);
2389 out:
2390         kfree_skb(skb);
2391         return -ENOTSUPP;
2392 }
2393
2394 /* Internal, non-exposed redirect flags. */
2395 enum {
2396         BPF_F_NEIGH     = (1ULL << 1),
2397         BPF_F_PEER      = (1ULL << 2),
2398         BPF_F_NEXTHOP   = (1ULL << 3),
2399 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP)
2400 };
2401
2402 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2403 {
2404         struct net_device *dev;
2405         struct sk_buff *clone;
2406         int ret;
2407
2408         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2409                 return -EINVAL;
2410
2411         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2412         if (unlikely(!dev))
2413                 return -EINVAL;
2414
2415         clone = skb_clone(skb, GFP_ATOMIC);
2416         if (unlikely(!clone))
2417                 return -ENOMEM;
2418
2419         /* For direct write, we need to keep the invariant that the skbs
2420          * we're dealing with need to be uncloned. Should uncloning fail
2421          * here, we need to free the just generated clone to unclone once
2422          * again.
2423          */
2424         ret = bpf_try_make_head_writable(skb);
2425         if (unlikely(ret)) {
2426                 kfree_skb(clone);
2427                 return -ENOMEM;
2428         }
2429
2430         return __bpf_redirect(clone, dev, flags);
2431 }
2432
2433 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2434         .func           = bpf_clone_redirect,
2435         .gpl_only       = false,
2436         .ret_type       = RET_INTEGER,
2437         .arg1_type      = ARG_PTR_TO_CTX,
2438         .arg2_type      = ARG_ANYTHING,
2439         .arg3_type      = ARG_ANYTHING,
2440 };
2441
2442 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2443 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2444
2445 int skb_do_redirect(struct sk_buff *skb)
2446 {
2447         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2448         struct net *net = dev_net(skb->dev);
2449         struct net_device *dev;
2450         u32 flags = ri->flags;
2451
2452         dev = dev_get_by_index_rcu(net, ri->tgt_index);
2453         ri->tgt_index = 0;
2454         ri->flags = 0;
2455         if (unlikely(!dev))
2456                 goto out_drop;
2457         if (flags & BPF_F_PEER) {
2458                 const struct net_device_ops *ops = dev->netdev_ops;
2459
2460                 if (unlikely(!ops->ndo_get_peer_dev ||
2461                              !skb_at_tc_ingress(skb)))
2462                         goto out_drop;
2463                 dev = ops->ndo_get_peer_dev(dev);
2464                 if (unlikely(!dev ||
2465                              !(dev->flags & IFF_UP) ||
2466                              net_eq(net, dev_net(dev))))
2467                         goto out_drop;
2468                 skb->dev = dev;
2469                 return -EAGAIN;
2470         }
2471         return flags & BPF_F_NEIGH ?
2472                __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ?
2473                                     &ri->nh : NULL) :
2474                __bpf_redirect(skb, dev, flags);
2475 out_drop:
2476         kfree_skb(skb);
2477         return -EINVAL;
2478 }
2479
2480 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2481 {
2482         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2483
2484         if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
2485                 return TC_ACT_SHOT;
2486
2487         ri->flags = flags;
2488         ri->tgt_index = ifindex;
2489
2490         return TC_ACT_REDIRECT;
2491 }
2492
2493 static const struct bpf_func_proto bpf_redirect_proto = {
2494         .func           = bpf_redirect,
2495         .gpl_only       = false,
2496         .ret_type       = RET_INTEGER,
2497         .arg1_type      = ARG_ANYTHING,
2498         .arg2_type      = ARG_ANYTHING,
2499 };
2500
2501 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
2502 {
2503         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2504
2505         if (unlikely(flags))
2506                 return TC_ACT_SHOT;
2507
2508         ri->flags = BPF_F_PEER;
2509         ri->tgt_index = ifindex;
2510
2511         return TC_ACT_REDIRECT;
2512 }
2513
2514 static const struct bpf_func_proto bpf_redirect_peer_proto = {
2515         .func           = bpf_redirect_peer,
2516         .gpl_only       = false,
2517         .ret_type       = RET_INTEGER,
2518         .arg1_type      = ARG_ANYTHING,
2519         .arg2_type      = ARG_ANYTHING,
2520 };
2521
2522 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
2523            int, plen, u64, flags)
2524 {
2525         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2526
2527         if (unlikely((plen && plen < sizeof(*params)) || flags))
2528                 return TC_ACT_SHOT;
2529
2530         ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0);
2531         ri->tgt_index = ifindex;
2532
2533         BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params));
2534         if (plen)
2535                 memcpy(&ri->nh, params, sizeof(ri->nh));
2536
2537         return TC_ACT_REDIRECT;
2538 }
2539
2540 static const struct bpf_func_proto bpf_redirect_neigh_proto = {
2541         .func           = bpf_redirect_neigh,
2542         .gpl_only       = false,
2543         .ret_type       = RET_INTEGER,
2544         .arg1_type      = ARG_ANYTHING,
2545         .arg2_type      = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
2546         .arg3_type      = ARG_CONST_SIZE_OR_ZERO,
2547         .arg4_type      = ARG_ANYTHING,
2548 };
2549
2550 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2551 {
2552         msg->apply_bytes = bytes;
2553         return 0;
2554 }
2555
2556 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2557         .func           = bpf_msg_apply_bytes,
2558         .gpl_only       = false,
2559         .ret_type       = RET_INTEGER,
2560         .arg1_type      = ARG_PTR_TO_CTX,
2561         .arg2_type      = ARG_ANYTHING,
2562 };
2563
2564 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2565 {
2566         msg->cork_bytes = bytes;
2567         return 0;
2568 }
2569
2570 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2571         .func           = bpf_msg_cork_bytes,
2572         .gpl_only       = false,
2573         .ret_type       = RET_INTEGER,
2574         .arg1_type      = ARG_PTR_TO_CTX,
2575         .arg2_type      = ARG_ANYTHING,
2576 };
2577
2578 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2579            u32, end, u64, flags)
2580 {
2581         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2582         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2583         struct scatterlist *sge;
2584         u8 *raw, *to, *from;
2585         struct page *page;
2586
2587         if (unlikely(flags || end <= start))
2588                 return -EINVAL;
2589
2590         /* First find the starting scatterlist element */
2591         i = msg->sg.start;
2592         do {
2593                 offset += len;
2594                 len = sk_msg_elem(msg, i)->length;
2595                 if (start < offset + len)
2596                         break;
2597                 sk_msg_iter_var_next(i);
2598         } while (i != msg->sg.end);
2599
2600         if (unlikely(start >= offset + len))
2601                 return -EINVAL;
2602
2603         first_sge = i;
2604         /* The start may point into the sg element so we need to also
2605          * account for the headroom.
2606          */
2607         bytes_sg_total = start - offset + bytes;
2608         if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len)
2609                 goto out;
2610
2611         /* At this point we need to linearize multiple scatterlist
2612          * elements or a single shared page. Either way we need to
2613          * copy into a linear buffer exclusively owned by BPF. Then
2614          * place the buffer in the scatterlist and fixup the original
2615          * entries by removing the entries now in the linear buffer
2616          * and shifting the remaining entries. For now we do not try
2617          * to copy partial entries to avoid complexity of running out
2618          * of sg_entry slots. The downside is reading a single byte
2619          * will copy the entire sg entry.
2620          */
2621         do {
2622                 copy += sk_msg_elem(msg, i)->length;
2623                 sk_msg_iter_var_next(i);
2624                 if (bytes_sg_total <= copy)
2625                         break;
2626         } while (i != msg->sg.end);
2627         last_sge = i;
2628
2629         if (unlikely(bytes_sg_total > copy))
2630                 return -EINVAL;
2631
2632         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2633                            get_order(copy));
2634         if (unlikely(!page))
2635                 return -ENOMEM;
2636
2637         raw = page_address(page);
2638         i = first_sge;
2639         do {
2640                 sge = sk_msg_elem(msg, i);
2641                 from = sg_virt(sge);
2642                 len = sge->length;
2643                 to = raw + poffset;
2644
2645                 memcpy(to, from, len);
2646                 poffset += len;
2647                 sge->length = 0;
2648                 put_page(sg_page(sge));
2649
2650                 sk_msg_iter_var_next(i);
2651         } while (i != last_sge);
2652
2653         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2654
2655         /* To repair sg ring we need to shift entries. If we only
2656          * had a single entry though we can just replace it and
2657          * be done. Otherwise walk the ring and shift the entries.
2658          */
2659         WARN_ON_ONCE(last_sge == first_sge);
2660         shift = last_sge > first_sge ?
2661                 last_sge - first_sge - 1 :
2662                 NR_MSG_FRAG_IDS - first_sge + last_sge - 1;
2663         if (!shift)
2664                 goto out;
2665
2666         i = first_sge;
2667         sk_msg_iter_var_next(i);
2668         do {
2669                 u32 move_from;
2670
2671                 if (i + shift >= NR_MSG_FRAG_IDS)
2672                         move_from = i + shift - NR_MSG_FRAG_IDS;
2673                 else
2674                         move_from = i + shift;
2675                 if (move_from == msg->sg.end)
2676                         break;
2677
2678                 msg->sg.data[i] = msg->sg.data[move_from];
2679                 msg->sg.data[move_from].length = 0;
2680                 msg->sg.data[move_from].page_link = 0;
2681                 msg->sg.data[move_from].offset = 0;
2682                 sk_msg_iter_var_next(i);
2683         } while (1);
2684
2685         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2686                       msg->sg.end - shift + NR_MSG_FRAG_IDS :
2687                       msg->sg.end - shift;
2688 out:
2689         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2690         msg->data_end = msg->data + bytes;
2691         return 0;
2692 }
2693
2694 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2695         .func           = bpf_msg_pull_data,
2696         .gpl_only       = false,
2697         .ret_type       = RET_INTEGER,
2698         .arg1_type      = ARG_PTR_TO_CTX,
2699         .arg2_type      = ARG_ANYTHING,
2700         .arg3_type      = ARG_ANYTHING,
2701         .arg4_type      = ARG_ANYTHING,
2702 };
2703
2704 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2705            u32, len, u64, flags)
2706 {
2707         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2708         u32 new, i = 0, l = 0, space, copy = 0, offset = 0;
2709         u8 *raw, *to, *from;
2710         struct page *page;
2711
2712         if (unlikely(flags))
2713                 return -EINVAL;
2714
2715         if (unlikely(len == 0))
2716                 return 0;
2717
2718         /* First find the starting scatterlist element */
2719         i = msg->sg.start;
2720         do {
2721                 offset += l;
2722                 l = sk_msg_elem(msg, i)->length;
2723
2724                 if (start < offset + l)
2725                         break;
2726                 sk_msg_iter_var_next(i);
2727         } while (i != msg->sg.end);
2728
2729         if (start >= offset + l)
2730                 return -EINVAL;
2731
2732         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2733
2734         /* If no space available will fallback to copy, we need at
2735          * least one scatterlist elem available to push data into
2736          * when start aligns to the beginning of an element or two
2737          * when it falls inside an element. We handle the start equals
2738          * offset case because its the common case for inserting a
2739          * header.
2740          */
2741         if (!space || (space == 1 && start != offset))
2742                 copy = msg->sg.data[i].length;
2743
2744         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2745                            get_order(copy + len));
2746         if (unlikely(!page))
2747                 return -ENOMEM;
2748
2749         if (copy) {
2750                 int front, back;
2751
2752                 raw = page_address(page);
2753
2754                 psge = sk_msg_elem(msg, i);
2755                 front = start - offset;
2756                 back = psge->length - front;
2757                 from = sg_virt(psge);
2758
2759                 if (front)
2760                         memcpy(raw, from, front);
2761
2762                 if (back) {
2763                         from += front;
2764                         to = raw + front + len;
2765
2766                         memcpy(to, from, back);
2767                 }
2768
2769                 put_page(sg_page(psge));
2770         } else if (start - offset) {
2771                 psge = sk_msg_elem(msg, i);
2772                 rsge = sk_msg_elem_cpy(msg, i);
2773
2774                 psge->length = start - offset;
2775                 rsge.length -= psge->length;
2776                 rsge.offset += start;
2777
2778                 sk_msg_iter_var_next(i);
2779                 sg_unmark_end(psge);
2780                 sg_unmark_end(&rsge);
2781                 sk_msg_iter_next(msg, end);
2782         }
2783
2784         /* Slot(s) to place newly allocated data */
2785         new = i;
2786
2787         /* Shift one or two slots as needed */
2788         if (!copy) {
2789                 sge = sk_msg_elem_cpy(msg, i);
2790
2791                 sk_msg_iter_var_next(i);
2792                 sg_unmark_end(&sge);
2793                 sk_msg_iter_next(msg, end);
2794
2795                 nsge = sk_msg_elem_cpy(msg, i);
2796                 if (rsge.length) {
2797                         sk_msg_iter_var_next(i);
2798                         nnsge = sk_msg_elem_cpy(msg, i);
2799                 }
2800
2801                 while (i != msg->sg.end) {
2802                         msg->sg.data[i] = sge;
2803                         sge = nsge;
2804                         sk_msg_iter_var_next(i);
2805                         if (rsge.length) {
2806                                 nsge = nnsge;
2807                                 nnsge = sk_msg_elem_cpy(msg, i);
2808                         } else {
2809                                 nsge = sk_msg_elem_cpy(msg, i);
2810                         }
2811                 }
2812         }
2813
2814         /* Place newly allocated data buffer */
2815         sk_mem_charge(msg->sk, len);
2816         msg->sg.size += len;
2817         __clear_bit(new, msg->sg.copy);
2818         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2819         if (rsge.length) {
2820                 get_page(sg_page(&rsge));
2821                 sk_msg_iter_var_next(new);
2822                 msg->sg.data[new] = rsge;
2823         }
2824
2825         sk_msg_compute_data_pointers(msg);
2826         return 0;
2827 }
2828
2829 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2830         .func           = bpf_msg_push_data,
2831         .gpl_only       = false,
2832         .ret_type       = RET_INTEGER,
2833         .arg1_type      = ARG_PTR_TO_CTX,
2834         .arg2_type      = ARG_ANYTHING,
2835         .arg3_type      = ARG_ANYTHING,
2836         .arg4_type      = ARG_ANYTHING,
2837 };
2838
2839 static void sk_msg_shift_left(struct sk_msg *msg, int i)
2840 {
2841         int prev;
2842
2843         do {
2844                 prev = i;
2845                 sk_msg_iter_var_next(i);
2846                 msg->sg.data[prev] = msg->sg.data[i];
2847         } while (i != msg->sg.end);
2848
2849         sk_msg_iter_prev(msg, end);
2850 }
2851
2852 static void sk_msg_shift_right(struct sk_msg *msg, int i)
2853 {
2854         struct scatterlist tmp, sge;
2855
2856         sk_msg_iter_next(msg, end);
2857         sge = sk_msg_elem_cpy(msg, i);
2858         sk_msg_iter_var_next(i);
2859         tmp = sk_msg_elem_cpy(msg, i);
2860
2861         while (i != msg->sg.end) {
2862                 msg->sg.data[i] = sge;
2863                 sk_msg_iter_var_next(i);
2864                 sge = tmp;
2865                 tmp = sk_msg_elem_cpy(msg, i);
2866         }
2867 }
2868
2869 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start,
2870            u32, len, u64, flags)
2871 {
2872         u32 i = 0, l = 0, space, offset = 0;
2873         u64 last = start + len;
2874         int pop;
2875
2876         if (unlikely(flags))
2877                 return -EINVAL;
2878
2879         /* First find the starting scatterlist element */
2880         i = msg->sg.start;
2881         do {
2882                 offset += l;
2883                 l = sk_msg_elem(msg, i)->length;
2884
2885                 if (start < offset + l)
2886                         break;
2887                 sk_msg_iter_var_next(i);
2888         } while (i != msg->sg.end);
2889
2890         /* Bounds checks: start and pop must be inside message */
2891         if (start >= offset + l || last >= msg->sg.size)
2892                 return -EINVAL;
2893
2894         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2895
2896         pop = len;
2897         /* --------------| offset
2898          * -| start      |-------- len -------|
2899          *
2900          *  |----- a ----|-------- pop -------|----- b ----|
2901          *  |______________________________________________| length
2902          *
2903          *
2904          * a:   region at front of scatter element to save
2905          * b:   region at back of scatter element to save when length > A + pop
2906          * pop: region to pop from element, same as input 'pop' here will be
2907          *      decremented below per iteration.
2908          *
2909          * Two top-level cases to handle when start != offset, first B is non
2910          * zero and second B is zero corresponding to when a pop includes more
2911          * than one element.
2912          *
2913          * Then if B is non-zero AND there is no space allocate space and
2914          * compact A, B regions into page. If there is space shift ring to
2915          * the rigth free'ing the next element in ring to place B, leaving
2916          * A untouched except to reduce length.
2917          */
2918         if (start != offset) {
2919                 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i);
2920                 int a = start;
2921                 int b = sge->length - pop - a;
2922
2923                 sk_msg_iter_var_next(i);
2924
2925                 if (pop < sge->length - a) {
2926                         if (space) {
2927                                 sge->length = a;
2928                                 sk_msg_shift_right(msg, i);
2929                                 nsge = sk_msg_elem(msg, i);
2930                                 get_page(sg_page(sge));
2931                                 sg_set_page(nsge,
2932                                             sg_page(sge),
2933                                             b, sge->offset + pop + a);
2934                         } else {
2935                                 struct page *page, *orig;
2936                                 u8 *to, *from;
2937
2938                                 page = alloc_pages(__GFP_NOWARN |
2939                                                    __GFP_COMP   | GFP_ATOMIC,
2940                                                    get_order(a + b));
2941                                 if (unlikely(!page))
2942                                         return -ENOMEM;
2943
2944                                 sge->length = a;
2945                                 orig = sg_page(sge);
2946                                 from = sg_virt(sge);
2947                                 to = page_address(page);
2948                                 memcpy(to, from, a);
2949                                 memcpy(to + a, from + a + pop, b);
2950                                 sg_set_page(sge, page, a + b, 0);
2951                                 put_page(orig);
2952                         }
2953                         pop = 0;
2954                 } else if (pop >= sge->length - a) {
2955                         pop -= (sge->length - a);
2956                         sge->length = a;
2957                 }
2958         }
2959
2960         /* From above the current layout _must_ be as follows,
2961          *
2962          * -| offset
2963          * -| start
2964          *
2965          *  |---- pop ---|---------------- b ------------|
2966          *  |____________________________________________| length
2967          *
2968          * Offset and start of the current msg elem are equal because in the
2969          * previous case we handled offset != start and either consumed the
2970          * entire element and advanced to the next element OR pop == 0.
2971          *
2972          * Two cases to handle here are first pop is less than the length
2973          * leaving some remainder b above. Simply adjust the element's layout
2974          * in this case. Or pop >= length of the element so that b = 0. In this
2975          * case advance to next element decrementing pop.
2976          */
2977         while (pop) {
2978                 struct scatterlist *sge = sk_msg_elem(msg, i);
2979
2980                 if (pop < sge->length) {
2981                         sge->length -= pop;
2982                         sge->offset += pop;
2983                         pop = 0;
2984                 } else {
2985                         pop -= sge->length;
2986                         sk_msg_shift_left(msg, i);
2987                 }
2988                 sk_msg_iter_var_next(i);
2989         }
2990
2991         sk_mem_uncharge(msg->sk, len - pop);
2992         msg->sg.size -= (len - pop);
2993         sk_msg_compute_data_pointers(msg);
2994         return 0;
2995 }
2996
2997 static const struct bpf_func_proto bpf_msg_pop_data_proto = {
2998         .func           = bpf_msg_pop_data,
2999         .gpl_only       = false,
3000         .ret_type       = RET_INTEGER,
3001         .arg1_type      = ARG_PTR_TO_CTX,
3002         .arg2_type      = ARG_ANYTHING,
3003         .arg3_type      = ARG_ANYTHING,
3004         .arg4_type      = ARG_ANYTHING,
3005 };
3006
3007 #ifdef CONFIG_CGROUP_NET_CLASSID
3008 BPF_CALL_0(bpf_get_cgroup_classid_curr)
3009 {
3010         return __task_get_classid(current);
3011 }
3012
3013 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = {
3014         .func           = bpf_get_cgroup_classid_curr,
3015         .gpl_only       = false,
3016         .ret_type       = RET_INTEGER,
3017 };
3018
3019 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb)
3020 {
3021         struct sock *sk = skb_to_full_sk(skb);
3022
3023         if (!sk || !sk_fullsock(sk))
3024                 return 0;
3025
3026         return sock_cgroup_classid(&sk->sk_cgrp_data);
3027 }
3028
3029 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = {
3030         .func           = bpf_skb_cgroup_classid,
3031         .gpl_only       = false,
3032         .ret_type       = RET_INTEGER,
3033         .arg1_type      = ARG_PTR_TO_CTX,
3034 };
3035 #endif
3036
3037 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
3038 {
3039         return task_get_classid(skb);
3040 }
3041
3042 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
3043         .func           = bpf_get_cgroup_classid,
3044         .gpl_only       = false,
3045         .ret_type       = RET_INTEGER,
3046         .arg1_type      = ARG_PTR_TO_CTX,
3047 };
3048
3049 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
3050 {
3051         return dst_tclassid(skb);
3052 }
3053
3054 static const struct bpf_func_proto bpf_get_route_realm_proto = {
3055         .func           = bpf_get_route_realm,
3056         .gpl_only       = false,
3057         .ret_type       = RET_INTEGER,
3058         .arg1_type      = ARG_PTR_TO_CTX,
3059 };
3060
3061 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
3062 {
3063         /* If skb_clear_hash() was called due to mangling, we can
3064          * trigger SW recalculation here. Later access to hash
3065          * can then use the inline skb->hash via context directly
3066          * instead of calling this helper again.
3067          */
3068         return skb_get_hash(skb);
3069 }
3070
3071 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
3072         .func           = bpf_get_hash_recalc,
3073         .gpl_only       = false,
3074         .ret_type       = RET_INTEGER,
3075         .arg1_type      = ARG_PTR_TO_CTX,
3076 };
3077
3078 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
3079 {
3080         /* After all direct packet write, this can be used once for
3081          * triggering a lazy recalc on next skb_get_hash() invocation.
3082          */
3083         skb_clear_hash(skb);
3084         return 0;
3085 }
3086
3087 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
3088         .func           = bpf_set_hash_invalid,
3089         .gpl_only       = false,
3090         .ret_type       = RET_INTEGER,
3091         .arg1_type      = ARG_PTR_TO_CTX,
3092 };
3093
3094 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
3095 {
3096         /* Set user specified hash as L4(+), so that it gets returned
3097          * on skb_get_hash() call unless BPF prog later on triggers a
3098          * skb_clear_hash().
3099          */
3100         __skb_set_sw_hash(skb, hash, true);
3101         return 0;
3102 }
3103
3104 static const struct bpf_func_proto bpf_set_hash_proto = {
3105         .func           = bpf_set_hash,
3106         .gpl_only       = false,
3107         .ret_type       = RET_INTEGER,
3108         .arg1_type      = ARG_PTR_TO_CTX,
3109         .arg2_type      = ARG_ANYTHING,
3110 };
3111
3112 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
3113            u16, vlan_tci)
3114 {
3115         int ret;
3116
3117         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
3118                      vlan_proto != htons(ETH_P_8021AD)))
3119                 vlan_proto = htons(ETH_P_8021Q);
3120
3121         bpf_push_mac_rcsum(skb);
3122         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
3123         bpf_pull_mac_rcsum(skb);
3124
3125         bpf_compute_data_pointers(skb);
3126         return ret;
3127 }
3128
3129 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
3130         .func           = bpf_skb_vlan_push,
3131         .gpl_only       = false,
3132         .ret_type       = RET_INTEGER,
3133         .arg1_type      = ARG_PTR_TO_CTX,
3134         .arg2_type      = ARG_ANYTHING,
3135         .arg3_type      = ARG_ANYTHING,
3136 };
3137
3138 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
3139 {
3140         int ret;
3141
3142         bpf_push_mac_rcsum(skb);
3143         ret = skb_vlan_pop(skb);
3144         bpf_pull_mac_rcsum(skb);
3145
3146         bpf_compute_data_pointers(skb);
3147         return ret;
3148 }
3149
3150 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
3151         .func           = bpf_skb_vlan_pop,
3152         .gpl_only       = false,
3153         .ret_type       = RET_INTEGER,
3154         .arg1_type      = ARG_PTR_TO_CTX,
3155 };
3156
3157 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
3158 {
3159         /* Caller already did skb_cow() with len as headroom,
3160          * so no need to do it here.
3161          */
3162         skb_push(skb, len);
3163         memmove(skb->data, skb->data + len, off);
3164         memset(skb->data + off, 0, len);
3165
3166         /* No skb_postpush_rcsum(skb, skb->data + off, len)
3167          * needed here as it does not change the skb->csum
3168          * result for checksum complete when summing over
3169          * zeroed blocks.
3170          */
3171         return 0;
3172 }
3173
3174 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
3175 {
3176         /* skb_ensure_writable() is not needed here, as we're
3177          * already working on an uncloned skb.
3178          */
3179         if (unlikely(!pskb_may_pull(skb, off + len)))
3180                 return -ENOMEM;
3181
3182         skb_postpull_rcsum(skb, skb->data + off, len);
3183         memmove(skb->data + len, skb->data, off);
3184         __skb_pull(skb, len);
3185
3186         return 0;
3187 }
3188
3189 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
3190 {
3191         bool trans_same = skb->transport_header == skb->network_header;
3192         int ret;
3193
3194         /* There's no need for __skb_push()/__skb_pull() pair to
3195          * get to the start of the mac header as we're guaranteed
3196          * to always start from here under eBPF.
3197          */
3198         ret = bpf_skb_generic_push(skb, off, len);
3199         if (likely(!ret)) {
3200                 skb->mac_header -= len;
3201                 skb->network_header -= len;
3202                 if (trans_same)
3203                         skb->transport_header = skb->network_header;
3204         }
3205
3206         return ret;
3207 }
3208
3209 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
3210 {
3211         bool trans_same = skb->transport_header == skb->network_header;
3212         int ret;
3213
3214         /* Same here, __skb_push()/__skb_pull() pair not needed. */
3215         ret = bpf_skb_generic_pop(skb, off, len);
3216         if (likely(!ret)) {
3217                 skb->mac_header += len;
3218                 skb->network_header += len;
3219                 if (trans_same)
3220                         skb->transport_header = skb->network_header;
3221         }
3222
3223         return ret;
3224 }
3225
3226 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
3227 {
3228         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3229         u32 off = skb_mac_header_len(skb);
3230         int ret;
3231
3232         ret = skb_cow(skb, len_diff);
3233         if (unlikely(ret < 0))
3234                 return ret;
3235
3236         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3237         if (unlikely(ret < 0))
3238                 return ret;
3239
3240         if (skb_is_gso(skb)) {
3241                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3242
3243                 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */
3244                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
3245                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
3246                         shinfo->gso_type |=  SKB_GSO_TCPV6;
3247                 }
3248         }
3249
3250         skb->protocol = htons(ETH_P_IPV6);
3251         skb_clear_hash(skb);
3252
3253         return 0;
3254 }
3255
3256 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
3257 {
3258         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
3259         u32 off = skb_mac_header_len(skb);
3260         int ret;
3261
3262         ret = skb_unclone(skb, GFP_ATOMIC);
3263         if (unlikely(ret < 0))
3264                 return ret;
3265
3266         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3267         if (unlikely(ret < 0))
3268                 return ret;
3269
3270         if (skb_is_gso(skb)) {
3271                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3272
3273                 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */
3274                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
3275                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
3276                         shinfo->gso_type |=  SKB_GSO_TCPV4;
3277                 }
3278         }
3279
3280         skb->protocol = htons(ETH_P_IP);
3281         skb_clear_hash(skb);
3282
3283         return 0;
3284 }
3285
3286 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
3287 {
3288         __be16 from_proto = skb->protocol;
3289
3290         if (from_proto == htons(ETH_P_IP) &&
3291               to_proto == htons(ETH_P_IPV6))
3292                 return bpf_skb_proto_4_to_6(skb);
3293
3294         if (from_proto == htons(ETH_P_IPV6) &&
3295               to_proto == htons(ETH_P_IP))
3296                 return bpf_skb_proto_6_to_4(skb);
3297
3298         return -ENOTSUPP;
3299 }
3300
3301 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
3302            u64, flags)
3303 {
3304         int ret;
3305
3306         if (unlikely(flags))
3307                 return -EINVAL;
3308
3309         /* General idea is that this helper does the basic groundwork
3310          * needed for changing the protocol, and eBPF program fills the
3311          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
3312          * and other helpers, rather than passing a raw buffer here.
3313          *
3314          * The rationale is to keep this minimal and without a need to
3315          * deal with raw packet data. F.e. even if we would pass buffers
3316          * here, the program still needs to call the bpf_lX_csum_replace()
3317          * helpers anyway. Plus, this way we keep also separation of
3318          * concerns, since f.e. bpf_skb_store_bytes() should only take
3319          * care of stores.
3320          *
3321          * Currently, additional options and extension header space are
3322          * not supported, but flags register is reserved so we can adapt
3323          * that. For offloads, we mark packet as dodgy, so that headers
3324          * need to be verified first.
3325          */
3326         ret = bpf_skb_proto_xlat(skb, proto);
3327         bpf_compute_data_pointers(skb);
3328         return ret;
3329 }
3330
3331 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
3332         .func           = bpf_skb_change_proto,
3333         .gpl_only       = false,
3334         .ret_type       = RET_INTEGER,
3335         .arg1_type      = ARG_PTR_TO_CTX,
3336         .arg2_type      = ARG_ANYTHING,
3337         .arg3_type      = ARG_ANYTHING,
3338 };
3339
3340 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
3341 {
3342         /* We only allow a restricted subset to be changed for now. */
3343         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
3344                      !skb_pkt_type_ok(pkt_type)))
3345                 return -EINVAL;
3346
3347         skb->pkt_type = pkt_type;
3348         return 0;
3349 }
3350
3351 static const struct bpf_func_proto bpf_skb_change_type_proto = {
3352         .func           = bpf_skb_change_type,
3353         .gpl_only       = false,
3354         .ret_type       = RET_INTEGER,
3355         .arg1_type      = ARG_PTR_TO_CTX,
3356         .arg2_type      = ARG_ANYTHING,
3357 };
3358
3359 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
3360 {
3361         switch (skb->protocol) {
3362         case htons(ETH_P_IP):
3363                 return sizeof(struct iphdr);
3364         case htons(ETH_P_IPV6):
3365                 return sizeof(struct ipv6hdr);
3366         default:
3367                 return ~0U;
3368         }
3369 }
3370
3371 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK    (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \
3372                                          BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3373
3374 #define BPF_F_ADJ_ROOM_MASK             (BPF_F_ADJ_ROOM_FIXED_GSO | \
3375                                          BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \
3376                                          BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \
3377                                          BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \
3378                                          BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \
3379                                          BPF_F_ADJ_ROOM_ENCAP_L2( \
3380                                           BPF_ADJ_ROOM_ENCAP_L2_MASK))
3381
3382 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff,
3383                             u64 flags)
3384 {
3385         u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT;
3386         bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK;
3387         u16 mac_len = 0, inner_net = 0, inner_trans = 0;
3388         unsigned int gso_type = SKB_GSO_DODGY;
3389         int ret;
3390
3391         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3392                 /* udp gso_size delineates datagrams, only allow if fixed */
3393                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3394                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3395                         return -ENOTSUPP;
3396         }
3397
3398         ret = skb_cow_head(skb, len_diff);
3399         if (unlikely(ret < 0))
3400                 return ret;
3401
3402         if (encap) {
3403                 if (skb->protocol != htons(ETH_P_IP) &&
3404                     skb->protocol != htons(ETH_P_IPV6))
3405                         return -ENOTSUPP;
3406
3407                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 &&
3408                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3409                         return -EINVAL;
3410
3411                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE &&
3412                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3413                         return -EINVAL;
3414
3415                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH &&
3416                     inner_mac_len < ETH_HLEN)
3417                         return -EINVAL;
3418
3419                 if (skb->encapsulation)
3420                         return -EALREADY;
3421
3422                 mac_len = skb->network_header - skb->mac_header;
3423                 inner_net = skb->network_header;
3424                 if (inner_mac_len > len_diff)
3425                         return -EINVAL;
3426                 inner_trans = skb->transport_header;
3427         }
3428
3429         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
3430         if (unlikely(ret < 0))
3431                 return ret;
3432
3433         if (encap) {
3434                 skb->inner_mac_header = inner_net - inner_mac_len;
3435                 skb->inner_network_header = inner_net;
3436                 skb->inner_transport_header = inner_trans;
3437
3438                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH)
3439                         skb_set_inner_protocol(skb, htons(ETH_P_TEB));
3440                 else
3441                         skb_set_inner_protocol(skb, skb->protocol);
3442
3443                 skb->encapsulation = 1;
3444                 skb_set_network_header(skb, mac_len);
3445
3446                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP)
3447                         gso_type |= SKB_GSO_UDP_TUNNEL;
3448                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE)
3449                         gso_type |= SKB_GSO_GRE;
3450                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3451                         gso_type |= SKB_GSO_IPXIP6;
3452                 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3453                         gso_type |= SKB_GSO_IPXIP4;
3454
3455                 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE ||
3456                     flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) {
3457                         int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ?
3458                                         sizeof(struct ipv6hdr) :
3459                                         sizeof(struct iphdr);
3460
3461                         skb_set_transport_header(skb, mac_len + nh_len);
3462                 }
3463
3464                 /* Match skb->protocol to new outer l3 protocol */
3465                 if (skb->protocol == htons(ETH_P_IP) &&
3466                     flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6)
3467                         skb->protocol = htons(ETH_P_IPV6);
3468                 else if (skb->protocol == htons(ETH_P_IPV6) &&
3469                          flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4)
3470                         skb->protocol = htons(ETH_P_IP);
3471         }
3472
3473         if (skb_is_gso(skb)) {
3474                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3475
3476                 /* Due to header grow, MSS needs to be downgraded. */
3477                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3478                         skb_decrease_gso_size(shinfo, len_diff);
3479
3480                 /* Header must be checked, and gso_segs recomputed. */
3481                 shinfo->gso_type |= gso_type;
3482                 shinfo->gso_segs = 0;
3483         }
3484
3485         return 0;
3486 }
3487
3488 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff,
3489                               u64 flags)
3490 {
3491         int ret;
3492
3493         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO |
3494                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3495                 return -EINVAL;
3496
3497         if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) {
3498                 /* udp gso_size delineates datagrams, only allow if fixed */
3499                 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ||
3500                     !(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3501                         return -ENOTSUPP;
3502         }
3503
3504         ret = skb_unclone(skb, GFP_ATOMIC);
3505         if (unlikely(ret < 0))
3506                 return ret;
3507
3508         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
3509         if (unlikely(ret < 0))
3510                 return ret;
3511
3512         if (skb_is_gso(skb)) {
3513                 struct skb_shared_info *shinfo = skb_shinfo(skb);
3514
3515                 /* Due to header shrink, MSS can be upgraded. */
3516                 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO))
3517                         skb_increase_gso_size(shinfo, len_diff);
3518
3519                 /* Header must be checked, and gso_segs recomputed. */
3520                 shinfo->gso_type |= SKB_GSO_DODGY;
3521                 shinfo->gso_segs = 0;
3522         }
3523
3524         return 0;
3525 }
3526
3527 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC
3528
3529 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3530            u32, mode, u64, flags)
3531 {
3532         u32 len_diff_abs = abs(len_diff);
3533         bool shrink = len_diff < 0;
3534         int ret = 0;
3535
3536         if (unlikely(flags || mode))
3537                 return -EINVAL;
3538         if (unlikely(len_diff_abs > 0xfffU))
3539                 return -EFAULT;
3540
3541         if (!shrink) {
3542                 ret = skb_cow(skb, len_diff);
3543                 if (unlikely(ret < 0))
3544                         return ret;
3545                 __skb_push(skb, len_diff_abs);
3546                 memset(skb->data, 0, len_diff_abs);
3547         } else {
3548                 if (unlikely(!pskb_may_pull(skb, len_diff_abs)))
3549                         return -ENOMEM;
3550                 __skb_pull(skb, len_diff_abs);
3551         }
3552         if (tls_sw_has_ctx_rx(skb->sk)) {
3553                 struct strp_msg *rxm = strp_msg(skb);
3554
3555                 rxm->full_len += len_diff;
3556         }
3557         return ret;
3558 }
3559
3560 static const struct bpf_func_proto sk_skb_adjust_room_proto = {
3561         .func           = sk_skb_adjust_room,
3562         .gpl_only       = false,
3563         .ret_type       = RET_INTEGER,
3564         .arg1_type      = ARG_PTR_TO_CTX,
3565         .arg2_type      = ARG_ANYTHING,
3566         .arg3_type      = ARG_ANYTHING,
3567         .arg4_type      = ARG_ANYTHING,
3568 };
3569
3570 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
3571            u32, mode, u64, flags)
3572 {
3573         u32 len_cur, len_diff_abs = abs(len_diff);
3574         u32 len_min = bpf_skb_net_base_len(skb);
3575         u32 len_max = BPF_SKB_MAX_LEN;
3576         __be16 proto = skb->protocol;
3577         bool shrink = len_diff < 0;
3578         u32 off;
3579         int ret;
3580
3581         if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK |
3582                                BPF_F_ADJ_ROOM_NO_CSUM_RESET)))
3583                 return -EINVAL;
3584         if (unlikely(len_diff_abs > 0xfffU))
3585                 return -EFAULT;
3586         if (unlikely(proto != htons(ETH_P_IP) &&
3587                      proto != htons(ETH_P_IPV6)))
3588                 return -ENOTSUPP;
3589
3590         off = skb_mac_header_len(skb);
3591         switch (mode) {
3592         case BPF_ADJ_ROOM_NET:
3593                 off += bpf_skb_net_base_len(skb);
3594                 break;
3595         case BPF_ADJ_ROOM_MAC:
3596                 break;
3597         default:
3598                 return -ENOTSUPP;
3599         }
3600
3601         len_cur = skb->len - skb_network_offset(skb);
3602         if ((shrink && (len_diff_abs >= len_cur ||
3603                         len_cur - len_diff_abs < len_min)) ||
3604             (!shrink && (skb->len + len_diff_abs > len_max &&
3605                          !skb_is_gso(skb))))
3606                 return -ENOTSUPP;
3607
3608         ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) :
3609                        bpf_skb_net_grow(skb, off, len_diff_abs, flags);
3610         if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET))
3611                 __skb_reset_checksum_unnecessary(skb);
3612
3613         bpf_compute_data_pointers(skb);
3614         return ret;
3615 }
3616
3617 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
3618         .func           = bpf_skb_adjust_room,
3619         .gpl_only       = false,
3620         .ret_type       = RET_INTEGER,
3621         .arg1_type      = ARG_PTR_TO_CTX,
3622         .arg2_type      = ARG_ANYTHING,
3623         .arg3_type      = ARG_ANYTHING,
3624         .arg4_type      = ARG_ANYTHING,
3625 };
3626
3627 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
3628 {
3629         u32 min_len = skb_network_offset(skb);
3630
3631         if (skb_transport_header_was_set(skb))
3632                 min_len = skb_transport_offset(skb);
3633         if (skb->ip_summed == CHECKSUM_PARTIAL)
3634                 min_len = skb_checksum_start_offset(skb) +
3635                           skb->csum_offset + sizeof(__sum16);
3636         return min_len;
3637 }
3638
3639 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
3640 {
3641         unsigned int old_len = skb->len;
3642         int ret;
3643
3644         ret = __skb_grow_rcsum(skb, new_len);
3645         if (!ret)
3646                 memset(skb->data + old_len, 0, new_len - old_len);
3647         return ret;
3648 }
3649
3650 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
3651 {
3652         return __skb_trim_rcsum(skb, new_len);
3653 }
3654
3655 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
3656                                         u64 flags)
3657 {
3658         u32 max_len = BPF_SKB_MAX_LEN;
3659         u32 min_len = __bpf_skb_min_len(skb);
3660         int ret;
3661
3662         if (unlikely(flags || new_len > max_len || new_len < min_len))
3663                 return -EINVAL;
3664         if (skb->encapsulation)
3665                 return -ENOTSUPP;
3666
3667         /* The basic idea of this helper is that it's performing the
3668          * needed work to either grow or trim an skb, and eBPF program
3669          * rewrites the rest via helpers like bpf_skb_store_bytes(),
3670          * bpf_lX_csum_replace() and others rather than passing a raw
3671          * buffer here. This one is a slow path helper and intended
3672          * for replies with control messages.
3673          *
3674          * Like in bpf_skb_change_proto(), we want to keep this rather
3675          * minimal and without protocol specifics so that we are able
3676          * to separate concerns as in bpf_skb_store_bytes() should only
3677          * be the one responsible for writing buffers.
3678          *
3679          * It's really expected to be a slow path operation here for
3680          * control message replies, so we're implicitly linearizing,
3681          * uncloning and drop offloads from the skb by this.
3682          */
3683         ret = __bpf_try_make_writable(skb, skb->len);
3684         if (!ret) {
3685                 if (new_len > skb->len)
3686                         ret = bpf_skb_grow_rcsum(skb, new_len);
3687                 else if (new_len < skb->len)
3688                         ret = bpf_skb_trim_rcsum(skb, new_len);
3689                 if (!ret && skb_is_gso(skb))
3690                         skb_gso_reset(skb);
3691         }
3692         return ret;
3693 }
3694
3695 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3696            u64, flags)
3697 {
3698         int ret = __bpf_skb_change_tail(skb, new_len, flags);
3699
3700         bpf_compute_data_pointers(skb);
3701         return ret;
3702 }
3703
3704 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
3705         .func           = bpf_skb_change_tail,
3706         .gpl_only       = false,
3707         .ret_type       = RET_INTEGER,
3708         .arg1_type      = ARG_PTR_TO_CTX,
3709         .arg2_type      = ARG_ANYTHING,
3710         .arg3_type      = ARG_ANYTHING,
3711 };
3712
3713 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
3714            u64, flags)
3715 {
3716         return __bpf_skb_change_tail(skb, new_len, flags);
3717 }
3718
3719 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3720         .func           = sk_skb_change_tail,
3721         .gpl_only       = false,
3722         .ret_type       = RET_INTEGER,
3723         .arg1_type      = ARG_PTR_TO_CTX,
3724         .arg2_type      = ARG_ANYTHING,
3725         .arg3_type      = ARG_ANYTHING,
3726 };
3727
3728 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3729                                         u64 flags)
3730 {
3731         u32 max_len = BPF_SKB_MAX_LEN;
3732         u32 new_len = skb->len + head_room;
3733         int ret;
3734
3735         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3736                      new_len < skb->len))
3737                 return -EINVAL;
3738
3739         ret = skb_cow(skb, head_room);
3740         if (likely(!ret)) {
3741                 /* Idea for this helper is that we currently only
3742                  * allow to expand on mac header. This means that
3743                  * skb->protocol network header, etc, stay as is.
3744                  * Compared to bpf_skb_change_tail(), we're more
3745                  * flexible due to not needing to linearize or
3746                  * reset GSO. Intention for this helper is to be
3747                  * used by an L3 skb that needs to push mac header
3748                  * for redirection into L2 device.
3749                  */
3750                 __skb_push(skb, head_room);
3751                 memset(skb->data, 0, head_room);
3752                 skb_reset_mac_header(skb);
3753                 skb_reset_mac_len(skb);
3754         }
3755
3756         return ret;
3757 }
3758
3759 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3760            u64, flags)
3761 {
3762         int ret = __bpf_skb_change_head(skb, head_room, flags);
3763
3764         bpf_compute_data_pointers(skb);
3765         return ret;
3766 }
3767
3768 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3769         .func           = bpf_skb_change_head,
3770         .gpl_only       = false,
3771         .ret_type       = RET_INTEGER,
3772         .arg1_type      = ARG_PTR_TO_CTX,
3773         .arg2_type      = ARG_ANYTHING,
3774         .arg3_type      = ARG_ANYTHING,
3775 };
3776
3777 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3778            u64, flags)
3779 {
3780         return __bpf_skb_change_head(skb, head_room, flags);
3781 }
3782
3783 static const struct bpf_func_proto sk_skb_change_head_proto = {
3784         .func           = sk_skb_change_head,
3785         .gpl_only       = false,
3786         .ret_type       = RET_INTEGER,
3787         .arg1_type      = ARG_PTR_TO_CTX,
3788         .arg2_type      = ARG_ANYTHING,
3789         .arg3_type      = ARG_ANYTHING,
3790 };
3791
3792 BPF_CALL_1(bpf_xdp_get_buff_len, struct  xdp_buff*, xdp)
3793 {
3794         return xdp_get_buff_len(xdp);
3795 }
3796
3797 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = {
3798         .func           = bpf_xdp_get_buff_len,
3799         .gpl_only       = false,
3800         .ret_type       = RET_INTEGER,
3801         .arg1_type      = ARG_PTR_TO_CTX,
3802 };
3803
3804 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff)
3805
3806 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = {
3807         .func           = bpf_xdp_get_buff_len,
3808         .gpl_only       = false,
3809         .arg1_type      = ARG_PTR_TO_BTF_ID,
3810         .arg1_btf_id    = &bpf_xdp_get_buff_len_bpf_ids[0],
3811 };
3812
3813 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3814 {
3815         return xdp_data_meta_unsupported(xdp) ? 0 :
3816                xdp->data - xdp->data_meta;
3817 }
3818
3819 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3820 {
3821         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3822         unsigned long metalen = xdp_get_metalen(xdp);
3823         void *data_start = xdp_frame_end + metalen;
3824         void *data = xdp->data + offset;
3825
3826         if (unlikely(data < data_start ||
3827                      data > xdp->data_end - ETH_HLEN))
3828                 return -EINVAL;
3829
3830         if (metalen)
3831                 memmove(xdp->data_meta + offset,
3832                         xdp->data_meta, metalen);
3833         xdp->data_meta += offset;
3834         xdp->data = data;
3835
3836         return 0;
3837 }
3838
3839 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3840         .func           = bpf_xdp_adjust_head,
3841         .gpl_only       = false,
3842         .ret_type       = RET_INTEGER,
3843         .arg1_type      = ARG_PTR_TO_CTX,
3844         .arg2_type      = ARG_ANYTHING,
3845 };
3846
3847 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off,
3848                              void *buf, unsigned long len, bool flush)
3849 {
3850         unsigned long ptr_len, ptr_off = 0;
3851         skb_frag_t *next_frag, *end_frag;
3852         struct skb_shared_info *sinfo;
3853         void *src, *dst;
3854         u8 *ptr_buf;
3855
3856         if (likely(xdp->data_end - xdp->data >= off + len)) {
3857                 src = flush ? buf : xdp->data + off;
3858                 dst = flush ? xdp->data + off : buf;
3859                 memcpy(dst, src, len);
3860                 return;
3861         }
3862
3863         sinfo = xdp_get_shared_info_from_buff(xdp);
3864         end_frag = &sinfo->frags[sinfo->nr_frags];
3865         next_frag = &sinfo->frags[0];
3866
3867         ptr_len = xdp->data_end - xdp->data;
3868         ptr_buf = xdp->data;
3869
3870         while (true) {
3871                 if (off < ptr_off + ptr_len) {
3872                         unsigned long copy_off = off - ptr_off;
3873                         unsigned long copy_len = min(len, ptr_len - copy_off);
3874
3875                         src = flush ? buf : ptr_buf + copy_off;
3876                         dst = flush ? ptr_buf + copy_off : buf;
3877                         memcpy(dst, src, copy_len);
3878
3879                         off += copy_len;
3880                         len -= copy_len;
3881                         buf += copy_len;
3882                 }
3883
3884                 if (!len || next_frag == end_frag)
3885                         break;
3886
3887                 ptr_off += ptr_len;
3888                 ptr_buf = skb_frag_address(next_frag);
3889                 ptr_len = skb_frag_size(next_frag);
3890                 next_frag++;
3891         }
3892 }
3893
3894 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len)
3895 {
3896         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3897         u32 size = xdp->data_end - xdp->data;
3898         void *addr = xdp->data;
3899         int i;
3900
3901         if (unlikely(offset > 0xffff || len > 0xffff))
3902                 return ERR_PTR(-EFAULT);
3903
3904         if (offset + len > xdp_get_buff_len(xdp))
3905                 return ERR_PTR(-EINVAL);
3906
3907         if (offset < size) /* linear area */
3908                 goto out;
3909
3910         offset -= size;
3911         for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */
3912                 u32 frag_size = skb_frag_size(&sinfo->frags[i]);
3913
3914                 if  (offset < frag_size) {
3915                         addr = skb_frag_address(&sinfo->frags[i]);
3916                         size = frag_size;
3917                         break;
3918                 }
3919                 offset -= frag_size;
3920         }
3921 out:
3922         return offset + len <= size ? addr + offset : NULL;
3923 }
3924
3925 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset,
3926            void *, buf, u32, len)
3927 {
3928         void *ptr;
3929
3930         ptr = bpf_xdp_pointer(xdp, offset, len);
3931         if (IS_ERR(ptr))
3932                 return PTR_ERR(ptr);
3933
3934         if (!ptr)
3935                 bpf_xdp_copy_buf(xdp, offset, buf, len, false);
3936         else
3937                 memcpy(buf, ptr, len);
3938
3939         return 0;
3940 }
3941
3942 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = {
3943         .func           = bpf_xdp_load_bytes,
3944         .gpl_only       = false,
3945         .ret_type       = RET_INTEGER,
3946         .arg1_type      = ARG_PTR_TO_CTX,
3947         .arg2_type      = ARG_ANYTHING,
3948         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
3949         .arg4_type      = ARG_CONST_SIZE,
3950 };
3951
3952 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset,
3953            void *, buf, u32, len)
3954 {
3955         void *ptr;
3956
3957         ptr = bpf_xdp_pointer(xdp, offset, len);
3958         if (IS_ERR(ptr))
3959                 return PTR_ERR(ptr);
3960
3961         if (!ptr)
3962                 bpf_xdp_copy_buf(xdp, offset, buf, len, true);
3963         else
3964                 memcpy(ptr, buf, len);
3965
3966         return 0;
3967 }
3968
3969 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = {
3970         .func           = bpf_xdp_store_bytes,
3971         .gpl_only       = false,
3972         .ret_type       = RET_INTEGER,
3973         .arg1_type      = ARG_PTR_TO_CTX,
3974         .arg2_type      = ARG_ANYTHING,
3975         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
3976         .arg4_type      = ARG_CONST_SIZE,
3977 };
3978
3979 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset)
3980 {
3981         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
3982         skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1];
3983         struct xdp_rxq_info *rxq = xdp->rxq;
3984         unsigned int tailroom;
3985
3986         if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz)
3987                 return -EOPNOTSUPP;
3988
3989         tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag);
3990         if (unlikely(offset > tailroom))
3991                 return -EINVAL;
3992
3993         memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset);
3994         skb_frag_size_add(frag, offset);
3995         sinfo->xdp_frags_size += offset;
3996
3997         return 0;
3998 }
3999
4000 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset)
4001 {
4002         struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
4003         int i, n_frags_free = 0, len_free = 0;
4004
4005         if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN))
4006                 return -EINVAL;
4007
4008         for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) {
4009                 skb_frag_t *frag = &sinfo->frags[i];
4010                 int shrink = min_t(int, offset, skb_frag_size(frag));
4011
4012                 len_free += shrink;
4013                 offset -= shrink;
4014
4015                 if (skb_frag_size(frag) == shrink) {
4016                         struct page *page = skb_frag_page(frag);
4017
4018                         __xdp_return(page_address(page), &xdp->rxq->mem,
4019                                      false, NULL);
4020                         n_frags_free++;
4021                 } else {
4022                         skb_frag_size_sub(frag, shrink);
4023                         break;
4024                 }
4025         }
4026         sinfo->nr_frags -= n_frags_free;
4027         sinfo->xdp_frags_size -= len_free;
4028
4029         if (unlikely(!sinfo->nr_frags)) {
4030                 xdp_buff_clear_frags_flag(xdp);
4031                 xdp->data_end -= offset;
4032         }
4033
4034         return 0;
4035 }
4036
4037 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
4038 {
4039         void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */
4040         void *data_end = xdp->data_end + offset;
4041
4042         if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */
4043                 if (offset < 0)
4044                         return bpf_xdp_frags_shrink_tail(xdp, -offset);
4045
4046                 return bpf_xdp_frags_increase_tail(xdp, offset);
4047         }
4048
4049         /* Notice that xdp_data_hard_end have reserved some tailroom */
4050         if (unlikely(data_end > data_hard_end))
4051                 return -EINVAL;
4052
4053         /* ALL drivers MUST init xdp->frame_sz, chicken check below */
4054         if (unlikely(xdp->frame_sz > PAGE_SIZE)) {
4055                 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz);
4056                 return -EINVAL;
4057         }
4058
4059         if (unlikely(data_end < xdp->data + ETH_HLEN))
4060                 return -EINVAL;
4061
4062         /* Clear memory area on grow, can contain uninit kernel memory */
4063         if (offset > 0)
4064                 memset(xdp->data_end, 0, offset);
4065
4066         xdp->data_end = data_end;
4067
4068         return 0;
4069 }
4070
4071 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
4072         .func           = bpf_xdp_adjust_tail,
4073         .gpl_only       = false,
4074         .ret_type       = RET_INTEGER,
4075         .arg1_type      = ARG_PTR_TO_CTX,
4076         .arg2_type      = ARG_ANYTHING,
4077 };
4078
4079 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
4080 {
4081         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
4082         void *meta = xdp->data_meta + offset;
4083         unsigned long metalen = xdp->data - meta;
4084
4085         if (xdp_data_meta_unsupported(xdp))
4086                 return -ENOTSUPP;
4087         if (unlikely(meta < xdp_frame_end ||
4088                      meta > xdp->data))
4089                 return -EINVAL;
4090         if (unlikely(xdp_metalen_invalid(metalen)))
4091                 return -EACCES;
4092
4093         xdp->data_meta = meta;
4094
4095         return 0;
4096 }
4097
4098 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
4099         .func           = bpf_xdp_adjust_meta,
4100         .gpl_only       = false,
4101         .ret_type       = RET_INTEGER,
4102         .arg1_type      = ARG_PTR_TO_CTX,
4103         .arg2_type      = ARG_ANYTHING,
4104 };
4105
4106 /* XDP_REDIRECT works by a three-step process, implemented in the functions
4107  * below:
4108  *
4109  * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target
4110  *    of the redirect and store it (along with some other metadata) in a per-CPU
4111  *    struct bpf_redirect_info.
4112  *
4113  * 2. When the program returns the XDP_REDIRECT return code, the driver will
4114  *    call xdp_do_redirect() which will use the information in struct
4115  *    bpf_redirect_info to actually enqueue the frame into a map type-specific
4116  *    bulk queue structure.
4117  *
4118  * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(),
4119  *    which will flush all the different bulk queues, thus completing the
4120  *    redirect.
4121  *
4122  * Pointers to the map entries will be kept around for this whole sequence of
4123  * steps, protected by RCU. However, there is no top-level rcu_read_lock() in
4124  * the core code; instead, the RCU protection relies on everything happening
4125  * inside a single NAPI poll sequence, which means it's between a pair of calls
4126  * to local_bh_disable()/local_bh_enable().
4127  *
4128  * The map entries are marked as __rcu and the map code makes sure to
4129  * dereference those pointers with rcu_dereference_check() in a way that works
4130  * for both sections that to hold an rcu_read_lock() and sections that are
4131  * called from NAPI without a separate rcu_read_lock(). The code below does not
4132  * use RCU annotations, but relies on those in the map code.
4133  */
4134 void xdp_do_flush(void)
4135 {
4136         __dev_flush();
4137         __cpu_map_flush();
4138         __xsk_map_flush();
4139 }
4140 EXPORT_SYMBOL_GPL(xdp_do_flush);
4141
4142 void bpf_clear_redirect_map(struct bpf_map *map)
4143 {
4144         struct bpf_redirect_info *ri;
4145         int cpu;
4146
4147         for_each_possible_cpu(cpu) {
4148                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
4149                 /* Avoid polluting remote cacheline due to writes if
4150                  * not needed. Once we pass this test, we need the
4151                  * cmpxchg() to make sure it hasn't been changed in
4152                  * the meantime by remote CPU.
4153                  */
4154                 if (unlikely(READ_ONCE(ri->map) == map))
4155                         cmpxchg(&ri->map, map, NULL);
4156         }
4157 }
4158
4159 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
4160 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
4161
4162 u32 xdp_master_redirect(struct xdp_buff *xdp)
4163 {
4164         struct net_device *master, *slave;
4165         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4166
4167         master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
4168         slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
4169         if (slave && slave != xdp->rxq->dev) {
4170                 /* The target device is different from the receiving device, so
4171                  * redirect it to the new device.
4172                  * Using XDP_REDIRECT gets the correct behaviour from XDP enabled
4173                  * drivers to unmap the packet from their rx ring.
4174                  */
4175                 ri->tgt_index = slave->ifindex;
4176                 ri->map_id = INT_MAX;
4177                 ri->map_type = BPF_MAP_TYPE_UNSPEC;
4178                 return XDP_REDIRECT;
4179         }
4180         return XDP_TX;
4181 }
4182 EXPORT_SYMBOL_GPL(xdp_master_redirect);
4183
4184 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri,
4185                                         struct net_device *dev,
4186                                         struct xdp_buff *xdp,
4187                                         struct bpf_prog *xdp_prog)
4188 {
4189         enum bpf_map_type map_type = ri->map_type;
4190         void *fwd = ri->tgt_value;
4191         u32 map_id = ri->map_id;
4192         int err;
4193
4194         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4195         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4196
4197         err = __xsk_map_redirect(fwd, xdp);
4198         if (unlikely(err))
4199                 goto err;
4200
4201         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4202         return 0;
4203 err:
4204         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4205         return err;
4206 }
4207
4208 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
4209                                                    struct net_device *dev,
4210                                                    struct xdp_frame *xdpf,
4211                                                    struct bpf_prog *xdp_prog)
4212 {
4213         enum bpf_map_type map_type = ri->map_type;
4214         void *fwd = ri->tgt_value;
4215         u32 map_id = ri->map_id;
4216         struct bpf_map *map;
4217         int err;
4218
4219         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4220         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4221
4222         if (unlikely(!xdpf)) {
4223                 err = -EOVERFLOW;
4224                 goto err;
4225         }
4226
4227         switch (map_type) {
4228         case BPF_MAP_TYPE_DEVMAP:
4229                 fallthrough;
4230         case BPF_MAP_TYPE_DEVMAP_HASH:
4231                 map = READ_ONCE(ri->map);
4232                 if (unlikely(map)) {
4233                         WRITE_ONCE(ri->map, NULL);
4234                         err = dev_map_enqueue_multi(xdpf, dev, map,
4235                                                     ri->flags & BPF_F_EXCLUDE_INGRESS);
4236                 } else {
4237                         err = dev_map_enqueue(fwd, xdpf, dev);
4238                 }
4239                 break;
4240         case BPF_MAP_TYPE_CPUMAP:
4241                 err = cpu_map_enqueue(fwd, xdpf, dev);
4242                 break;
4243         case BPF_MAP_TYPE_UNSPEC:
4244                 if (map_id == INT_MAX) {
4245                         fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4246                         if (unlikely(!fwd)) {
4247                                 err = -EINVAL;
4248                                 break;
4249                         }
4250                         err = dev_xdp_enqueue(fwd, xdpf, dev);
4251                         break;
4252                 }
4253                 fallthrough;
4254         default:
4255                 err = -EBADRQC;
4256         }
4257
4258         if (unlikely(err))
4259                 goto err;
4260
4261         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4262         return 0;
4263 err:
4264         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4265         return err;
4266 }
4267
4268 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
4269                     struct bpf_prog *xdp_prog)
4270 {
4271         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4272         enum bpf_map_type map_type = ri->map_type;
4273
4274         /* XDP_REDIRECT is not fully supported yet for xdp frags since
4275          * not all XDP capable drivers can map non-linear xdp_frame in
4276          * ndo_xdp_xmit.
4277          */
4278         if (unlikely(xdp_buff_has_frags(xdp) &&
4279                      map_type != BPF_MAP_TYPE_CPUMAP))
4280                 return -EOPNOTSUPP;
4281
4282         if (map_type == BPF_MAP_TYPE_XSKMAP)
4283                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4284
4285         return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp),
4286                                        xdp_prog);
4287 }
4288 EXPORT_SYMBOL_GPL(xdp_do_redirect);
4289
4290 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
4291                           struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
4292 {
4293         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4294         enum bpf_map_type map_type = ri->map_type;
4295
4296         if (map_type == BPF_MAP_TYPE_XSKMAP)
4297                 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog);
4298
4299         return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog);
4300 }
4301 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame);
4302
4303 static int xdp_do_generic_redirect_map(struct net_device *dev,
4304                                        struct sk_buff *skb,
4305                                        struct xdp_buff *xdp,
4306                                        struct bpf_prog *xdp_prog,
4307                                        void *fwd,
4308                                        enum bpf_map_type map_type, u32 map_id)
4309 {
4310         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4311         struct bpf_map *map;
4312         int err;
4313
4314         switch (map_type) {
4315         case BPF_MAP_TYPE_DEVMAP:
4316                 fallthrough;
4317         case BPF_MAP_TYPE_DEVMAP_HASH:
4318                 map = READ_ONCE(ri->map);
4319                 if (unlikely(map)) {
4320                         WRITE_ONCE(ri->map, NULL);
4321                         err = dev_map_redirect_multi(dev, skb, xdp_prog, map,
4322                                                      ri->flags & BPF_F_EXCLUDE_INGRESS);
4323                 } else {
4324                         err = dev_map_generic_redirect(fwd, skb, xdp_prog);
4325                 }
4326                 if (unlikely(err))
4327                         goto err;
4328                 break;
4329         case BPF_MAP_TYPE_XSKMAP:
4330                 err = xsk_generic_rcv(fwd, xdp);
4331                 if (err)
4332                         goto err;
4333                 consume_skb(skb);
4334                 break;
4335         case BPF_MAP_TYPE_CPUMAP:
4336                 err = cpu_map_generic_redirect(fwd, skb);
4337                 if (unlikely(err))
4338                         goto err;
4339                 break;
4340         default:
4341                 err = -EBADRQC;
4342                 goto err;
4343         }
4344
4345         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index);
4346         return 0;
4347 err:
4348         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err);
4349         return err;
4350 }
4351
4352 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
4353                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
4354 {
4355         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4356         enum bpf_map_type map_type = ri->map_type;
4357         void *fwd = ri->tgt_value;
4358         u32 map_id = ri->map_id;
4359         int err;
4360
4361         ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */
4362         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4363
4364         if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) {
4365                 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index);
4366                 if (unlikely(!fwd)) {
4367                         err = -EINVAL;
4368                         goto err;
4369                 }
4370
4371                 err = xdp_ok_fwd_dev(fwd, skb->len);
4372                 if (unlikely(err))
4373                         goto err;
4374
4375                 skb->dev = fwd;
4376                 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index);
4377                 generic_xdp_tx(skb, xdp_prog);
4378                 return 0;
4379         }
4380
4381         return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id);
4382 err:
4383         _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err);
4384         return err;
4385 }
4386
4387 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
4388 {
4389         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
4390
4391         if (unlikely(flags))
4392                 return XDP_ABORTED;
4393
4394         /* NB! Map type UNSPEC and map_id == INT_MAX (never generated
4395          * by map_idr) is used for ifindex based XDP redirect.
4396          */
4397         ri->tgt_index = ifindex;
4398         ri->map_id = INT_MAX;
4399         ri->map_type = BPF_MAP_TYPE_UNSPEC;
4400
4401         return XDP_REDIRECT;
4402 }
4403
4404 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
4405         .func           = bpf_xdp_redirect,
4406         .gpl_only       = false,
4407         .ret_type       = RET_INTEGER,
4408         .arg1_type      = ARG_ANYTHING,
4409         .arg2_type      = ARG_ANYTHING,
4410 };
4411
4412 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
4413            u64, flags)
4414 {
4415         return map->ops->map_redirect(map, ifindex, flags);
4416 }
4417
4418 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
4419         .func           = bpf_xdp_redirect_map,
4420         .gpl_only       = false,
4421         .ret_type       = RET_INTEGER,
4422         .arg1_type      = ARG_CONST_MAP_PTR,
4423         .arg2_type      = ARG_ANYTHING,
4424         .arg3_type      = ARG_ANYTHING,
4425 };
4426
4427 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
4428                                   unsigned long off, unsigned long len)
4429 {
4430         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
4431
4432         if (unlikely(!ptr))
4433                 return len;
4434         if (ptr != dst_buff)
4435                 memcpy(dst_buff, ptr, len);
4436
4437         return 0;
4438 }
4439
4440 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
4441            u64, flags, void *, meta, u64, meta_size)
4442 {
4443         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4444
4445         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4446                 return -EINVAL;
4447         if (unlikely(!skb || skb_size > skb->len))
4448                 return -EFAULT;
4449
4450         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
4451                                 bpf_skb_copy);
4452 }
4453
4454 static const struct bpf_func_proto bpf_skb_event_output_proto = {
4455         .func           = bpf_skb_event_output,
4456         .gpl_only       = true,
4457         .ret_type       = RET_INTEGER,
4458         .arg1_type      = ARG_PTR_TO_CTX,
4459         .arg2_type      = ARG_CONST_MAP_PTR,
4460         .arg3_type      = ARG_ANYTHING,
4461         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4462         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4463 };
4464
4465 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff)
4466
4467 const struct bpf_func_proto bpf_skb_output_proto = {
4468         .func           = bpf_skb_event_output,
4469         .gpl_only       = true,
4470         .ret_type       = RET_INTEGER,
4471         .arg1_type      = ARG_PTR_TO_BTF_ID,
4472         .arg1_btf_id    = &bpf_skb_output_btf_ids[0],
4473         .arg2_type      = ARG_CONST_MAP_PTR,
4474         .arg3_type      = ARG_ANYTHING,
4475         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4476         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4477 };
4478
4479 static unsigned short bpf_tunnel_key_af(u64 flags)
4480 {
4481         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
4482 }
4483
4484 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
4485            u32, size, u64, flags)
4486 {
4487         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4488         u8 compat[sizeof(struct bpf_tunnel_key)];
4489         void *to_orig = to;
4490         int err;
4491
4492         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 |
4493                                          BPF_F_TUNINFO_FLAGS)))) {
4494                 err = -EINVAL;
4495                 goto err_clear;
4496         }
4497         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
4498                 err = -EPROTO;
4499                 goto err_clear;
4500         }
4501         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4502                 err = -EINVAL;
4503                 switch (size) {
4504                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4505                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4506                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4507                         goto set_compat;
4508                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4509                         /* Fixup deprecated structure layouts here, so we have
4510                          * a common path later on.
4511                          */
4512                         if (ip_tunnel_info_af(info) != AF_INET)
4513                                 goto err_clear;
4514 set_compat:
4515                         to = (struct bpf_tunnel_key *)compat;
4516                         break;
4517                 default:
4518                         goto err_clear;
4519                 }
4520         }
4521
4522         to->tunnel_id = be64_to_cpu(info->key.tun_id);
4523         to->tunnel_tos = info->key.tos;
4524         to->tunnel_ttl = info->key.ttl;
4525         if (flags & BPF_F_TUNINFO_FLAGS)
4526                 to->tunnel_flags = info->key.tun_flags;
4527         else
4528                 to->tunnel_ext = 0;
4529
4530         if (flags & BPF_F_TUNINFO_IPV6) {
4531                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
4532                        sizeof(to->remote_ipv6));
4533                 memcpy(to->local_ipv6, &info->key.u.ipv6.dst,
4534                        sizeof(to->local_ipv6));
4535                 to->tunnel_label = be32_to_cpu(info->key.label);
4536         } else {
4537                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
4538                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4539                 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst);
4540                 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3);
4541                 to->tunnel_label = 0;
4542         }
4543
4544         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
4545                 memcpy(to_orig, to, size);
4546
4547         return 0;
4548 err_clear:
4549         memset(to_orig, 0, size);
4550         return err;
4551 }
4552
4553 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
4554         .func           = bpf_skb_get_tunnel_key,
4555         .gpl_only       = false,
4556         .ret_type       = RET_INTEGER,
4557         .arg1_type      = ARG_PTR_TO_CTX,
4558         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4559         .arg3_type      = ARG_CONST_SIZE,
4560         .arg4_type      = ARG_ANYTHING,
4561 };
4562
4563 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
4564 {
4565         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
4566         int err;
4567
4568         if (unlikely(!info ||
4569                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
4570                 err = -ENOENT;
4571                 goto err_clear;
4572         }
4573         if (unlikely(size < info->options_len)) {
4574                 err = -ENOMEM;
4575                 goto err_clear;
4576         }
4577
4578         ip_tunnel_info_opts_get(to, info);
4579         if (size > info->options_len)
4580                 memset(to + info->options_len, 0, size - info->options_len);
4581
4582         return info->options_len;
4583 err_clear:
4584         memset(to, 0, size);
4585         return err;
4586 }
4587
4588 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
4589         .func           = bpf_skb_get_tunnel_opt,
4590         .gpl_only       = false,
4591         .ret_type       = RET_INTEGER,
4592         .arg1_type      = ARG_PTR_TO_CTX,
4593         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
4594         .arg3_type      = ARG_CONST_SIZE,
4595 };
4596
4597 static struct metadata_dst __percpu *md_dst;
4598
4599 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
4600            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
4601 {
4602         struct metadata_dst *md = this_cpu_ptr(md_dst);
4603         u8 compat[sizeof(struct bpf_tunnel_key)];
4604         struct ip_tunnel_info *info;
4605
4606         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
4607                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
4608                 return -EINVAL;
4609         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
4610                 switch (size) {
4611                 case offsetof(struct bpf_tunnel_key, local_ipv6[0]):
4612                 case offsetof(struct bpf_tunnel_key, tunnel_label):
4613                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
4614                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
4615                         /* Fixup deprecated structure layouts here, so we have
4616                          * a common path later on.
4617                          */
4618                         memcpy(compat, from, size);
4619                         memset(compat + size, 0, sizeof(compat) - size);
4620                         from = (const struct bpf_tunnel_key *) compat;
4621                         break;
4622                 default:
4623                         return -EINVAL;
4624                 }
4625         }
4626         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
4627                      from->tunnel_ext))
4628                 return -EINVAL;
4629
4630         skb_dst_drop(skb);
4631         dst_hold((struct dst_entry *) md);
4632         skb_dst_set(skb, (struct dst_entry *) md);
4633
4634         info = &md->u.tun_info;
4635         memset(info, 0, sizeof(*info));
4636         info->mode = IP_TUNNEL_INFO_TX;
4637
4638         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
4639         if (flags & BPF_F_DONT_FRAGMENT)
4640                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
4641         if (flags & BPF_F_ZERO_CSUM_TX)
4642                 info->key.tun_flags &= ~TUNNEL_CSUM;
4643         if (flags & BPF_F_SEQ_NUMBER)
4644                 info->key.tun_flags |= TUNNEL_SEQ;
4645
4646         info->key.tun_id = cpu_to_be64(from->tunnel_id);
4647         info->key.tos = from->tunnel_tos;
4648         info->key.ttl = from->tunnel_ttl;
4649
4650         if (flags & BPF_F_TUNINFO_IPV6) {
4651                 info->mode |= IP_TUNNEL_INFO_IPV6;
4652                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
4653                        sizeof(from->remote_ipv6));
4654                 memcpy(&info->key.u.ipv6.src, from->local_ipv6,
4655                        sizeof(from->local_ipv6));
4656                 info->key.label = cpu_to_be32(from->tunnel_label) &
4657                                   IPV6_FLOWLABEL_MASK;
4658         } else {
4659                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
4660                 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4);
4661                 info->key.flow_flags = FLOWI_FLAG_ANYSRC;
4662         }
4663
4664         return 0;
4665 }
4666
4667 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
4668         .func           = bpf_skb_set_tunnel_key,
4669         .gpl_only       = false,
4670         .ret_type       = RET_INTEGER,
4671         .arg1_type      = ARG_PTR_TO_CTX,
4672         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4673         .arg3_type      = ARG_CONST_SIZE,
4674         .arg4_type      = ARG_ANYTHING,
4675 };
4676
4677 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
4678            const u8 *, from, u32, size)
4679 {
4680         struct ip_tunnel_info *info = skb_tunnel_info(skb);
4681         const struct metadata_dst *md = this_cpu_ptr(md_dst);
4682
4683         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
4684                 return -EINVAL;
4685         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
4686                 return -ENOMEM;
4687
4688         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
4689
4690         return 0;
4691 }
4692
4693 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
4694         .func           = bpf_skb_set_tunnel_opt,
4695         .gpl_only       = false,
4696         .ret_type       = RET_INTEGER,
4697         .arg1_type      = ARG_PTR_TO_CTX,
4698         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4699         .arg3_type      = ARG_CONST_SIZE,
4700 };
4701
4702 static const struct bpf_func_proto *
4703 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
4704 {
4705         if (!md_dst) {
4706                 struct metadata_dst __percpu *tmp;
4707
4708                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
4709                                                 METADATA_IP_TUNNEL,
4710                                                 GFP_KERNEL);
4711                 if (!tmp)
4712                         return NULL;
4713                 if (cmpxchg(&md_dst, NULL, tmp))
4714                         metadata_dst_free_percpu(tmp);
4715         }
4716
4717         switch (which) {
4718         case BPF_FUNC_skb_set_tunnel_key:
4719                 return &bpf_skb_set_tunnel_key_proto;
4720         case BPF_FUNC_skb_set_tunnel_opt:
4721                 return &bpf_skb_set_tunnel_opt_proto;
4722         default:
4723                 return NULL;
4724         }
4725 }
4726
4727 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
4728            u32, idx)
4729 {
4730         struct bpf_array *array = container_of(map, struct bpf_array, map);
4731         struct cgroup *cgrp;
4732         struct sock *sk;
4733
4734         sk = skb_to_full_sk(skb);
4735         if (!sk || !sk_fullsock(sk))
4736                 return -ENOENT;
4737         if (unlikely(idx >= array->map.max_entries))
4738                 return -E2BIG;
4739
4740         cgrp = READ_ONCE(array->ptrs[idx]);
4741         if (unlikely(!cgrp))
4742                 return -EAGAIN;
4743
4744         return sk_under_cgroup_hierarchy(sk, cgrp);
4745 }
4746
4747 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
4748         .func           = bpf_skb_under_cgroup,
4749         .gpl_only       = false,
4750         .ret_type       = RET_INTEGER,
4751         .arg1_type      = ARG_PTR_TO_CTX,
4752         .arg2_type      = ARG_CONST_MAP_PTR,
4753         .arg3_type      = ARG_ANYTHING,
4754 };
4755
4756 #ifdef CONFIG_SOCK_CGROUP_DATA
4757 static inline u64 __bpf_sk_cgroup_id(struct sock *sk)
4758 {
4759         struct cgroup *cgrp;
4760
4761         sk = sk_to_full_sk(sk);
4762         if (!sk || !sk_fullsock(sk))
4763                 return 0;
4764
4765         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4766         return cgroup_id(cgrp);
4767 }
4768
4769 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
4770 {
4771         return __bpf_sk_cgroup_id(skb->sk);
4772 }
4773
4774 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
4775         .func           = bpf_skb_cgroup_id,
4776         .gpl_only       = false,
4777         .ret_type       = RET_INTEGER,
4778         .arg1_type      = ARG_PTR_TO_CTX,
4779 };
4780
4781 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk,
4782                                               int ancestor_level)
4783 {
4784         struct cgroup *ancestor;
4785         struct cgroup *cgrp;
4786
4787         sk = sk_to_full_sk(sk);
4788         if (!sk || !sk_fullsock(sk))
4789                 return 0;
4790
4791         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
4792         ancestor = cgroup_ancestor(cgrp, ancestor_level);
4793         if (!ancestor)
4794                 return 0;
4795
4796         return cgroup_id(ancestor);
4797 }
4798
4799 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
4800            ancestor_level)
4801 {
4802         return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level);
4803 }
4804
4805 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
4806         .func           = bpf_skb_ancestor_cgroup_id,
4807         .gpl_only       = false,
4808         .ret_type       = RET_INTEGER,
4809         .arg1_type      = ARG_PTR_TO_CTX,
4810         .arg2_type      = ARG_ANYTHING,
4811 };
4812
4813 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk)
4814 {
4815         return __bpf_sk_cgroup_id(sk);
4816 }
4817
4818 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = {
4819         .func           = bpf_sk_cgroup_id,
4820         .gpl_only       = false,
4821         .ret_type       = RET_INTEGER,
4822         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4823 };
4824
4825 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level)
4826 {
4827         return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level);
4828 }
4829
4830 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = {
4831         .func           = bpf_sk_ancestor_cgroup_id,
4832         .gpl_only       = false,
4833         .ret_type       = RET_INTEGER,
4834         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4835         .arg2_type      = ARG_ANYTHING,
4836 };
4837 #endif
4838
4839 static unsigned long bpf_xdp_copy(void *dst, const void *ctx,
4840                                   unsigned long off, unsigned long len)
4841 {
4842         struct xdp_buff *xdp = (struct xdp_buff *)ctx;
4843
4844         bpf_xdp_copy_buf(xdp, off, dst, len, false);
4845         return 0;
4846 }
4847
4848 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
4849            u64, flags, void *, meta, u64, meta_size)
4850 {
4851         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
4852
4853         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
4854                 return -EINVAL;
4855
4856         if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp)))
4857                 return -EFAULT;
4858
4859         return bpf_event_output(map, flags, meta, meta_size, xdp,
4860                                 xdp_size, bpf_xdp_copy);
4861 }
4862
4863 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
4864         .func           = bpf_xdp_event_output,
4865         .gpl_only       = true,
4866         .ret_type       = RET_INTEGER,
4867         .arg1_type      = ARG_PTR_TO_CTX,
4868         .arg2_type      = ARG_CONST_MAP_PTR,
4869         .arg3_type      = ARG_ANYTHING,
4870         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4871         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4872 };
4873
4874 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff)
4875
4876 const struct bpf_func_proto bpf_xdp_output_proto = {
4877         .func           = bpf_xdp_event_output,
4878         .gpl_only       = true,
4879         .ret_type       = RET_INTEGER,
4880         .arg1_type      = ARG_PTR_TO_BTF_ID,
4881         .arg1_btf_id    = &bpf_xdp_output_btf_ids[0],
4882         .arg2_type      = ARG_CONST_MAP_PTR,
4883         .arg3_type      = ARG_ANYTHING,
4884         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
4885         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
4886 };
4887
4888 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
4889 {
4890         return skb->sk ? __sock_gen_cookie(skb->sk) : 0;
4891 }
4892
4893 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
4894         .func           = bpf_get_socket_cookie,
4895         .gpl_only       = false,
4896         .ret_type       = RET_INTEGER,
4897         .arg1_type      = ARG_PTR_TO_CTX,
4898 };
4899
4900 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4901 {
4902         return __sock_gen_cookie(ctx->sk);
4903 }
4904
4905 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
4906         .func           = bpf_get_socket_cookie_sock_addr,
4907         .gpl_only       = false,
4908         .ret_type       = RET_INTEGER,
4909         .arg1_type      = ARG_PTR_TO_CTX,
4910 };
4911
4912 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx)
4913 {
4914         return __sock_gen_cookie(ctx);
4915 }
4916
4917 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = {
4918         .func           = bpf_get_socket_cookie_sock,
4919         .gpl_only       = false,
4920         .ret_type       = RET_INTEGER,
4921         .arg1_type      = ARG_PTR_TO_CTX,
4922 };
4923
4924 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk)
4925 {
4926         return sk ? sock_gen_cookie(sk) : 0;
4927 }
4928
4929 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = {
4930         .func           = bpf_get_socket_ptr_cookie,
4931         .gpl_only       = false,
4932         .ret_type       = RET_INTEGER,
4933         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
4934 };
4935
4936 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4937 {
4938         return __sock_gen_cookie(ctx->sk);
4939 }
4940
4941 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
4942         .func           = bpf_get_socket_cookie_sock_ops,
4943         .gpl_only       = false,
4944         .ret_type       = RET_INTEGER,
4945         .arg1_type      = ARG_PTR_TO_CTX,
4946 };
4947
4948 static u64 __bpf_get_netns_cookie(struct sock *sk)
4949 {
4950         const struct net *net = sk ? sock_net(sk) : &init_net;
4951
4952         return net->net_cookie;
4953 }
4954
4955 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx)
4956 {
4957         return __bpf_get_netns_cookie(ctx);
4958 }
4959
4960 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = {
4961         .func           = bpf_get_netns_cookie_sock,
4962         .gpl_only       = false,
4963         .ret_type       = RET_INTEGER,
4964         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4965 };
4966
4967 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
4968 {
4969         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4970 }
4971
4972 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = {
4973         .func           = bpf_get_netns_cookie_sock_addr,
4974         .gpl_only       = false,
4975         .ret_type       = RET_INTEGER,
4976         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4977 };
4978
4979 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
4980 {
4981         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4982 }
4983
4984 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = {
4985         .func           = bpf_get_netns_cookie_sock_ops,
4986         .gpl_only       = false,
4987         .ret_type       = RET_INTEGER,
4988         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
4989 };
4990
4991 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx)
4992 {
4993         return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL);
4994 }
4995
4996 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = {
4997         .func           = bpf_get_netns_cookie_sk_msg,
4998         .gpl_only       = false,
4999         .ret_type       = RET_INTEGER,
5000         .arg1_type      = ARG_PTR_TO_CTX_OR_NULL,
5001 };
5002
5003 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
5004 {
5005         struct sock *sk = sk_to_full_sk(skb->sk);
5006         kuid_t kuid;
5007
5008         if (!sk || !sk_fullsock(sk))
5009                 return overflowuid;
5010         kuid = sock_net_uid(sock_net(sk), sk);
5011         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
5012 }
5013
5014 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
5015         .func           = bpf_get_socket_uid,
5016         .gpl_only       = false,
5017         .ret_type       = RET_INTEGER,
5018         .arg1_type      = ARG_PTR_TO_CTX,
5019 };
5020
5021 static int sol_socket_sockopt(struct sock *sk, int optname,
5022                               char *optval, int *optlen,
5023                               bool getopt)
5024 {
5025         switch (optname) {
5026         case SO_REUSEADDR:
5027         case SO_SNDBUF:
5028         case SO_RCVBUF:
5029         case SO_KEEPALIVE:
5030         case SO_PRIORITY:
5031         case SO_REUSEPORT:
5032         case SO_RCVLOWAT:
5033         case SO_MARK:
5034         case SO_MAX_PACING_RATE:
5035         case SO_BINDTOIFINDEX:
5036         case SO_TXREHASH:
5037                 if (*optlen != sizeof(int))
5038                         return -EINVAL;
5039                 break;
5040         case SO_BINDTODEVICE:
5041                 break;
5042         default:
5043                 return -EINVAL;
5044         }
5045
5046         if (getopt) {
5047                 if (optname == SO_BINDTODEVICE)
5048                         return -EINVAL;
5049                 return sk_getsockopt(sk, SOL_SOCKET, optname,
5050                                      KERNEL_SOCKPTR(optval),
5051                                      KERNEL_SOCKPTR(optlen));
5052         }
5053
5054         return sk_setsockopt(sk, SOL_SOCKET, optname,
5055                              KERNEL_SOCKPTR(optval), *optlen);
5056 }
5057
5058 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname,
5059                                   char *optval, int optlen)
5060 {
5061         struct tcp_sock *tp = tcp_sk(sk);
5062         unsigned long timeout;
5063         int val;
5064
5065         if (optlen != sizeof(int))
5066                 return -EINVAL;
5067
5068         val = *(int *)optval;
5069
5070         /* Only some options are supported */
5071         switch (optname) {
5072         case TCP_BPF_IW:
5073                 if (val <= 0 || tp->data_segs_out > tp->syn_data)
5074                         return -EINVAL;
5075                 tcp_snd_cwnd_set(tp, val);
5076                 break;
5077         case TCP_BPF_SNDCWND_CLAMP:
5078                 if (val <= 0)
5079                         return -EINVAL;
5080                 tp->snd_cwnd_clamp = val;
5081                 tp->snd_ssthresh = val;
5082                 break;
5083         case TCP_BPF_DELACK_MAX:
5084                 timeout = usecs_to_jiffies(val);
5085                 if (timeout > TCP_DELACK_MAX ||
5086                     timeout < TCP_TIMEOUT_MIN)
5087                         return -EINVAL;
5088                 inet_csk(sk)->icsk_delack_max = timeout;
5089                 break;
5090         case TCP_BPF_RTO_MIN:
5091                 timeout = usecs_to_jiffies(val);
5092                 if (timeout > TCP_RTO_MIN ||
5093                     timeout < TCP_TIMEOUT_MIN)
5094                         return -EINVAL;
5095                 inet_csk(sk)->icsk_rto_min = timeout;
5096                 break;
5097         default:
5098                 return -EINVAL;
5099         }
5100
5101         return 0;
5102 }
5103
5104 static int sol_tcp_sockopt(struct sock *sk, int optname,
5105                            char *optval, int *optlen,
5106                            bool getopt)
5107 {
5108         if (sk->sk_prot->setsockopt != tcp_setsockopt)
5109                 return -EINVAL;
5110
5111         switch (optname) {
5112         case TCP_NODELAY:
5113         case TCP_MAXSEG:
5114         case TCP_KEEPIDLE:
5115         case TCP_KEEPINTVL:
5116         case TCP_KEEPCNT:
5117         case TCP_SYNCNT:
5118         case TCP_WINDOW_CLAMP:
5119         case TCP_THIN_LINEAR_TIMEOUTS:
5120         case TCP_USER_TIMEOUT:
5121         case TCP_NOTSENT_LOWAT:
5122         case TCP_SAVE_SYN:
5123                 if (*optlen != sizeof(int))
5124                         return -EINVAL;
5125                 break;
5126         case TCP_CONGESTION:
5127                 if (*optlen < 2)
5128                         return -EINVAL;
5129                 break;
5130         case TCP_SAVED_SYN:
5131                 if (*optlen < 1)
5132                         return -EINVAL;
5133                 break;
5134         default:
5135                 if (getopt)
5136                         return -EINVAL;
5137                 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen);
5138         }
5139
5140         if (getopt) {
5141                 if (optname == TCP_SAVED_SYN) {
5142                         struct tcp_sock *tp = tcp_sk(sk);
5143
5144                         if (!tp->saved_syn ||
5145                             *optlen > tcp_saved_syn_len(tp->saved_syn))
5146                                 return -EINVAL;
5147                         memcpy(optval, tp->saved_syn->data, *optlen);
5148                         /* It cannot free tp->saved_syn here because it
5149                          * does not know if the user space still needs it.
5150                          */
5151                         return 0;
5152                 }
5153
5154                 if (optname == TCP_CONGESTION) {
5155                         if (!inet_csk(sk)->icsk_ca_ops)
5156                                 return -EINVAL;
5157                         /* BPF expects NULL-terminated tcp-cc string */
5158                         optval[--(*optlen)] = '\0';
5159                 }
5160
5161                 return do_tcp_getsockopt(sk, SOL_TCP, optname,
5162                                          KERNEL_SOCKPTR(optval),
5163                                          KERNEL_SOCKPTR(optlen));
5164         }
5165
5166         return do_tcp_setsockopt(sk, SOL_TCP, optname,
5167                                  KERNEL_SOCKPTR(optval), *optlen);
5168 }
5169
5170 static int sol_ip_sockopt(struct sock *sk, int optname,
5171                           char *optval, int *optlen,
5172                           bool getopt)
5173 {
5174         if (sk->sk_family != AF_INET)
5175                 return -EINVAL;
5176
5177         switch (optname) {
5178         case IP_TOS:
5179                 if (*optlen != sizeof(int))
5180                         return -EINVAL;
5181                 break;
5182         default:
5183                 return -EINVAL;
5184         }
5185
5186         if (getopt)
5187                 return do_ip_getsockopt(sk, SOL_IP, optname,
5188                                         KERNEL_SOCKPTR(optval),
5189                                         KERNEL_SOCKPTR(optlen));
5190
5191         return do_ip_setsockopt(sk, SOL_IP, optname,
5192                                 KERNEL_SOCKPTR(optval), *optlen);
5193 }
5194
5195 static int sol_ipv6_sockopt(struct sock *sk, int optname,
5196                             char *optval, int *optlen,
5197                             bool getopt)
5198 {
5199         if (sk->sk_family != AF_INET6)
5200                 return -EINVAL;
5201
5202         switch (optname) {
5203         case IPV6_TCLASS:
5204         case IPV6_AUTOFLOWLABEL:
5205                 if (*optlen != sizeof(int))
5206                         return -EINVAL;
5207                 break;
5208         default:
5209                 return -EINVAL;
5210         }
5211
5212         if (getopt)
5213                 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname,
5214                                                       KERNEL_SOCKPTR(optval),
5215                                                       KERNEL_SOCKPTR(optlen));
5216
5217         return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname,
5218                                               KERNEL_SOCKPTR(optval), *optlen);
5219 }
5220
5221 static int __bpf_setsockopt(struct sock *sk, int level, int optname,
5222                             char *optval, int optlen)
5223 {
5224         if (!sk_fullsock(sk))
5225                 return -EINVAL;
5226
5227         if (level == SOL_SOCKET)
5228                 return sol_socket_sockopt(sk, optname, optval, &optlen, false);
5229         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5230                 return sol_ip_sockopt(sk, optname, optval, &optlen, false);
5231         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5232                 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false);
5233         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5234                 return sol_tcp_sockopt(sk, optname, optval, &optlen, false);
5235
5236         return -EINVAL;
5237 }
5238
5239 static int _bpf_setsockopt(struct sock *sk, int level, int optname,
5240                            char *optval, int optlen)
5241 {
5242         if (sk_fullsock(sk))
5243                 sock_owned_by_me(sk);
5244         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5245 }
5246
5247 static int __bpf_getsockopt(struct sock *sk, int level, int optname,
5248                             char *optval, int optlen)
5249 {
5250         int err, saved_optlen = optlen;
5251
5252         if (!sk_fullsock(sk)) {
5253                 err = -EINVAL;
5254                 goto done;
5255         }
5256
5257         if (level == SOL_SOCKET)
5258                 err = sol_socket_sockopt(sk, optname, optval, &optlen, true);
5259         else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP)
5260                 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true);
5261         else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP)
5262                 err = sol_ip_sockopt(sk, optname, optval, &optlen, true);
5263         else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6)
5264                 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true);
5265         else
5266                 err = -EINVAL;
5267
5268 done:
5269         if (err)
5270                 optlen = 0;
5271         if (optlen < saved_optlen)
5272                 memset(optval + optlen, 0, saved_optlen - optlen);
5273         return err;
5274 }
5275
5276 static int _bpf_getsockopt(struct sock *sk, int level, int optname,
5277                            char *optval, int optlen)
5278 {
5279         if (sk_fullsock(sk))
5280                 sock_owned_by_me(sk);
5281         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5282 }
5283
5284 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level,
5285            int, optname, char *, optval, int, optlen)
5286 {
5287         if (level == SOL_TCP && optname == TCP_CONGESTION) {
5288                 if (optlen >= sizeof("cdg") - 1 &&
5289                     !strncmp("cdg", optval, optlen))
5290                         return -ENOTSUPP;
5291         }
5292
5293         return _bpf_setsockopt(sk, level, optname, optval, optlen);
5294 }
5295
5296 const struct bpf_func_proto bpf_sk_setsockopt_proto = {
5297         .func           = bpf_sk_setsockopt,
5298         .gpl_only       = false,
5299         .ret_type       = RET_INTEGER,
5300         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5301         .arg2_type      = ARG_ANYTHING,
5302         .arg3_type      = ARG_ANYTHING,
5303         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5304         .arg5_type      = ARG_CONST_SIZE,
5305 };
5306
5307 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level,
5308            int, optname, char *, optval, int, optlen)
5309 {
5310         return _bpf_getsockopt(sk, level, optname, optval, optlen);
5311 }
5312
5313 const struct bpf_func_proto bpf_sk_getsockopt_proto = {
5314         .func           = bpf_sk_getsockopt,
5315         .gpl_only       = false,
5316         .ret_type       = RET_INTEGER,
5317         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5318         .arg2_type      = ARG_ANYTHING,
5319         .arg3_type      = ARG_ANYTHING,
5320         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5321         .arg5_type      = ARG_CONST_SIZE,
5322 };
5323
5324 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level,
5325            int, optname, char *, optval, int, optlen)
5326 {
5327         return __bpf_setsockopt(sk, level, optname, optval, optlen);
5328 }
5329
5330 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = {
5331         .func           = bpf_unlocked_sk_setsockopt,
5332         .gpl_only       = false,
5333         .ret_type       = RET_INTEGER,
5334         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5335         .arg2_type      = ARG_ANYTHING,
5336         .arg3_type      = ARG_ANYTHING,
5337         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5338         .arg5_type      = ARG_CONST_SIZE,
5339 };
5340
5341 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level,
5342            int, optname, char *, optval, int, optlen)
5343 {
5344         return __bpf_getsockopt(sk, level, optname, optval, optlen);
5345 }
5346
5347 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = {
5348         .func           = bpf_unlocked_sk_getsockopt,
5349         .gpl_only       = false,
5350         .ret_type       = RET_INTEGER,
5351         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
5352         .arg2_type      = ARG_ANYTHING,
5353         .arg3_type      = ARG_ANYTHING,
5354         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5355         .arg5_type      = ARG_CONST_SIZE,
5356 };
5357
5358 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx,
5359            int, level, int, optname, char *, optval, int, optlen)
5360 {
5361         return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen);
5362 }
5363
5364 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = {
5365         .func           = bpf_sock_addr_setsockopt,
5366         .gpl_only       = false,
5367         .ret_type       = RET_INTEGER,
5368         .arg1_type      = ARG_PTR_TO_CTX,
5369         .arg2_type      = ARG_ANYTHING,
5370         .arg3_type      = ARG_ANYTHING,
5371         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5372         .arg5_type      = ARG_CONST_SIZE,
5373 };
5374
5375 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx,
5376            int, level, int, optname, char *, optval, int, optlen)
5377 {
5378         return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen);
5379 }
5380
5381 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = {
5382         .func           = bpf_sock_addr_getsockopt,
5383         .gpl_only       = false,
5384         .ret_type       = RET_INTEGER,
5385         .arg1_type      = ARG_PTR_TO_CTX,
5386         .arg2_type      = ARG_ANYTHING,
5387         .arg3_type      = ARG_ANYTHING,
5388         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5389         .arg5_type      = ARG_CONST_SIZE,
5390 };
5391
5392 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5393            int, level, int, optname, char *, optval, int, optlen)
5394 {
5395         return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen);
5396 }
5397
5398 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = {
5399         .func           = bpf_sock_ops_setsockopt,
5400         .gpl_only       = false,
5401         .ret_type       = RET_INTEGER,
5402         .arg1_type      = ARG_PTR_TO_CTX,
5403         .arg2_type      = ARG_ANYTHING,
5404         .arg3_type      = ARG_ANYTHING,
5405         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5406         .arg5_type      = ARG_CONST_SIZE,
5407 };
5408
5409 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock,
5410                                 int optname, const u8 **start)
5411 {
5412         struct sk_buff *syn_skb = bpf_sock->syn_skb;
5413         const u8 *hdr_start;
5414         int ret;
5415
5416         if (syn_skb) {
5417                 /* sk is a request_sock here */
5418
5419                 if (optname == TCP_BPF_SYN) {
5420                         hdr_start = syn_skb->data;
5421                         ret = tcp_hdrlen(syn_skb);
5422                 } else if (optname == TCP_BPF_SYN_IP) {
5423                         hdr_start = skb_network_header(syn_skb);
5424                         ret = skb_network_header_len(syn_skb) +
5425                                 tcp_hdrlen(syn_skb);
5426                 } else {
5427                         /* optname == TCP_BPF_SYN_MAC */
5428                         hdr_start = skb_mac_header(syn_skb);
5429                         ret = skb_mac_header_len(syn_skb) +
5430                                 skb_network_header_len(syn_skb) +
5431                                 tcp_hdrlen(syn_skb);
5432                 }
5433         } else {
5434                 struct sock *sk = bpf_sock->sk;
5435                 struct saved_syn *saved_syn;
5436
5437                 if (sk->sk_state == TCP_NEW_SYN_RECV)
5438                         /* synack retransmit. bpf_sock->syn_skb will
5439                          * not be available.  It has to resort to
5440                          * saved_syn (if it is saved).
5441                          */
5442                         saved_syn = inet_reqsk(sk)->saved_syn;
5443                 else
5444                         saved_syn = tcp_sk(sk)->saved_syn;
5445
5446                 if (!saved_syn)
5447                         return -ENOENT;
5448
5449                 if (optname == TCP_BPF_SYN) {
5450                         hdr_start = saved_syn->data +
5451                                 saved_syn->mac_hdrlen +
5452                                 saved_syn->network_hdrlen;
5453                         ret = saved_syn->tcp_hdrlen;
5454                 } else if (optname == TCP_BPF_SYN_IP) {
5455                         hdr_start = saved_syn->data +
5456                                 saved_syn->mac_hdrlen;
5457                         ret = saved_syn->network_hdrlen +
5458                                 saved_syn->tcp_hdrlen;
5459                 } else {
5460                         /* optname == TCP_BPF_SYN_MAC */
5461
5462                         /* TCP_SAVE_SYN may not have saved the mac hdr */
5463                         if (!saved_syn->mac_hdrlen)
5464                                 return -ENOENT;
5465
5466                         hdr_start = saved_syn->data;
5467                         ret = saved_syn->mac_hdrlen +
5468                                 saved_syn->network_hdrlen +
5469                                 saved_syn->tcp_hdrlen;
5470                 }
5471         }
5472
5473         *start = hdr_start;
5474         return ret;
5475 }
5476
5477 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
5478            int, level, int, optname, char *, optval, int, optlen)
5479 {
5480         if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP &&
5481             optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) {
5482                 int ret, copy_len = 0;
5483                 const u8 *start;
5484
5485                 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start);
5486                 if (ret > 0) {
5487                         copy_len = ret;
5488                         if (optlen < copy_len) {
5489                                 copy_len = optlen;
5490                                 ret = -ENOSPC;
5491                         }
5492
5493                         memcpy(optval, start, copy_len);
5494                 }
5495
5496                 /* Zero out unused buffer at the end */
5497                 memset(optval + copy_len, 0, optlen - copy_len);
5498
5499                 return ret;
5500         }
5501
5502         return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen);
5503 }
5504
5505 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = {
5506         .func           = bpf_sock_ops_getsockopt,
5507         .gpl_only       = false,
5508         .ret_type       = RET_INTEGER,
5509         .arg1_type      = ARG_PTR_TO_CTX,
5510         .arg2_type      = ARG_ANYTHING,
5511         .arg3_type      = ARG_ANYTHING,
5512         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
5513         .arg5_type      = ARG_CONST_SIZE,
5514 };
5515
5516 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
5517            int, argval)
5518 {
5519         struct sock *sk = bpf_sock->sk;
5520         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
5521
5522         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
5523                 return -EINVAL;
5524
5525         tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
5526
5527         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
5528 }
5529
5530 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
5531         .func           = bpf_sock_ops_cb_flags_set,
5532         .gpl_only       = false,
5533         .ret_type       = RET_INTEGER,
5534         .arg1_type      = ARG_PTR_TO_CTX,
5535         .arg2_type      = ARG_ANYTHING,
5536 };
5537
5538 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
5539 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
5540
5541 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
5542            int, addr_len)
5543 {
5544 #ifdef CONFIG_INET
5545         struct sock *sk = ctx->sk;
5546         u32 flags = BIND_FROM_BPF;
5547         int err;
5548
5549         err = -EINVAL;
5550         if (addr_len < offsetofend(struct sockaddr, sa_family))
5551                 return err;
5552         if (addr->sa_family == AF_INET) {
5553                 if (addr_len < sizeof(struct sockaddr_in))
5554                         return err;
5555                 if (((struct sockaddr_in *)addr)->sin_port == htons(0))
5556                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5557                 return __inet_bind(sk, addr, addr_len, flags);
5558 #if IS_ENABLED(CONFIG_IPV6)
5559         } else if (addr->sa_family == AF_INET6) {
5560                 if (addr_len < SIN6_LEN_RFC2133)
5561                         return err;
5562                 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0))
5563                         flags |= BIND_FORCE_ADDRESS_NO_PORT;
5564                 /* ipv6_bpf_stub cannot be NULL, since it's called from
5565                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
5566                  */
5567                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags);
5568 #endif /* CONFIG_IPV6 */
5569         }
5570 #endif /* CONFIG_INET */
5571
5572         return -EAFNOSUPPORT;
5573 }
5574
5575 static const struct bpf_func_proto bpf_bind_proto = {
5576         .func           = bpf_bind,
5577         .gpl_only       = false,
5578         .ret_type       = RET_INTEGER,
5579         .arg1_type      = ARG_PTR_TO_CTX,
5580         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
5581         .arg3_type      = ARG_CONST_SIZE,
5582 };
5583
5584 #ifdef CONFIG_XFRM
5585 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
5586            struct bpf_xfrm_state *, to, u32, size, u64, flags)
5587 {
5588         const struct sec_path *sp = skb_sec_path(skb);
5589         const struct xfrm_state *x;
5590
5591         if (!sp || unlikely(index >= sp->len || flags))
5592                 goto err_clear;
5593
5594         x = sp->xvec[index];
5595
5596         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
5597                 goto err_clear;
5598
5599         to->reqid = x->props.reqid;
5600         to->spi = x->id.spi;
5601         to->family = x->props.family;
5602         to->ext = 0;
5603
5604         if (to->family == AF_INET6) {
5605                 memcpy(to->remote_ipv6, x->props.saddr.a6,
5606                        sizeof(to->remote_ipv6));
5607         } else {
5608                 to->remote_ipv4 = x->props.saddr.a4;
5609                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
5610         }
5611
5612         return 0;
5613 err_clear:
5614         memset(to, 0, size);
5615         return -EINVAL;
5616 }
5617
5618 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
5619         .func           = bpf_skb_get_xfrm_state,
5620         .gpl_only       = false,
5621         .ret_type       = RET_INTEGER,
5622         .arg1_type      = ARG_PTR_TO_CTX,
5623         .arg2_type      = ARG_ANYTHING,
5624         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
5625         .arg4_type      = ARG_CONST_SIZE,
5626         .arg5_type      = ARG_ANYTHING,
5627 };
5628 #endif
5629
5630 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
5631 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
5632                                   const struct neighbour *neigh,
5633                                   const struct net_device *dev, u32 mtu)
5634 {
5635         memcpy(params->dmac, neigh->ha, ETH_ALEN);
5636         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
5637         params->h_vlan_TCI = 0;
5638         params->h_vlan_proto = 0;
5639         if (mtu)
5640                 params->mtu_result = mtu; /* union with tot_len */
5641
5642         return 0;
5643 }
5644 #endif
5645
5646 #if IS_ENABLED(CONFIG_INET)
5647 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5648                                u32 flags, bool check_mtu)
5649 {
5650         struct fib_nh_common *nhc;
5651         struct in_device *in_dev;
5652         struct neighbour *neigh;
5653         struct net_device *dev;
5654         struct fib_result res;
5655         struct flowi4 fl4;
5656         u32 mtu = 0;
5657         int err;
5658
5659         dev = dev_get_by_index_rcu(net, params->ifindex);
5660         if (unlikely(!dev))
5661                 return -ENODEV;
5662
5663         /* verify forwarding is enabled on this interface */
5664         in_dev = __in_dev_get_rcu(dev);
5665         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
5666                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5667
5668         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5669                 fl4.flowi4_iif = 1;
5670                 fl4.flowi4_oif = params->ifindex;
5671         } else {
5672                 fl4.flowi4_iif = params->ifindex;
5673                 fl4.flowi4_oif = 0;
5674         }
5675         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
5676         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
5677         fl4.flowi4_flags = 0;
5678
5679         fl4.flowi4_proto = params->l4_protocol;
5680         fl4.daddr = params->ipv4_dst;
5681         fl4.saddr = params->ipv4_src;
5682         fl4.fl4_sport = params->sport;
5683         fl4.fl4_dport = params->dport;
5684         fl4.flowi4_multipath_hash = 0;
5685
5686         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5687                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5688                 struct fib_table *tb;
5689
5690                 tb = fib_get_table(net, tbid);
5691                 if (unlikely(!tb))
5692                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5693
5694                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
5695         } else {
5696                 fl4.flowi4_mark = 0;
5697                 fl4.flowi4_secid = 0;
5698                 fl4.flowi4_tun_key.tun_id = 0;
5699                 fl4.flowi4_uid = sock_net_uid(net, NULL);
5700
5701                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
5702         }
5703
5704         if (err) {
5705                 /* map fib lookup errors to RTN_ type */
5706                 if (err == -EINVAL)
5707                         return BPF_FIB_LKUP_RET_BLACKHOLE;
5708                 if (err == -EHOSTUNREACH)
5709                         return BPF_FIB_LKUP_RET_UNREACHABLE;
5710                 if (err == -EACCES)
5711                         return BPF_FIB_LKUP_RET_PROHIBIT;
5712
5713                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5714         }
5715
5716         if (res.type != RTN_UNICAST)
5717                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5718
5719         if (fib_info_num_path(res.fi) > 1)
5720                 fib_select_path(net, &res, &fl4, NULL);
5721
5722         if (check_mtu) {
5723                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
5724                 if (params->tot_len > mtu) {
5725                         params->mtu_result = mtu; /* union with tot_len */
5726                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5727                 }
5728         }
5729
5730         nhc = res.nhc;
5731
5732         /* do not handle lwt encaps right now */
5733         if (nhc->nhc_lwtstate)
5734                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5735
5736         dev = nhc->nhc_dev;
5737
5738         params->rt_metric = res.fi->fib_priority;
5739         params->ifindex = dev->ifindex;
5740
5741         /* xdp and cls_bpf programs are run in RCU-bh so
5742          * rcu_read_lock_bh is not needed here
5743          */
5744         if (likely(nhc->nhc_gw_family != AF_INET6)) {
5745                 if (nhc->nhc_gw_family)
5746                         params->ipv4_dst = nhc->nhc_gw.ipv4;
5747
5748                 neigh = __ipv4_neigh_lookup_noref(dev,
5749                                                  (__force u32)params->ipv4_dst);
5750         } else {
5751                 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst;
5752
5753                 params->family = AF_INET6;
5754                 *dst = nhc->nhc_gw.ipv6;
5755                 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5756         }
5757
5758         if (!neigh)
5759                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5760
5761         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5762 }
5763 #endif
5764
5765 #if IS_ENABLED(CONFIG_IPV6)
5766 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
5767                                u32 flags, bool check_mtu)
5768 {
5769         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
5770         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
5771         struct fib6_result res = {};
5772         struct neighbour *neigh;
5773         struct net_device *dev;
5774         struct inet6_dev *idev;
5775         struct flowi6 fl6;
5776         int strict = 0;
5777         int oif, err;
5778         u32 mtu = 0;
5779
5780         /* link local addresses are never forwarded */
5781         if (rt6_need_strict(dst) || rt6_need_strict(src))
5782                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5783
5784         dev = dev_get_by_index_rcu(net, params->ifindex);
5785         if (unlikely(!dev))
5786                 return -ENODEV;
5787
5788         idev = __in6_dev_get_safely(dev);
5789         if (unlikely(!idev || !idev->cnf.forwarding))
5790                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
5791
5792         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
5793                 fl6.flowi6_iif = 1;
5794                 oif = fl6.flowi6_oif = params->ifindex;
5795         } else {
5796                 oif = fl6.flowi6_iif = params->ifindex;
5797                 fl6.flowi6_oif = 0;
5798                 strict = RT6_LOOKUP_F_HAS_SADDR;
5799         }
5800         fl6.flowlabel = params->flowinfo;
5801         fl6.flowi6_scope = 0;
5802         fl6.flowi6_flags = 0;
5803         fl6.mp_hash = 0;
5804
5805         fl6.flowi6_proto = params->l4_protocol;
5806         fl6.daddr = *dst;
5807         fl6.saddr = *src;
5808         fl6.fl6_sport = params->sport;
5809         fl6.fl6_dport = params->dport;
5810
5811         if (flags & BPF_FIB_LOOKUP_DIRECT) {
5812                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
5813                 struct fib6_table *tb;
5814
5815                 tb = ipv6_stub->fib6_get_table(net, tbid);
5816                 if (unlikely(!tb))
5817                         return BPF_FIB_LKUP_RET_NOT_FWDED;
5818
5819                 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res,
5820                                                    strict);
5821         } else {
5822                 fl6.flowi6_mark = 0;
5823                 fl6.flowi6_secid = 0;
5824                 fl6.flowi6_tun_key.tun_id = 0;
5825                 fl6.flowi6_uid = sock_net_uid(net, NULL);
5826
5827                 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict);
5828         }
5829
5830         if (unlikely(err || IS_ERR_OR_NULL(res.f6i) ||
5831                      res.f6i == net->ipv6.fib6_null_entry))
5832                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5833
5834         switch (res.fib6_type) {
5835         /* only unicast is forwarded */
5836         case RTN_UNICAST:
5837                 break;
5838         case RTN_BLACKHOLE:
5839                 return BPF_FIB_LKUP_RET_BLACKHOLE;
5840         case RTN_UNREACHABLE:
5841                 return BPF_FIB_LKUP_RET_UNREACHABLE;
5842         case RTN_PROHIBIT:
5843                 return BPF_FIB_LKUP_RET_PROHIBIT;
5844         default:
5845                 return BPF_FIB_LKUP_RET_NOT_FWDED;
5846         }
5847
5848         ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif,
5849                                     fl6.flowi6_oif != 0, NULL, strict);
5850
5851         if (check_mtu) {
5852                 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src);
5853                 if (params->tot_len > mtu) {
5854                         params->mtu_result = mtu; /* union with tot_len */
5855                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
5856                 }
5857         }
5858
5859         if (res.nh->fib_nh_lws)
5860                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
5861
5862         if (res.nh->fib_nh_gw_family)
5863                 *dst = res.nh->fib_nh_gw6;
5864
5865         dev = res.nh->fib_nh_dev;
5866         params->rt_metric = res.f6i->fib6_metric;
5867         params->ifindex = dev->ifindex;
5868
5869         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
5870          * not needed here.
5871          */
5872         neigh = __ipv6_neigh_lookup_noref_stub(dev, dst);
5873         if (!neigh)
5874                 return BPF_FIB_LKUP_RET_NO_NEIGH;
5875
5876         return bpf_fib_set_fwd_params(params, neigh, dev, mtu);
5877 }
5878 #endif
5879
5880 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
5881            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5882 {
5883         if (plen < sizeof(*params))
5884                 return -EINVAL;
5885
5886         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5887                 return -EINVAL;
5888
5889         switch (params->family) {
5890 #if IS_ENABLED(CONFIG_INET)
5891         case AF_INET:
5892                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
5893                                            flags, true);
5894 #endif
5895 #if IS_ENABLED(CONFIG_IPV6)
5896         case AF_INET6:
5897                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
5898                                            flags, true);
5899 #endif
5900         }
5901         return -EAFNOSUPPORT;
5902 }
5903
5904 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
5905         .func           = bpf_xdp_fib_lookup,
5906         .gpl_only       = true,
5907         .ret_type       = RET_INTEGER,
5908         .arg1_type      = ARG_PTR_TO_CTX,
5909         .arg2_type      = ARG_PTR_TO_MEM,
5910         .arg3_type      = ARG_CONST_SIZE,
5911         .arg4_type      = ARG_ANYTHING,
5912 };
5913
5914 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
5915            struct bpf_fib_lookup *, params, int, plen, u32, flags)
5916 {
5917         struct net *net = dev_net(skb->dev);
5918         int rc = -EAFNOSUPPORT;
5919         bool check_mtu = false;
5920
5921         if (plen < sizeof(*params))
5922                 return -EINVAL;
5923
5924         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
5925                 return -EINVAL;
5926
5927         if (params->tot_len)
5928                 check_mtu = true;
5929
5930         switch (params->family) {
5931 #if IS_ENABLED(CONFIG_INET)
5932         case AF_INET:
5933                 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu);
5934                 break;
5935 #endif
5936 #if IS_ENABLED(CONFIG_IPV6)
5937         case AF_INET6:
5938                 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu);
5939                 break;
5940 #endif
5941         }
5942
5943         if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) {
5944                 struct net_device *dev;
5945
5946                 /* When tot_len isn't provided by user, check skb
5947                  * against MTU of FIB lookup resulting net_device
5948                  */
5949                 dev = dev_get_by_index_rcu(net, params->ifindex);
5950                 if (!is_skb_forwardable(dev, skb))
5951                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
5952
5953                 params->mtu_result = dev->mtu; /* union with tot_len */
5954         }
5955
5956         return rc;
5957 }
5958
5959 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
5960         .func           = bpf_skb_fib_lookup,
5961         .gpl_only       = true,
5962         .ret_type       = RET_INTEGER,
5963         .arg1_type      = ARG_PTR_TO_CTX,
5964         .arg2_type      = ARG_PTR_TO_MEM,
5965         .arg3_type      = ARG_CONST_SIZE,
5966         .arg4_type      = ARG_ANYTHING,
5967 };
5968
5969 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr,
5970                                             u32 ifindex)
5971 {
5972         struct net *netns = dev_net(dev_curr);
5973
5974         /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */
5975         if (ifindex == 0)
5976                 return dev_curr;
5977
5978         return dev_get_by_index_rcu(netns, ifindex);
5979 }
5980
5981 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb,
5982            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
5983 {
5984         int ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
5985         struct net_device *dev = skb->dev;
5986         int skb_len, dev_len;
5987         int mtu;
5988
5989         if (unlikely(flags & ~(BPF_MTU_CHK_SEGS)))
5990                 return -EINVAL;
5991
5992         if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len)))
5993                 return -EINVAL;
5994
5995         dev = __dev_via_ifindex(dev, ifindex);
5996         if (unlikely(!dev))
5997                 return -ENODEV;
5998
5999         mtu = READ_ONCE(dev->mtu);
6000
6001         dev_len = mtu + dev->hard_header_len;
6002
6003         /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6004         skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len;
6005
6006         skb_len += len_diff; /* minus result pass check */
6007         if (skb_len <= dev_len) {
6008                 ret = BPF_MTU_CHK_RET_SUCCESS;
6009                 goto out;
6010         }
6011         /* At this point, skb->len exceed MTU, but as it include length of all
6012          * segments, it can still be below MTU.  The SKB can possibly get
6013          * re-segmented in transmit path (see validate_xmit_skb).  Thus, user
6014          * must choose if segs are to be MTU checked.
6015          */
6016         if (skb_is_gso(skb)) {
6017                 ret = BPF_MTU_CHK_RET_SUCCESS;
6018
6019                 if (flags & BPF_MTU_CHK_SEGS &&
6020                     !skb_gso_validate_network_len(skb, mtu))
6021                         ret = BPF_MTU_CHK_RET_SEGS_TOOBIG;
6022         }
6023 out:
6024         /* BPF verifier guarantees valid pointer */
6025         *mtu_len = mtu;
6026
6027         return ret;
6028 }
6029
6030 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp,
6031            u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags)
6032 {
6033         struct net_device *dev = xdp->rxq->dev;
6034         int xdp_len = xdp->data_end - xdp->data;
6035         int ret = BPF_MTU_CHK_RET_SUCCESS;
6036         int mtu, dev_len;
6037
6038         /* XDP variant doesn't support multi-buffer segment check (yet) */
6039         if (unlikely(flags))
6040                 return -EINVAL;
6041
6042         dev = __dev_via_ifindex(dev, ifindex);
6043         if (unlikely(!dev))
6044                 return -ENODEV;
6045
6046         mtu = READ_ONCE(dev->mtu);
6047
6048         /* Add L2-header as dev MTU is L3 size */
6049         dev_len = mtu + dev->hard_header_len;
6050
6051         /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */
6052         if (*mtu_len)
6053                 xdp_len = *mtu_len + dev->hard_header_len;
6054
6055         xdp_len += len_diff; /* minus result pass check */
6056         if (xdp_len > dev_len)
6057                 ret = BPF_MTU_CHK_RET_FRAG_NEEDED;
6058
6059         /* BPF verifier guarantees valid pointer */
6060         *mtu_len = mtu;
6061
6062         return ret;
6063 }
6064
6065 static const struct bpf_func_proto bpf_skb_check_mtu_proto = {
6066         .func           = bpf_skb_check_mtu,
6067         .gpl_only       = true,
6068         .ret_type       = RET_INTEGER,
6069         .arg1_type      = ARG_PTR_TO_CTX,
6070         .arg2_type      = ARG_ANYTHING,
6071         .arg3_type      = ARG_PTR_TO_INT,
6072         .arg4_type      = ARG_ANYTHING,
6073         .arg5_type      = ARG_ANYTHING,
6074 };
6075
6076 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = {
6077         .func           = bpf_xdp_check_mtu,
6078         .gpl_only       = true,
6079         .ret_type       = RET_INTEGER,
6080         .arg1_type      = ARG_PTR_TO_CTX,
6081         .arg2_type      = ARG_ANYTHING,
6082         .arg3_type      = ARG_PTR_TO_INT,
6083         .arg4_type      = ARG_ANYTHING,
6084         .arg5_type      = ARG_ANYTHING,
6085 };
6086
6087 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6088 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
6089 {
6090         int err;
6091         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
6092
6093         if (!seg6_validate_srh(srh, len, false))
6094                 return -EINVAL;
6095
6096         switch (type) {
6097         case BPF_LWT_ENCAP_SEG6_INLINE:
6098                 if (skb->protocol != htons(ETH_P_IPV6))
6099                         return -EBADMSG;
6100
6101                 err = seg6_do_srh_inline(skb, srh);
6102                 break;
6103         case BPF_LWT_ENCAP_SEG6:
6104                 skb_reset_inner_headers(skb);
6105                 skb->encapsulation = 1;
6106                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
6107                 break;
6108         default:
6109                 return -EINVAL;
6110         }
6111
6112         bpf_compute_data_pointers(skb);
6113         if (err)
6114                 return err;
6115
6116         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
6117
6118         return seg6_lookup_nexthop(skb, NULL, 0);
6119 }
6120 #endif /* CONFIG_IPV6_SEG6_BPF */
6121
6122 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6123 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len,
6124                              bool ingress)
6125 {
6126         return bpf_lwt_push_ip_encap(skb, hdr, len, ingress);
6127 }
6128 #endif
6129
6130 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
6131            u32, len)
6132 {
6133         switch (type) {
6134 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6135         case BPF_LWT_ENCAP_SEG6:
6136         case BPF_LWT_ENCAP_SEG6_INLINE:
6137                 return bpf_push_seg6_encap(skb, type, hdr, len);
6138 #endif
6139 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6140         case BPF_LWT_ENCAP_IP:
6141                 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */);
6142 #endif
6143         default:
6144                 return -EINVAL;
6145         }
6146 }
6147
6148 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type,
6149            void *, hdr, u32, len)
6150 {
6151         switch (type) {
6152 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF)
6153         case BPF_LWT_ENCAP_IP:
6154                 return bpf_push_ip_encap(skb, hdr, len, false /* egress */);
6155 #endif
6156         default:
6157                 return -EINVAL;
6158         }
6159 }
6160
6161 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = {
6162         .func           = bpf_lwt_in_push_encap,
6163         .gpl_only       = false,
6164         .ret_type       = RET_INTEGER,
6165         .arg1_type      = ARG_PTR_TO_CTX,
6166         .arg2_type      = ARG_ANYTHING,
6167         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6168         .arg4_type      = ARG_CONST_SIZE
6169 };
6170
6171 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = {
6172         .func           = bpf_lwt_xmit_push_encap,
6173         .gpl_only       = false,
6174         .ret_type       = RET_INTEGER,
6175         .arg1_type      = ARG_PTR_TO_CTX,
6176         .arg2_type      = ARG_ANYTHING,
6177         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6178         .arg4_type      = ARG_CONST_SIZE
6179 };
6180
6181 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
6182 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
6183            const void *, from, u32, len)
6184 {
6185         struct seg6_bpf_srh_state *srh_state =
6186                 this_cpu_ptr(&seg6_bpf_srh_states);
6187         struct ipv6_sr_hdr *srh = srh_state->srh;
6188         void *srh_tlvs, *srh_end, *ptr;
6189         int srhoff = 0;
6190
6191         if (srh == NULL)
6192                 return -EINVAL;
6193
6194         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
6195         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
6196
6197         ptr = skb->data + offset;
6198         if (ptr >= srh_tlvs && ptr + len <= srh_end)
6199                 srh_state->valid = false;
6200         else if (ptr < (void *)&srh->flags ||
6201                  ptr + len > (void *)&srh->segments)
6202                 return -EFAULT;
6203
6204         if (unlikely(bpf_try_make_writable(skb, offset + len)))
6205                 return -EFAULT;
6206         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6207                 return -EINVAL;
6208         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6209
6210         memcpy(skb->data + offset, from, len);
6211         return 0;
6212 }
6213
6214 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
6215         .func           = bpf_lwt_seg6_store_bytes,
6216         .gpl_only       = false,
6217         .ret_type       = RET_INTEGER,
6218         .arg1_type      = ARG_PTR_TO_CTX,
6219         .arg2_type      = ARG_ANYTHING,
6220         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6221         .arg4_type      = ARG_CONST_SIZE
6222 };
6223
6224 static void bpf_update_srh_state(struct sk_buff *skb)
6225 {
6226         struct seg6_bpf_srh_state *srh_state =
6227                 this_cpu_ptr(&seg6_bpf_srh_states);
6228         int srhoff = 0;
6229
6230         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
6231                 srh_state->srh = NULL;
6232         } else {
6233                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6234                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
6235                 srh_state->valid = true;
6236         }
6237 }
6238
6239 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
6240            u32, action, void *, param, u32, param_len)
6241 {
6242         struct seg6_bpf_srh_state *srh_state =
6243                 this_cpu_ptr(&seg6_bpf_srh_states);
6244         int hdroff = 0;
6245         int err;
6246
6247         switch (action) {
6248         case SEG6_LOCAL_ACTION_END_X:
6249                 if (!seg6_bpf_has_valid_srh(skb))
6250                         return -EBADMSG;
6251                 if (param_len != sizeof(struct in6_addr))
6252                         return -EINVAL;
6253                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
6254         case SEG6_LOCAL_ACTION_END_T:
6255                 if (!seg6_bpf_has_valid_srh(skb))
6256                         return -EBADMSG;
6257                 if (param_len != sizeof(int))
6258                         return -EINVAL;
6259                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6260         case SEG6_LOCAL_ACTION_END_DT6:
6261                 if (!seg6_bpf_has_valid_srh(skb))
6262                         return -EBADMSG;
6263                 if (param_len != sizeof(int))
6264                         return -EINVAL;
6265
6266                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
6267                         return -EBADMSG;
6268                 if (!pskb_pull(skb, hdroff))
6269                         return -EBADMSG;
6270
6271                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
6272                 skb_reset_network_header(skb);
6273                 skb_reset_transport_header(skb);
6274                 skb->encapsulation = 0;
6275
6276                 bpf_compute_data_pointers(skb);
6277                 bpf_update_srh_state(skb);
6278                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
6279         case SEG6_LOCAL_ACTION_END_B6:
6280                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6281                         return -EBADMSG;
6282                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
6283                                           param, param_len);
6284                 if (!err)
6285                         bpf_update_srh_state(skb);
6286
6287                 return err;
6288         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
6289                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
6290                         return -EBADMSG;
6291                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
6292                                           param, param_len);
6293                 if (!err)
6294                         bpf_update_srh_state(skb);
6295
6296                 return err;
6297         default:
6298                 return -EINVAL;
6299         }
6300 }
6301
6302 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
6303         .func           = bpf_lwt_seg6_action,
6304         .gpl_only       = false,
6305         .ret_type       = RET_INTEGER,
6306         .arg1_type      = ARG_PTR_TO_CTX,
6307         .arg2_type      = ARG_ANYTHING,
6308         .arg3_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6309         .arg4_type      = ARG_CONST_SIZE
6310 };
6311
6312 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
6313            s32, len)
6314 {
6315         struct seg6_bpf_srh_state *srh_state =
6316                 this_cpu_ptr(&seg6_bpf_srh_states);
6317         struct ipv6_sr_hdr *srh = srh_state->srh;
6318         void *srh_end, *srh_tlvs, *ptr;
6319         struct ipv6hdr *hdr;
6320         int srhoff = 0;
6321         int ret;
6322
6323         if (unlikely(srh == NULL))
6324                 return -EINVAL;
6325
6326         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
6327                         ((srh->first_segment + 1) << 4));
6328         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
6329                         srh_state->hdrlen);
6330         ptr = skb->data + offset;
6331
6332         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
6333                 return -EFAULT;
6334         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
6335                 return -EFAULT;
6336
6337         if (len > 0) {
6338                 ret = skb_cow_head(skb, len);
6339                 if (unlikely(ret < 0))
6340                         return ret;
6341
6342                 ret = bpf_skb_net_hdr_push(skb, offset, len);
6343         } else {
6344                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
6345         }
6346
6347         bpf_compute_data_pointers(skb);
6348         if (unlikely(ret < 0))
6349                 return ret;
6350
6351         hdr = (struct ipv6hdr *)skb->data;
6352         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
6353
6354         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
6355                 return -EINVAL;
6356         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
6357         srh_state->hdrlen += len;
6358         srh_state->valid = false;
6359         return 0;
6360 }
6361
6362 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
6363         .func           = bpf_lwt_seg6_adjust_srh,
6364         .gpl_only       = false,
6365         .ret_type       = RET_INTEGER,
6366         .arg1_type      = ARG_PTR_TO_CTX,
6367         .arg2_type      = ARG_ANYTHING,
6368         .arg3_type      = ARG_ANYTHING,
6369 };
6370 #endif /* CONFIG_IPV6_SEG6_BPF */
6371
6372 #ifdef CONFIG_INET
6373 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
6374                               int dif, int sdif, u8 family, u8 proto)
6375 {
6376         bool refcounted = false;
6377         struct sock *sk = NULL;
6378
6379         if (family == AF_INET) {
6380                 __be32 src4 = tuple->ipv4.saddr;
6381                 __be32 dst4 = tuple->ipv4.daddr;
6382
6383                 if (proto == IPPROTO_TCP)
6384                         sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0,
6385                                            src4, tuple->ipv4.sport,
6386                                            dst4, tuple->ipv4.dport,
6387                                            dif, sdif, &refcounted);
6388                 else
6389                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
6390                                                dst4, tuple->ipv4.dport,
6391                                                dif, sdif, &udp_table, NULL);
6392 #if IS_ENABLED(CONFIG_IPV6)
6393         } else {
6394                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
6395                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
6396
6397                 if (proto == IPPROTO_TCP)
6398                         sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0,
6399                                             src6, tuple->ipv6.sport,
6400                                             dst6, ntohs(tuple->ipv6.dport),
6401                                             dif, sdif, &refcounted);
6402                 else if (likely(ipv6_bpf_stub))
6403                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
6404                                                             src6, tuple->ipv6.sport,
6405                                                             dst6, tuple->ipv6.dport,
6406                                                             dif, sdif,
6407                                                             &udp_table, NULL);
6408 #endif
6409         }
6410
6411         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
6412                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6413                 sk = NULL;
6414         }
6415         return sk;
6416 }
6417
6418 /* bpf_skc_lookup performs the core lookup for different types of sockets,
6419  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
6420  */
6421 static struct sock *
6422 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6423                  struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6424                  u64 flags)
6425 {
6426         struct sock *sk = NULL;
6427         struct net *net;
6428         u8 family;
6429         int sdif;
6430
6431         if (len == sizeof(tuple->ipv4))
6432                 family = AF_INET;
6433         else if (len == sizeof(tuple->ipv6))
6434                 family = AF_INET6;
6435         else
6436                 return NULL;
6437
6438         if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX)))
6439                 goto out;
6440
6441         if (family == AF_INET)
6442                 sdif = inet_sdif(skb);
6443         else
6444                 sdif = inet6_sdif(skb);
6445
6446         if ((s32)netns_id < 0) {
6447                 net = caller_net;
6448                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6449         } else {
6450                 net = get_net_ns_by_id(caller_net, netns_id);
6451                 if (unlikely(!net))
6452                         goto out;
6453                 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto);
6454                 put_net(net);
6455         }
6456
6457 out:
6458         return sk;
6459 }
6460
6461 static struct sock *
6462 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6463                 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id,
6464                 u64 flags)
6465 {
6466         struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net,
6467                                            ifindex, proto, netns_id, flags);
6468
6469         if (sk) {
6470                 struct sock *sk2 = sk_to_full_sk(sk);
6471
6472                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6473                  * sock refcnt is decremented to prevent a request_sock leak.
6474                  */
6475                 if (!sk_fullsock(sk2))
6476                         sk2 = NULL;
6477                 if (sk2 != sk) {
6478                         sock_gen_put(sk);
6479                         /* Ensure there is no need to bump sk2 refcnt */
6480                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6481                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6482                                 return NULL;
6483                         }
6484                         sk = sk2;
6485                 }
6486         }
6487
6488         return sk;
6489 }
6490
6491 static struct sock *
6492 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6493                u8 proto, u64 netns_id, u64 flags)
6494 {
6495         struct net *caller_net;
6496         int ifindex;
6497
6498         if (skb->dev) {
6499                 caller_net = dev_net(skb->dev);
6500                 ifindex = skb->dev->ifindex;
6501         } else {
6502                 caller_net = sock_net(skb->sk);
6503                 ifindex = 0;
6504         }
6505
6506         return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto,
6507                                 netns_id, flags);
6508 }
6509
6510 static struct sock *
6511 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
6512               u8 proto, u64 netns_id, u64 flags)
6513 {
6514         struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id,
6515                                          flags);
6516
6517         if (sk) {
6518                 struct sock *sk2 = sk_to_full_sk(sk);
6519
6520                 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk
6521                  * sock refcnt is decremented to prevent a request_sock leak.
6522                  */
6523                 if (!sk_fullsock(sk2))
6524                         sk2 = NULL;
6525                 if (sk2 != sk) {
6526                         sock_gen_put(sk);
6527                         /* Ensure there is no need to bump sk2 refcnt */
6528                         if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) {
6529                                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
6530                                 return NULL;
6531                         }
6532                         sk = sk2;
6533                 }
6534         }
6535
6536         return sk;
6537 }
6538
6539 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb,
6540            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6541 {
6542         return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP,
6543                                              netns_id, flags);
6544 }
6545
6546 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = {
6547         .func           = bpf_skc_lookup_tcp,
6548         .gpl_only       = false,
6549         .pkt_access     = true,
6550         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6551         .arg1_type      = ARG_PTR_TO_CTX,
6552         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6553         .arg3_type      = ARG_CONST_SIZE,
6554         .arg4_type      = ARG_ANYTHING,
6555         .arg5_type      = ARG_ANYTHING,
6556 };
6557
6558 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
6559            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6560 {
6561         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP,
6562                                             netns_id, flags);
6563 }
6564
6565 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
6566         .func           = bpf_sk_lookup_tcp,
6567         .gpl_only       = false,
6568         .pkt_access     = true,
6569         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6570         .arg1_type      = ARG_PTR_TO_CTX,
6571         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6572         .arg3_type      = ARG_CONST_SIZE,
6573         .arg4_type      = ARG_ANYTHING,
6574         .arg5_type      = ARG_ANYTHING,
6575 };
6576
6577 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
6578            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6579 {
6580         return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP,
6581                                             netns_id, flags);
6582 }
6583
6584 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
6585         .func           = bpf_sk_lookup_udp,
6586         .gpl_only       = false,
6587         .pkt_access     = true,
6588         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6589         .arg1_type      = ARG_PTR_TO_CTX,
6590         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6591         .arg3_type      = ARG_CONST_SIZE,
6592         .arg4_type      = ARG_ANYTHING,
6593         .arg5_type      = ARG_ANYTHING,
6594 };
6595
6596 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
6597 {
6598         if (sk && sk_is_refcounted(sk))
6599                 sock_gen_put(sk);
6600         return 0;
6601 }
6602
6603 static const struct bpf_func_proto bpf_sk_release_proto = {
6604         .func           = bpf_sk_release,
6605         .gpl_only       = false,
6606         .ret_type       = RET_INTEGER,
6607         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE,
6608 };
6609
6610 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx,
6611            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6612 {
6613         struct net *caller_net = dev_net(ctx->rxq->dev);
6614         int ifindex = ctx->rxq->dev->ifindex;
6615
6616         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6617                                               ifindex, IPPROTO_UDP, netns_id,
6618                                               flags);
6619 }
6620
6621 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = {
6622         .func           = bpf_xdp_sk_lookup_udp,
6623         .gpl_only       = false,
6624         .pkt_access     = true,
6625         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6626         .arg1_type      = ARG_PTR_TO_CTX,
6627         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6628         .arg3_type      = ARG_CONST_SIZE,
6629         .arg4_type      = ARG_ANYTHING,
6630         .arg5_type      = ARG_ANYTHING,
6631 };
6632
6633 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx,
6634            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6635 {
6636         struct net *caller_net = dev_net(ctx->rxq->dev);
6637         int ifindex = ctx->rxq->dev->ifindex;
6638
6639         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net,
6640                                                ifindex, IPPROTO_TCP, netns_id,
6641                                                flags);
6642 }
6643
6644 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = {
6645         .func           = bpf_xdp_skc_lookup_tcp,
6646         .gpl_only       = false,
6647         .pkt_access     = true,
6648         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6649         .arg1_type      = ARG_PTR_TO_CTX,
6650         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6651         .arg3_type      = ARG_CONST_SIZE,
6652         .arg4_type      = ARG_ANYTHING,
6653         .arg5_type      = ARG_ANYTHING,
6654 };
6655
6656 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx,
6657            struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags)
6658 {
6659         struct net *caller_net = dev_net(ctx->rxq->dev);
6660         int ifindex = ctx->rxq->dev->ifindex;
6661
6662         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net,
6663                                               ifindex, IPPROTO_TCP, netns_id,
6664                                               flags);
6665 }
6666
6667 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = {
6668         .func           = bpf_xdp_sk_lookup_tcp,
6669         .gpl_only       = false,
6670         .pkt_access     = true,
6671         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6672         .arg1_type      = ARG_PTR_TO_CTX,
6673         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6674         .arg3_type      = ARG_CONST_SIZE,
6675         .arg4_type      = ARG_ANYTHING,
6676         .arg5_type      = ARG_ANYTHING,
6677 };
6678
6679 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6680            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6681 {
6682         return (unsigned long)__bpf_skc_lookup(NULL, tuple, len,
6683                                                sock_net(ctx->sk), 0,
6684                                                IPPROTO_TCP, netns_id, flags);
6685 }
6686
6687 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = {
6688         .func           = bpf_sock_addr_skc_lookup_tcp,
6689         .gpl_only       = false,
6690         .ret_type       = RET_PTR_TO_SOCK_COMMON_OR_NULL,
6691         .arg1_type      = ARG_PTR_TO_CTX,
6692         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6693         .arg3_type      = ARG_CONST_SIZE,
6694         .arg4_type      = ARG_ANYTHING,
6695         .arg5_type      = ARG_ANYTHING,
6696 };
6697
6698 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx,
6699            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6700 {
6701         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6702                                               sock_net(ctx->sk), 0, IPPROTO_TCP,
6703                                               netns_id, flags);
6704 }
6705
6706 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = {
6707         .func           = bpf_sock_addr_sk_lookup_tcp,
6708         .gpl_only       = false,
6709         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6710         .arg1_type      = ARG_PTR_TO_CTX,
6711         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6712         .arg3_type      = ARG_CONST_SIZE,
6713         .arg4_type      = ARG_ANYTHING,
6714         .arg5_type      = ARG_ANYTHING,
6715 };
6716
6717 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx,
6718            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
6719 {
6720         return (unsigned long)__bpf_sk_lookup(NULL, tuple, len,
6721                                               sock_net(ctx->sk), 0, IPPROTO_UDP,
6722                                               netns_id, flags);
6723 }
6724
6725 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = {
6726         .func           = bpf_sock_addr_sk_lookup_udp,
6727         .gpl_only       = false,
6728         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6729         .arg1_type      = ARG_PTR_TO_CTX,
6730         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
6731         .arg3_type      = ARG_CONST_SIZE,
6732         .arg4_type      = ARG_ANYTHING,
6733         .arg5_type      = ARG_ANYTHING,
6734 };
6735
6736 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6737                                   struct bpf_insn_access_aux *info)
6738 {
6739         if (off < 0 || off >= offsetofend(struct bpf_tcp_sock,
6740                                           icsk_retransmits))
6741                 return false;
6742
6743         if (off % size != 0)
6744                 return false;
6745
6746         switch (off) {
6747         case offsetof(struct bpf_tcp_sock, bytes_received):
6748         case offsetof(struct bpf_tcp_sock, bytes_acked):
6749                 return size == sizeof(__u64);
6750         default:
6751                 return size == sizeof(__u32);
6752         }
6753 }
6754
6755 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type,
6756                                     const struct bpf_insn *si,
6757                                     struct bpf_insn *insn_buf,
6758                                     struct bpf_prog *prog, u32 *target_size)
6759 {
6760         struct bpf_insn *insn = insn_buf;
6761
6762 #define BPF_TCP_SOCK_GET_COMMON(FIELD)                                  \
6763         do {                                                            \
6764                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) >     \
6765                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6766                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\
6767                                       si->dst_reg, si->src_reg,         \
6768                                       offsetof(struct tcp_sock, FIELD)); \
6769         } while (0)
6770
6771 #define BPF_INET_SOCK_GET_COMMON(FIELD)                                 \
6772         do {                                                            \
6773                 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock,  \
6774                                           FIELD) >                      \
6775                              sizeof_field(struct bpf_tcp_sock, FIELD)); \
6776                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                 \
6777                                         struct inet_connection_sock,    \
6778                                         FIELD),                         \
6779                                       si->dst_reg, si->src_reg,         \
6780                                       offsetof(                         \
6781                                         struct inet_connection_sock,    \
6782                                         FIELD));                        \
6783         } while (0)
6784
6785         if (insn > insn_buf)
6786                 return insn - insn_buf;
6787
6788         switch (si->off) {
6789         case offsetof(struct bpf_tcp_sock, rtt_min):
6790                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
6791                              sizeof(struct minmax));
6792                 BUILD_BUG_ON(sizeof(struct minmax) <
6793                              sizeof(struct minmax_sample));
6794
6795                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6796                                       offsetof(struct tcp_sock, rtt_min) +
6797                                       offsetof(struct minmax_sample, v));
6798                 break;
6799         case offsetof(struct bpf_tcp_sock, snd_cwnd):
6800                 BPF_TCP_SOCK_GET_COMMON(snd_cwnd);
6801                 break;
6802         case offsetof(struct bpf_tcp_sock, srtt_us):
6803                 BPF_TCP_SOCK_GET_COMMON(srtt_us);
6804                 break;
6805         case offsetof(struct bpf_tcp_sock, snd_ssthresh):
6806                 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh);
6807                 break;
6808         case offsetof(struct bpf_tcp_sock, rcv_nxt):
6809                 BPF_TCP_SOCK_GET_COMMON(rcv_nxt);
6810                 break;
6811         case offsetof(struct bpf_tcp_sock, snd_nxt):
6812                 BPF_TCP_SOCK_GET_COMMON(snd_nxt);
6813                 break;
6814         case offsetof(struct bpf_tcp_sock, snd_una):
6815                 BPF_TCP_SOCK_GET_COMMON(snd_una);
6816                 break;
6817         case offsetof(struct bpf_tcp_sock, mss_cache):
6818                 BPF_TCP_SOCK_GET_COMMON(mss_cache);
6819                 break;
6820         case offsetof(struct bpf_tcp_sock, ecn_flags):
6821                 BPF_TCP_SOCK_GET_COMMON(ecn_flags);
6822                 break;
6823         case offsetof(struct bpf_tcp_sock, rate_delivered):
6824                 BPF_TCP_SOCK_GET_COMMON(rate_delivered);
6825                 break;
6826         case offsetof(struct bpf_tcp_sock, rate_interval_us):
6827                 BPF_TCP_SOCK_GET_COMMON(rate_interval_us);
6828                 break;
6829         case offsetof(struct bpf_tcp_sock, packets_out):
6830                 BPF_TCP_SOCK_GET_COMMON(packets_out);
6831                 break;
6832         case offsetof(struct bpf_tcp_sock, retrans_out):
6833                 BPF_TCP_SOCK_GET_COMMON(retrans_out);
6834                 break;
6835         case offsetof(struct bpf_tcp_sock, total_retrans):
6836                 BPF_TCP_SOCK_GET_COMMON(total_retrans);
6837                 break;
6838         case offsetof(struct bpf_tcp_sock, segs_in):
6839                 BPF_TCP_SOCK_GET_COMMON(segs_in);
6840                 break;
6841         case offsetof(struct bpf_tcp_sock, data_segs_in):
6842                 BPF_TCP_SOCK_GET_COMMON(data_segs_in);
6843                 break;
6844         case offsetof(struct bpf_tcp_sock, segs_out):
6845                 BPF_TCP_SOCK_GET_COMMON(segs_out);
6846                 break;
6847         case offsetof(struct bpf_tcp_sock, data_segs_out):
6848                 BPF_TCP_SOCK_GET_COMMON(data_segs_out);
6849                 break;
6850         case offsetof(struct bpf_tcp_sock, lost_out):
6851                 BPF_TCP_SOCK_GET_COMMON(lost_out);
6852                 break;
6853         case offsetof(struct bpf_tcp_sock, sacked_out):
6854                 BPF_TCP_SOCK_GET_COMMON(sacked_out);
6855                 break;
6856         case offsetof(struct bpf_tcp_sock, bytes_received):
6857                 BPF_TCP_SOCK_GET_COMMON(bytes_received);
6858                 break;
6859         case offsetof(struct bpf_tcp_sock, bytes_acked):
6860                 BPF_TCP_SOCK_GET_COMMON(bytes_acked);
6861                 break;
6862         case offsetof(struct bpf_tcp_sock, dsack_dups):
6863                 BPF_TCP_SOCK_GET_COMMON(dsack_dups);
6864                 break;
6865         case offsetof(struct bpf_tcp_sock, delivered):
6866                 BPF_TCP_SOCK_GET_COMMON(delivered);
6867                 break;
6868         case offsetof(struct bpf_tcp_sock, delivered_ce):
6869                 BPF_TCP_SOCK_GET_COMMON(delivered_ce);
6870                 break;
6871         case offsetof(struct bpf_tcp_sock, icsk_retransmits):
6872                 BPF_INET_SOCK_GET_COMMON(icsk_retransmits);
6873                 break;
6874         }
6875
6876         return insn - insn_buf;
6877 }
6878
6879 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk)
6880 {
6881         if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
6882                 return (unsigned long)sk;
6883
6884         return (unsigned long)NULL;
6885 }
6886
6887 const struct bpf_func_proto bpf_tcp_sock_proto = {
6888         .func           = bpf_tcp_sock,
6889         .gpl_only       = false,
6890         .ret_type       = RET_PTR_TO_TCP_SOCK_OR_NULL,
6891         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6892 };
6893
6894 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk)
6895 {
6896         sk = sk_to_full_sk(sk);
6897
6898         if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE))
6899                 return (unsigned long)sk;
6900
6901         return (unsigned long)NULL;
6902 }
6903
6904 static const struct bpf_func_proto bpf_get_listener_sock_proto = {
6905         .func           = bpf_get_listener_sock,
6906         .gpl_only       = false,
6907         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
6908         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
6909 };
6910
6911 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb)
6912 {
6913         unsigned int iphdr_len;
6914
6915         switch (skb_protocol(skb, true)) {
6916         case cpu_to_be16(ETH_P_IP):
6917                 iphdr_len = sizeof(struct iphdr);
6918                 break;
6919         case cpu_to_be16(ETH_P_IPV6):
6920                 iphdr_len = sizeof(struct ipv6hdr);
6921                 break;
6922         default:
6923                 return 0;
6924         }
6925
6926         if (skb_headlen(skb) < iphdr_len)
6927                 return 0;
6928
6929         if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len))
6930                 return 0;
6931
6932         return INET_ECN_set_ce(skb);
6933 }
6934
6935 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type,
6936                                   struct bpf_insn_access_aux *info)
6937 {
6938         if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id))
6939                 return false;
6940
6941         if (off % size != 0)
6942                 return false;
6943
6944         switch (off) {
6945         default:
6946                 return size == sizeof(__u32);
6947         }
6948 }
6949
6950 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type,
6951                                     const struct bpf_insn *si,
6952                                     struct bpf_insn *insn_buf,
6953                                     struct bpf_prog *prog, u32 *target_size)
6954 {
6955         struct bpf_insn *insn = insn_buf;
6956
6957 #define BPF_XDP_SOCK_GET(FIELD)                                         \
6958         do {                                                            \
6959                 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) >     \
6960                              sizeof_field(struct bpf_xdp_sock, FIELD)); \
6961                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\
6962                                       si->dst_reg, si->src_reg,         \
6963                                       offsetof(struct xdp_sock, FIELD)); \
6964         } while (0)
6965
6966         switch (si->off) {
6967         case offsetof(struct bpf_xdp_sock, queue_id):
6968                 BPF_XDP_SOCK_GET(queue_id);
6969                 break;
6970         }
6971
6972         return insn - insn_buf;
6973 }
6974
6975 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = {
6976         .func           = bpf_skb_ecn_set_ce,
6977         .gpl_only       = false,
6978         .ret_type       = RET_INTEGER,
6979         .arg1_type      = ARG_PTR_TO_CTX,
6980 };
6981
6982 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
6983            struct tcphdr *, th, u32, th_len)
6984 {
6985 #ifdef CONFIG_SYN_COOKIES
6986         u32 cookie;
6987         int ret;
6988
6989         if (unlikely(!sk || th_len < sizeof(*th)))
6990                 return -EINVAL;
6991
6992         /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */
6993         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
6994                 return -EINVAL;
6995
6996         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
6997                 return -EINVAL;
6998
6999         if (!th->ack || th->rst || th->syn)
7000                 return -ENOENT;
7001
7002         if (unlikely(iph_len < sizeof(struct iphdr)))
7003                 return -EINVAL;
7004
7005         if (tcp_synq_no_recent_overflow(sk))
7006                 return -ENOENT;
7007
7008         cookie = ntohl(th->ack_seq) - 1;
7009
7010         /* Both struct iphdr and struct ipv6hdr have the version field at the
7011          * same offset so we can cast to the shorter header (struct iphdr).
7012          */
7013         switch (((struct iphdr *)iph)->version) {
7014         case 4:
7015                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7016                         return -EINVAL;
7017
7018                 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie);
7019                 break;
7020
7021 #if IS_BUILTIN(CONFIG_IPV6)
7022         case 6:
7023                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7024                         return -EINVAL;
7025
7026                 if (sk->sk_family != AF_INET6)
7027                         return -EINVAL;
7028
7029                 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie);
7030                 break;
7031 #endif /* CONFIG_IPV6 */
7032
7033         default:
7034                 return -EPROTONOSUPPORT;
7035         }
7036
7037         if (ret > 0)
7038                 return 0;
7039
7040         return -ENOENT;
7041 #else
7042         return -ENOTSUPP;
7043 #endif
7044 }
7045
7046 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = {
7047         .func           = bpf_tcp_check_syncookie,
7048         .gpl_only       = true,
7049         .pkt_access     = true,
7050         .ret_type       = RET_INTEGER,
7051         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7052         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7053         .arg3_type      = ARG_CONST_SIZE,
7054         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7055         .arg5_type      = ARG_CONST_SIZE,
7056 };
7057
7058 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len,
7059            struct tcphdr *, th, u32, th_len)
7060 {
7061 #ifdef CONFIG_SYN_COOKIES
7062         u32 cookie;
7063         u16 mss;
7064
7065         if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4))
7066                 return -EINVAL;
7067
7068         if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN)
7069                 return -EINVAL;
7070
7071         if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies))
7072                 return -ENOENT;
7073
7074         if (!th->syn || th->ack || th->fin || th->rst)
7075                 return -EINVAL;
7076
7077         if (unlikely(iph_len < sizeof(struct iphdr)))
7078                 return -EINVAL;
7079
7080         /* Both struct iphdr and struct ipv6hdr have the version field at the
7081          * same offset so we can cast to the shorter header (struct iphdr).
7082          */
7083         switch (((struct iphdr *)iph)->version) {
7084         case 4:
7085                 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk))
7086                         return -EINVAL;
7087
7088                 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie);
7089                 break;
7090
7091 #if IS_BUILTIN(CONFIG_IPV6)
7092         case 6:
7093                 if (unlikely(iph_len < sizeof(struct ipv6hdr)))
7094                         return -EINVAL;
7095
7096                 if (sk->sk_family != AF_INET6)
7097                         return -EINVAL;
7098
7099                 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie);
7100                 break;
7101 #endif /* CONFIG_IPV6 */
7102
7103         default:
7104                 return -EPROTONOSUPPORT;
7105         }
7106         if (mss == 0)
7107                 return -ENOENT;
7108
7109         return cookie | ((u64)mss << 32);
7110 #else
7111         return -EOPNOTSUPP;
7112 #endif /* CONFIG_SYN_COOKIES */
7113 }
7114
7115 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = {
7116         .func           = bpf_tcp_gen_syncookie,
7117         .gpl_only       = true, /* __cookie_v*_init_sequence() is GPL */
7118         .pkt_access     = true,
7119         .ret_type       = RET_INTEGER,
7120         .arg1_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7121         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7122         .arg3_type      = ARG_CONST_SIZE,
7123         .arg4_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7124         .arg5_type      = ARG_CONST_SIZE,
7125 };
7126
7127 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags)
7128 {
7129         if (!sk || flags != 0)
7130                 return -EINVAL;
7131         if (!skb_at_tc_ingress(skb))
7132                 return -EOPNOTSUPP;
7133         if (unlikely(dev_net(skb->dev) != sock_net(sk)))
7134                 return -ENETUNREACH;
7135         if (unlikely(sk_fullsock(sk) && sk->sk_reuseport))
7136                 return -ESOCKTNOSUPPORT;
7137         if (sk_is_refcounted(sk) &&
7138             unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
7139                 return -ENOENT;
7140
7141         skb_orphan(skb);
7142         skb->sk = sk;
7143         skb->destructor = sock_pfree;
7144
7145         return 0;
7146 }
7147
7148 static const struct bpf_func_proto bpf_sk_assign_proto = {
7149         .func           = bpf_sk_assign,
7150         .gpl_only       = false,
7151         .ret_type       = RET_INTEGER,
7152         .arg1_type      = ARG_PTR_TO_CTX,
7153         .arg2_type      = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
7154         .arg3_type      = ARG_ANYTHING,
7155 };
7156
7157 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend,
7158                                     u8 search_kind, const u8 *magic,
7159                                     u8 magic_len, bool *eol)
7160 {
7161         u8 kind, kind_len;
7162
7163         *eol = false;
7164
7165         while (op < opend) {
7166                 kind = op[0];
7167
7168                 if (kind == TCPOPT_EOL) {
7169                         *eol = true;
7170                         return ERR_PTR(-ENOMSG);
7171                 } else if (kind == TCPOPT_NOP) {
7172                         op++;
7173                         continue;
7174                 }
7175
7176                 if (opend - op < 2 || opend - op < op[1] || op[1] < 2)
7177                         /* Something is wrong in the received header.
7178                          * Follow the TCP stack's tcp_parse_options()
7179                          * and just bail here.
7180                          */
7181                         return ERR_PTR(-EFAULT);
7182
7183                 kind_len = op[1];
7184                 if (search_kind == kind) {
7185                         if (!magic_len)
7186                                 return op;
7187
7188                         if (magic_len > kind_len - 2)
7189                                 return ERR_PTR(-ENOMSG);
7190
7191                         if (!memcmp(&op[2], magic, magic_len))
7192                                 return op;
7193                 }
7194
7195                 op += kind_len;
7196         }
7197
7198         return ERR_PTR(-ENOMSG);
7199 }
7200
7201 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7202            void *, search_res, u32, len, u64, flags)
7203 {
7204         bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN;
7205         const u8 *op, *opend, *magic, *search = search_res;
7206         u8 search_kind, search_len, copy_len, magic_len;
7207         int ret;
7208
7209         /* 2 byte is the minimal option len except TCPOPT_NOP and
7210          * TCPOPT_EOL which are useless for the bpf prog to learn
7211          * and this helper disallow loading them also.
7212          */
7213         if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN)
7214                 return -EINVAL;
7215
7216         search_kind = search[0];
7217         search_len = search[1];
7218
7219         if (search_len > len || search_kind == TCPOPT_NOP ||
7220             search_kind == TCPOPT_EOL)
7221                 return -EINVAL;
7222
7223         if (search_kind == TCPOPT_EXP || search_kind == 253) {
7224                 /* 16 or 32 bit magic.  +2 for kind and kind length */
7225                 if (search_len != 4 && search_len != 6)
7226                         return -EINVAL;
7227                 magic = &search[2];
7228                 magic_len = search_len - 2;
7229         } else {
7230                 if (search_len)
7231                         return -EINVAL;
7232                 magic = NULL;
7233                 magic_len = 0;
7234         }
7235
7236         if (load_syn) {
7237                 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op);
7238                 if (ret < 0)
7239                         return ret;
7240
7241                 opend = op + ret;
7242                 op += sizeof(struct tcphdr);
7243         } else {
7244                 if (!bpf_sock->skb ||
7245                     bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7246                         /* This bpf_sock->op cannot call this helper */
7247                         return -EPERM;
7248
7249                 opend = bpf_sock->skb_data_end;
7250                 op = bpf_sock->skb->data + sizeof(struct tcphdr);
7251         }
7252
7253         op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len,
7254                                 &eol);
7255         if (IS_ERR(op))
7256                 return PTR_ERR(op);
7257
7258         copy_len = op[1];
7259         ret = copy_len;
7260         if (copy_len > len) {
7261                 ret = -ENOSPC;
7262                 copy_len = len;
7263         }
7264
7265         memcpy(search_res, op, copy_len);
7266         return ret;
7267 }
7268
7269 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = {
7270         .func           = bpf_sock_ops_load_hdr_opt,
7271         .gpl_only       = false,
7272         .ret_type       = RET_INTEGER,
7273         .arg1_type      = ARG_PTR_TO_CTX,
7274         .arg2_type      = ARG_PTR_TO_MEM,
7275         .arg3_type      = ARG_CONST_SIZE,
7276         .arg4_type      = ARG_ANYTHING,
7277 };
7278
7279 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7280            const void *, from, u32, len, u64, flags)
7281 {
7282         u8 new_kind, new_kind_len, magic_len = 0, *opend;
7283         const u8 *op, *new_op, *magic = NULL;
7284         struct sk_buff *skb;
7285         bool eol;
7286
7287         if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB)
7288                 return -EPERM;
7289
7290         if (len < 2 || flags)
7291                 return -EINVAL;
7292
7293         new_op = from;
7294         new_kind = new_op[0];
7295         new_kind_len = new_op[1];
7296
7297         if (new_kind_len > len || new_kind == TCPOPT_NOP ||
7298             new_kind == TCPOPT_EOL)
7299                 return -EINVAL;
7300
7301         if (new_kind_len > bpf_sock->remaining_opt_len)
7302                 return -ENOSPC;
7303
7304         /* 253 is another experimental kind */
7305         if (new_kind == TCPOPT_EXP || new_kind == 253)  {
7306                 if (new_kind_len < 4)
7307                         return -EINVAL;
7308                 /* Match for the 2 byte magic also.
7309                  * RFC 6994: the magic could be 2 or 4 bytes.
7310                  * Hence, matching by 2 byte only is on the
7311                  * conservative side but it is the right
7312                  * thing to do for the 'search-for-duplication'
7313                  * purpose.
7314                  */
7315                 magic = &new_op[2];
7316                 magic_len = 2;
7317         }
7318
7319         /* Check for duplication */
7320         skb = bpf_sock->skb;
7321         op = skb->data + sizeof(struct tcphdr);
7322         opend = bpf_sock->skb_data_end;
7323
7324         op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len,
7325                                 &eol);
7326         if (!IS_ERR(op))
7327                 return -EEXIST;
7328
7329         if (PTR_ERR(op) != -ENOMSG)
7330                 return PTR_ERR(op);
7331
7332         if (eol)
7333                 /* The option has been ended.  Treat it as no more
7334                  * header option can be written.
7335                  */
7336                 return -ENOSPC;
7337
7338         /* No duplication found.  Store the header option. */
7339         memcpy(opend, from, new_kind_len);
7340
7341         bpf_sock->remaining_opt_len -= new_kind_len;
7342         bpf_sock->skb_data_end += new_kind_len;
7343
7344         return 0;
7345 }
7346
7347 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = {
7348         .func           = bpf_sock_ops_store_hdr_opt,
7349         .gpl_only       = false,
7350         .ret_type       = RET_INTEGER,
7351         .arg1_type      = ARG_PTR_TO_CTX,
7352         .arg2_type      = ARG_PTR_TO_MEM | MEM_RDONLY,
7353         .arg3_type      = ARG_CONST_SIZE,
7354         .arg4_type      = ARG_ANYTHING,
7355 };
7356
7357 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock,
7358            u32, len, u64, flags)
7359 {
7360         if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB)
7361                 return -EPERM;
7362
7363         if (flags || len < 2)
7364                 return -EINVAL;
7365
7366         if (len > bpf_sock->remaining_opt_len)
7367                 return -ENOSPC;
7368
7369         bpf_sock->remaining_opt_len -= len;
7370
7371         return 0;
7372 }
7373
7374 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = {
7375         .func           = bpf_sock_ops_reserve_hdr_opt,
7376         .gpl_only       = false,
7377         .ret_type       = RET_INTEGER,
7378         .arg1_type      = ARG_PTR_TO_CTX,
7379         .arg2_type      = ARG_ANYTHING,
7380         .arg3_type      = ARG_ANYTHING,
7381 };
7382
7383 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb,
7384            u64, tstamp, u32, tstamp_type)
7385 {
7386         /* skb_clear_delivery_time() is done for inet protocol */
7387         if (skb->protocol != htons(ETH_P_IP) &&
7388             skb->protocol != htons(ETH_P_IPV6))
7389                 return -EOPNOTSUPP;
7390
7391         switch (tstamp_type) {
7392         case BPF_SKB_TSTAMP_DELIVERY_MONO:
7393                 if (!tstamp)
7394                         return -EINVAL;
7395                 skb->tstamp = tstamp;
7396                 skb->mono_delivery_time = 1;
7397                 break;
7398         case BPF_SKB_TSTAMP_UNSPEC:
7399                 if (tstamp)
7400                         return -EINVAL;
7401                 skb->tstamp = 0;
7402                 skb->mono_delivery_time = 0;
7403                 break;
7404         default:
7405                 return -EINVAL;
7406         }
7407
7408         return 0;
7409 }
7410
7411 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = {
7412         .func           = bpf_skb_set_tstamp,
7413         .gpl_only       = false,
7414         .ret_type       = RET_INTEGER,
7415         .arg1_type      = ARG_PTR_TO_CTX,
7416         .arg2_type      = ARG_ANYTHING,
7417         .arg3_type      = ARG_ANYTHING,
7418 };
7419
7420 #ifdef CONFIG_SYN_COOKIES
7421 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph,
7422            struct tcphdr *, th, u32, th_len)
7423 {
7424         u32 cookie;
7425         u16 mss;
7426
7427         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7428                 return -EINVAL;
7429
7430         mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT;
7431         cookie = __cookie_v4_init_sequence(iph, th, &mss);
7432
7433         return cookie | ((u64)mss << 32);
7434 }
7435
7436 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = {
7437         .func           = bpf_tcp_raw_gen_syncookie_ipv4,
7438         .gpl_only       = true, /* __cookie_v4_init_sequence() is GPL */
7439         .pkt_access     = true,
7440         .ret_type       = RET_INTEGER,
7441         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7442         .arg1_size      = sizeof(struct iphdr),
7443         .arg2_type      = ARG_PTR_TO_MEM,
7444         .arg3_type      = ARG_CONST_SIZE,
7445 };
7446
7447 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph,
7448            struct tcphdr *, th, u32, th_len)
7449 {
7450 #if IS_BUILTIN(CONFIG_IPV6)
7451         const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
7452                 sizeof(struct ipv6hdr);
7453         u32 cookie;
7454         u16 mss;
7455
7456         if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4))
7457                 return -EINVAL;
7458
7459         mss = tcp_parse_mss_option(th, 0) ?: mss_clamp;
7460         cookie = __cookie_v6_init_sequence(iph, th, &mss);
7461
7462         return cookie | ((u64)mss << 32);
7463 #else
7464         return -EPROTONOSUPPORT;
7465 #endif
7466 }
7467
7468 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = {
7469         .func           = bpf_tcp_raw_gen_syncookie_ipv6,
7470         .gpl_only       = true, /* __cookie_v6_init_sequence() is GPL */
7471         .pkt_access     = true,
7472         .ret_type       = RET_INTEGER,
7473         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7474         .arg1_size      = sizeof(struct ipv6hdr),
7475         .arg2_type      = ARG_PTR_TO_MEM,
7476         .arg3_type      = ARG_CONST_SIZE,
7477 };
7478
7479 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph,
7480            struct tcphdr *, th)
7481 {
7482         u32 cookie = ntohl(th->ack_seq) - 1;
7483
7484         if (__cookie_v4_check(iph, th, cookie) > 0)
7485                 return 0;
7486
7487         return -EACCES;
7488 }
7489
7490 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = {
7491         .func           = bpf_tcp_raw_check_syncookie_ipv4,
7492         .gpl_only       = true, /* __cookie_v4_check is GPL */
7493         .pkt_access     = true,
7494         .ret_type       = RET_INTEGER,
7495         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7496         .arg1_size      = sizeof(struct iphdr),
7497         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7498         .arg2_size      = sizeof(struct tcphdr),
7499 };
7500
7501 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph,
7502            struct tcphdr *, th)
7503 {
7504 #if IS_BUILTIN(CONFIG_IPV6)
7505         u32 cookie = ntohl(th->ack_seq) - 1;
7506
7507         if (__cookie_v6_check(iph, th, cookie) > 0)
7508                 return 0;
7509
7510         return -EACCES;
7511 #else
7512         return -EPROTONOSUPPORT;
7513 #endif
7514 }
7515
7516 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = {
7517         .func           = bpf_tcp_raw_check_syncookie_ipv6,
7518         .gpl_only       = true, /* __cookie_v6_check is GPL */
7519         .pkt_access     = true,
7520         .ret_type       = RET_INTEGER,
7521         .arg1_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7522         .arg1_size      = sizeof(struct ipv6hdr),
7523         .arg2_type      = ARG_PTR_TO_FIXED_SIZE_MEM,
7524         .arg2_size      = sizeof(struct tcphdr),
7525 };
7526 #endif /* CONFIG_SYN_COOKIES */
7527
7528 #endif /* CONFIG_INET */
7529
7530 bool bpf_helper_changes_pkt_data(void *func)
7531 {
7532         if (func == bpf_skb_vlan_push ||
7533             func == bpf_skb_vlan_pop ||
7534             func == bpf_skb_store_bytes ||
7535             func == bpf_skb_change_proto ||
7536             func == bpf_skb_change_head ||
7537             func == sk_skb_change_head ||
7538             func == bpf_skb_change_tail ||
7539             func == sk_skb_change_tail ||
7540             func == bpf_skb_adjust_room ||
7541             func == sk_skb_adjust_room ||
7542             func == bpf_skb_pull_data ||
7543             func == sk_skb_pull_data ||
7544             func == bpf_clone_redirect ||
7545             func == bpf_l3_csum_replace ||
7546             func == bpf_l4_csum_replace ||
7547             func == bpf_xdp_adjust_head ||
7548             func == bpf_xdp_adjust_meta ||
7549             func == bpf_msg_pull_data ||
7550             func == bpf_msg_push_data ||
7551             func == bpf_msg_pop_data ||
7552             func == bpf_xdp_adjust_tail ||
7553 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
7554             func == bpf_lwt_seg6_store_bytes ||
7555             func == bpf_lwt_seg6_adjust_srh ||
7556             func == bpf_lwt_seg6_action ||
7557 #endif
7558 #ifdef CONFIG_INET
7559             func == bpf_sock_ops_store_hdr_opt ||
7560 #endif
7561             func == bpf_lwt_in_push_encap ||
7562             func == bpf_lwt_xmit_push_encap)
7563                 return true;
7564
7565         return false;
7566 }
7567
7568 const struct bpf_func_proto bpf_event_output_data_proto __weak;
7569 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak;
7570
7571 static const struct bpf_func_proto *
7572 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7573 {
7574         const struct bpf_func_proto *func_proto;
7575
7576         func_proto = cgroup_common_func_proto(func_id, prog);
7577         if (func_proto)
7578                 return func_proto;
7579
7580         func_proto = cgroup_current_func_proto(func_id, prog);
7581         if (func_proto)
7582                 return func_proto;
7583
7584         switch (func_id) {
7585         case BPF_FUNC_get_socket_cookie:
7586                 return &bpf_get_socket_cookie_sock_proto;
7587         case BPF_FUNC_get_netns_cookie:
7588                 return &bpf_get_netns_cookie_sock_proto;
7589         case BPF_FUNC_perf_event_output:
7590                 return &bpf_event_output_data_proto;
7591         case BPF_FUNC_sk_storage_get:
7592                 return &bpf_sk_storage_get_cg_sock_proto;
7593         case BPF_FUNC_ktime_get_coarse_ns:
7594                 return &bpf_ktime_get_coarse_ns_proto;
7595         default:
7596                 return bpf_base_func_proto(func_id);
7597         }
7598 }
7599
7600 static const struct bpf_func_proto *
7601 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7602 {
7603         const struct bpf_func_proto *func_proto;
7604
7605         func_proto = cgroup_common_func_proto(func_id, prog);
7606         if (func_proto)
7607                 return func_proto;
7608
7609         func_proto = cgroup_current_func_proto(func_id, prog);
7610         if (func_proto)
7611                 return func_proto;
7612
7613         switch (func_id) {
7614         case BPF_FUNC_bind:
7615                 switch (prog->expected_attach_type) {
7616                 case BPF_CGROUP_INET4_CONNECT:
7617                 case BPF_CGROUP_INET6_CONNECT:
7618                         return &bpf_bind_proto;
7619                 default:
7620                         return NULL;
7621                 }
7622         case BPF_FUNC_get_socket_cookie:
7623                 return &bpf_get_socket_cookie_sock_addr_proto;
7624         case BPF_FUNC_get_netns_cookie:
7625                 return &bpf_get_netns_cookie_sock_addr_proto;
7626         case BPF_FUNC_perf_event_output:
7627                 return &bpf_event_output_data_proto;
7628 #ifdef CONFIG_INET
7629         case BPF_FUNC_sk_lookup_tcp:
7630                 return &bpf_sock_addr_sk_lookup_tcp_proto;
7631         case BPF_FUNC_sk_lookup_udp:
7632                 return &bpf_sock_addr_sk_lookup_udp_proto;
7633         case BPF_FUNC_sk_release:
7634                 return &bpf_sk_release_proto;
7635         case BPF_FUNC_skc_lookup_tcp:
7636                 return &bpf_sock_addr_skc_lookup_tcp_proto;
7637 #endif /* CONFIG_INET */
7638         case BPF_FUNC_sk_storage_get:
7639                 return &bpf_sk_storage_get_proto;
7640         case BPF_FUNC_sk_storage_delete:
7641                 return &bpf_sk_storage_delete_proto;
7642         case BPF_FUNC_setsockopt:
7643                 switch (prog->expected_attach_type) {
7644                 case BPF_CGROUP_INET4_BIND:
7645                 case BPF_CGROUP_INET6_BIND:
7646                 case BPF_CGROUP_INET4_CONNECT:
7647                 case BPF_CGROUP_INET6_CONNECT:
7648                 case BPF_CGROUP_UDP4_RECVMSG:
7649                 case BPF_CGROUP_UDP6_RECVMSG:
7650                 case BPF_CGROUP_UDP4_SENDMSG:
7651                 case BPF_CGROUP_UDP6_SENDMSG:
7652                 case BPF_CGROUP_INET4_GETPEERNAME:
7653                 case BPF_CGROUP_INET6_GETPEERNAME:
7654                 case BPF_CGROUP_INET4_GETSOCKNAME:
7655                 case BPF_CGROUP_INET6_GETSOCKNAME:
7656                         return &bpf_sock_addr_setsockopt_proto;
7657                 default:
7658                         return NULL;
7659                 }
7660         case BPF_FUNC_getsockopt:
7661                 switch (prog->expected_attach_type) {
7662                 case BPF_CGROUP_INET4_BIND:
7663                 case BPF_CGROUP_INET6_BIND:
7664                 case BPF_CGROUP_INET4_CONNECT:
7665                 case BPF_CGROUP_INET6_CONNECT:
7666                 case BPF_CGROUP_UDP4_RECVMSG:
7667                 case BPF_CGROUP_UDP6_RECVMSG:
7668                 case BPF_CGROUP_UDP4_SENDMSG:
7669                 case BPF_CGROUP_UDP6_SENDMSG:
7670                 case BPF_CGROUP_INET4_GETPEERNAME:
7671                 case BPF_CGROUP_INET6_GETPEERNAME:
7672                 case BPF_CGROUP_INET4_GETSOCKNAME:
7673                 case BPF_CGROUP_INET6_GETSOCKNAME:
7674                         return &bpf_sock_addr_getsockopt_proto;
7675                 default:
7676                         return NULL;
7677                 }
7678         default:
7679                 return bpf_sk_base_func_proto(func_id);
7680         }
7681 }
7682
7683 static const struct bpf_func_proto *
7684 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7685 {
7686         switch (func_id) {
7687         case BPF_FUNC_skb_load_bytes:
7688                 return &bpf_skb_load_bytes_proto;
7689         case BPF_FUNC_skb_load_bytes_relative:
7690                 return &bpf_skb_load_bytes_relative_proto;
7691         case BPF_FUNC_get_socket_cookie:
7692                 return &bpf_get_socket_cookie_proto;
7693         case BPF_FUNC_get_socket_uid:
7694                 return &bpf_get_socket_uid_proto;
7695         case BPF_FUNC_perf_event_output:
7696                 return &bpf_skb_event_output_proto;
7697         default:
7698                 return bpf_sk_base_func_proto(func_id);
7699         }
7700 }
7701
7702 const struct bpf_func_proto bpf_sk_storage_get_proto __weak;
7703 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak;
7704
7705 static const struct bpf_func_proto *
7706 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7707 {
7708         const struct bpf_func_proto *func_proto;
7709
7710         func_proto = cgroup_common_func_proto(func_id, prog);
7711         if (func_proto)
7712                 return func_proto;
7713
7714         switch (func_id) {
7715         case BPF_FUNC_sk_fullsock:
7716                 return &bpf_sk_fullsock_proto;
7717         case BPF_FUNC_sk_storage_get:
7718                 return &bpf_sk_storage_get_proto;
7719         case BPF_FUNC_sk_storage_delete:
7720                 return &bpf_sk_storage_delete_proto;
7721         case BPF_FUNC_perf_event_output:
7722                 return &bpf_skb_event_output_proto;
7723 #ifdef CONFIG_SOCK_CGROUP_DATA
7724         case BPF_FUNC_skb_cgroup_id:
7725                 return &bpf_skb_cgroup_id_proto;
7726         case BPF_FUNC_skb_ancestor_cgroup_id:
7727                 return &bpf_skb_ancestor_cgroup_id_proto;
7728         case BPF_FUNC_sk_cgroup_id:
7729                 return &bpf_sk_cgroup_id_proto;
7730         case BPF_FUNC_sk_ancestor_cgroup_id:
7731                 return &bpf_sk_ancestor_cgroup_id_proto;
7732 #endif
7733 #ifdef CONFIG_INET
7734         case BPF_FUNC_sk_lookup_tcp:
7735                 return &bpf_sk_lookup_tcp_proto;
7736         case BPF_FUNC_sk_lookup_udp:
7737                 return &bpf_sk_lookup_udp_proto;
7738         case BPF_FUNC_sk_release:
7739                 return &bpf_sk_release_proto;
7740         case BPF_FUNC_skc_lookup_tcp:
7741                 return &bpf_skc_lookup_tcp_proto;
7742         case BPF_FUNC_tcp_sock:
7743                 return &bpf_tcp_sock_proto;
7744         case BPF_FUNC_get_listener_sock:
7745                 return &bpf_get_listener_sock_proto;
7746         case BPF_FUNC_skb_ecn_set_ce:
7747                 return &bpf_skb_ecn_set_ce_proto;
7748 #endif
7749         default:
7750                 return sk_filter_func_proto(func_id, prog);
7751         }
7752 }
7753
7754 static const struct bpf_func_proto *
7755 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7756 {
7757         switch (func_id) {
7758         case BPF_FUNC_skb_store_bytes:
7759                 return &bpf_skb_store_bytes_proto;
7760         case BPF_FUNC_skb_load_bytes:
7761                 return &bpf_skb_load_bytes_proto;
7762         case BPF_FUNC_skb_load_bytes_relative:
7763                 return &bpf_skb_load_bytes_relative_proto;
7764         case BPF_FUNC_skb_pull_data:
7765                 return &bpf_skb_pull_data_proto;
7766         case BPF_FUNC_csum_diff:
7767                 return &bpf_csum_diff_proto;
7768         case BPF_FUNC_csum_update:
7769                 return &bpf_csum_update_proto;
7770         case BPF_FUNC_csum_level:
7771                 return &bpf_csum_level_proto;
7772         case BPF_FUNC_l3_csum_replace:
7773                 return &bpf_l3_csum_replace_proto;
7774         case BPF_FUNC_l4_csum_replace:
7775                 return &bpf_l4_csum_replace_proto;
7776         case BPF_FUNC_clone_redirect:
7777                 return &bpf_clone_redirect_proto;
7778         case BPF_FUNC_get_cgroup_classid:
7779                 return &bpf_get_cgroup_classid_proto;
7780         case BPF_FUNC_skb_vlan_push:
7781                 return &bpf_skb_vlan_push_proto;
7782         case BPF_FUNC_skb_vlan_pop:
7783                 return &bpf_skb_vlan_pop_proto;
7784         case BPF_FUNC_skb_change_proto:
7785                 return &bpf_skb_change_proto_proto;
7786         case BPF_FUNC_skb_change_type:
7787                 return &bpf_skb_change_type_proto;
7788         case BPF_FUNC_skb_adjust_room:
7789                 return &bpf_skb_adjust_room_proto;
7790         case BPF_FUNC_skb_change_tail:
7791                 return &bpf_skb_change_tail_proto;
7792         case BPF_FUNC_skb_change_head:
7793                 return &bpf_skb_change_head_proto;
7794         case BPF_FUNC_skb_get_tunnel_key:
7795                 return &bpf_skb_get_tunnel_key_proto;
7796         case BPF_FUNC_skb_set_tunnel_key:
7797                 return bpf_get_skb_set_tunnel_proto(func_id);
7798         case BPF_FUNC_skb_get_tunnel_opt:
7799                 return &bpf_skb_get_tunnel_opt_proto;
7800         case BPF_FUNC_skb_set_tunnel_opt:
7801                 return bpf_get_skb_set_tunnel_proto(func_id);
7802         case BPF_FUNC_redirect:
7803                 return &bpf_redirect_proto;
7804         case BPF_FUNC_redirect_neigh:
7805                 return &bpf_redirect_neigh_proto;
7806         case BPF_FUNC_redirect_peer:
7807                 return &bpf_redirect_peer_proto;
7808         case BPF_FUNC_get_route_realm:
7809                 return &bpf_get_route_realm_proto;
7810         case BPF_FUNC_get_hash_recalc:
7811                 return &bpf_get_hash_recalc_proto;
7812         case BPF_FUNC_set_hash_invalid:
7813                 return &bpf_set_hash_invalid_proto;
7814         case BPF_FUNC_set_hash:
7815                 return &bpf_set_hash_proto;
7816         case BPF_FUNC_perf_event_output:
7817                 return &bpf_skb_event_output_proto;
7818         case BPF_FUNC_get_smp_processor_id:
7819                 return &bpf_get_smp_processor_id_proto;
7820         case BPF_FUNC_skb_under_cgroup:
7821                 return &bpf_skb_under_cgroup_proto;
7822         case BPF_FUNC_get_socket_cookie:
7823                 return &bpf_get_socket_cookie_proto;
7824         case BPF_FUNC_get_socket_uid:
7825                 return &bpf_get_socket_uid_proto;
7826         case BPF_FUNC_fib_lookup:
7827                 return &bpf_skb_fib_lookup_proto;
7828         case BPF_FUNC_check_mtu:
7829                 return &bpf_skb_check_mtu_proto;
7830         case BPF_FUNC_sk_fullsock:
7831                 return &bpf_sk_fullsock_proto;
7832         case BPF_FUNC_sk_storage_get:
7833                 return &bpf_sk_storage_get_proto;
7834         case BPF_FUNC_sk_storage_delete:
7835                 return &bpf_sk_storage_delete_proto;
7836 #ifdef CONFIG_XFRM
7837         case BPF_FUNC_skb_get_xfrm_state:
7838                 return &bpf_skb_get_xfrm_state_proto;
7839 #endif
7840 #ifdef CONFIG_CGROUP_NET_CLASSID
7841         case BPF_FUNC_skb_cgroup_classid:
7842                 return &bpf_skb_cgroup_classid_proto;
7843 #endif
7844 #ifdef CONFIG_SOCK_CGROUP_DATA
7845         case BPF_FUNC_skb_cgroup_id:
7846                 return &bpf_skb_cgroup_id_proto;
7847         case BPF_FUNC_skb_ancestor_cgroup_id:
7848                 return &bpf_skb_ancestor_cgroup_id_proto;
7849 #endif
7850 #ifdef CONFIG_INET
7851         case BPF_FUNC_sk_lookup_tcp:
7852                 return &bpf_sk_lookup_tcp_proto;
7853         case BPF_FUNC_sk_lookup_udp:
7854                 return &bpf_sk_lookup_udp_proto;
7855         case BPF_FUNC_sk_release:
7856                 return &bpf_sk_release_proto;
7857         case BPF_FUNC_tcp_sock:
7858                 return &bpf_tcp_sock_proto;
7859         case BPF_FUNC_get_listener_sock:
7860                 return &bpf_get_listener_sock_proto;
7861         case BPF_FUNC_skc_lookup_tcp:
7862                 return &bpf_skc_lookup_tcp_proto;
7863         case BPF_FUNC_tcp_check_syncookie:
7864                 return &bpf_tcp_check_syncookie_proto;
7865         case BPF_FUNC_skb_ecn_set_ce:
7866                 return &bpf_skb_ecn_set_ce_proto;
7867         case BPF_FUNC_tcp_gen_syncookie:
7868                 return &bpf_tcp_gen_syncookie_proto;
7869         case BPF_FUNC_sk_assign:
7870                 return &bpf_sk_assign_proto;
7871         case BPF_FUNC_skb_set_tstamp:
7872                 return &bpf_skb_set_tstamp_proto;
7873 #ifdef CONFIG_SYN_COOKIES
7874         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7875                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7876         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7877                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7878         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7879                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7880         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7881                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7882 #endif
7883 #endif
7884         default:
7885                 return bpf_sk_base_func_proto(func_id);
7886         }
7887 }
7888
7889 static const struct bpf_func_proto *
7890 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7891 {
7892         switch (func_id) {
7893         case BPF_FUNC_perf_event_output:
7894                 return &bpf_xdp_event_output_proto;
7895         case BPF_FUNC_get_smp_processor_id:
7896                 return &bpf_get_smp_processor_id_proto;
7897         case BPF_FUNC_csum_diff:
7898                 return &bpf_csum_diff_proto;
7899         case BPF_FUNC_xdp_adjust_head:
7900                 return &bpf_xdp_adjust_head_proto;
7901         case BPF_FUNC_xdp_adjust_meta:
7902                 return &bpf_xdp_adjust_meta_proto;
7903         case BPF_FUNC_redirect:
7904                 return &bpf_xdp_redirect_proto;
7905         case BPF_FUNC_redirect_map:
7906                 return &bpf_xdp_redirect_map_proto;
7907         case BPF_FUNC_xdp_adjust_tail:
7908                 return &bpf_xdp_adjust_tail_proto;
7909         case BPF_FUNC_xdp_get_buff_len:
7910                 return &bpf_xdp_get_buff_len_proto;
7911         case BPF_FUNC_xdp_load_bytes:
7912                 return &bpf_xdp_load_bytes_proto;
7913         case BPF_FUNC_xdp_store_bytes:
7914                 return &bpf_xdp_store_bytes_proto;
7915         case BPF_FUNC_fib_lookup:
7916                 return &bpf_xdp_fib_lookup_proto;
7917         case BPF_FUNC_check_mtu:
7918                 return &bpf_xdp_check_mtu_proto;
7919 #ifdef CONFIG_INET
7920         case BPF_FUNC_sk_lookup_udp:
7921                 return &bpf_xdp_sk_lookup_udp_proto;
7922         case BPF_FUNC_sk_lookup_tcp:
7923                 return &bpf_xdp_sk_lookup_tcp_proto;
7924         case BPF_FUNC_sk_release:
7925                 return &bpf_sk_release_proto;
7926         case BPF_FUNC_skc_lookup_tcp:
7927                 return &bpf_xdp_skc_lookup_tcp_proto;
7928         case BPF_FUNC_tcp_check_syncookie:
7929                 return &bpf_tcp_check_syncookie_proto;
7930         case BPF_FUNC_tcp_gen_syncookie:
7931                 return &bpf_tcp_gen_syncookie_proto;
7932 #ifdef CONFIG_SYN_COOKIES
7933         case BPF_FUNC_tcp_raw_gen_syncookie_ipv4:
7934                 return &bpf_tcp_raw_gen_syncookie_ipv4_proto;
7935         case BPF_FUNC_tcp_raw_gen_syncookie_ipv6:
7936                 return &bpf_tcp_raw_gen_syncookie_ipv6_proto;
7937         case BPF_FUNC_tcp_raw_check_syncookie_ipv4:
7938                 return &bpf_tcp_raw_check_syncookie_ipv4_proto;
7939         case BPF_FUNC_tcp_raw_check_syncookie_ipv6:
7940                 return &bpf_tcp_raw_check_syncookie_ipv6_proto;
7941 #endif
7942 #endif
7943         default:
7944                 return bpf_sk_base_func_proto(func_id);
7945         }
7946 }
7947
7948 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
7949 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
7950
7951 static const struct bpf_func_proto *
7952 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
7953 {
7954         const struct bpf_func_proto *func_proto;
7955
7956         func_proto = cgroup_common_func_proto(func_id, prog);
7957         if (func_proto)
7958                 return func_proto;
7959
7960         switch (func_id) {
7961         case BPF_FUNC_setsockopt:
7962                 return &bpf_sock_ops_setsockopt_proto;
7963         case BPF_FUNC_getsockopt:
7964                 return &bpf_sock_ops_getsockopt_proto;
7965         case BPF_FUNC_sock_ops_cb_flags_set:
7966                 return &bpf_sock_ops_cb_flags_set_proto;
7967         case BPF_FUNC_sock_map_update:
7968                 return &bpf_sock_map_update_proto;
7969         case BPF_FUNC_sock_hash_update:
7970                 return &bpf_sock_hash_update_proto;
7971         case BPF_FUNC_get_socket_cookie:
7972                 return &bpf_get_socket_cookie_sock_ops_proto;
7973         case BPF_FUNC_perf_event_output:
7974                 return &bpf_event_output_data_proto;
7975         case BPF_FUNC_sk_storage_get:
7976                 return &bpf_sk_storage_get_proto;
7977         case BPF_FUNC_sk_storage_delete:
7978                 return &bpf_sk_storage_delete_proto;
7979         case BPF_FUNC_get_netns_cookie:
7980                 return &bpf_get_netns_cookie_sock_ops_proto;
7981 #ifdef CONFIG_INET
7982         case BPF_FUNC_load_hdr_opt:
7983                 return &bpf_sock_ops_load_hdr_opt_proto;
7984         case BPF_FUNC_store_hdr_opt:
7985                 return &bpf_sock_ops_store_hdr_opt_proto;
7986         case BPF_FUNC_reserve_hdr_opt:
7987                 return &bpf_sock_ops_reserve_hdr_opt_proto;
7988         case BPF_FUNC_tcp_sock:
7989                 return &bpf_tcp_sock_proto;
7990 #endif /* CONFIG_INET */
7991         default:
7992                 return bpf_sk_base_func_proto(func_id);
7993         }
7994 }
7995
7996 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
7997 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
7998
7999 static const struct bpf_func_proto *
8000 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8001 {
8002         switch (func_id) {
8003         case BPF_FUNC_msg_redirect_map:
8004                 return &bpf_msg_redirect_map_proto;
8005         case BPF_FUNC_msg_redirect_hash:
8006                 return &bpf_msg_redirect_hash_proto;
8007         case BPF_FUNC_msg_apply_bytes:
8008                 return &bpf_msg_apply_bytes_proto;
8009         case BPF_FUNC_msg_cork_bytes:
8010                 return &bpf_msg_cork_bytes_proto;
8011         case BPF_FUNC_msg_pull_data:
8012                 return &bpf_msg_pull_data_proto;
8013         case BPF_FUNC_msg_push_data:
8014                 return &bpf_msg_push_data_proto;
8015         case BPF_FUNC_msg_pop_data:
8016                 return &bpf_msg_pop_data_proto;
8017         case BPF_FUNC_perf_event_output:
8018                 return &bpf_event_output_data_proto;
8019         case BPF_FUNC_get_current_uid_gid:
8020                 return &bpf_get_current_uid_gid_proto;
8021         case BPF_FUNC_get_current_pid_tgid:
8022                 return &bpf_get_current_pid_tgid_proto;
8023         case BPF_FUNC_sk_storage_get:
8024                 return &bpf_sk_storage_get_proto;
8025         case BPF_FUNC_sk_storage_delete:
8026                 return &bpf_sk_storage_delete_proto;
8027         case BPF_FUNC_get_netns_cookie:
8028                 return &bpf_get_netns_cookie_sk_msg_proto;
8029 #ifdef CONFIG_CGROUPS
8030         case BPF_FUNC_get_current_cgroup_id:
8031                 return &bpf_get_current_cgroup_id_proto;
8032         case BPF_FUNC_get_current_ancestor_cgroup_id:
8033                 return &bpf_get_current_ancestor_cgroup_id_proto;
8034 #endif
8035 #ifdef CONFIG_CGROUP_NET_CLASSID
8036         case BPF_FUNC_get_cgroup_classid:
8037                 return &bpf_get_cgroup_classid_curr_proto;
8038 #endif
8039         default:
8040                 return bpf_sk_base_func_proto(func_id);
8041         }
8042 }
8043
8044 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
8045 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
8046
8047 static const struct bpf_func_proto *
8048 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8049 {
8050         switch (func_id) {
8051         case BPF_FUNC_skb_store_bytes:
8052                 return &bpf_skb_store_bytes_proto;
8053         case BPF_FUNC_skb_load_bytes:
8054                 return &bpf_skb_load_bytes_proto;
8055         case BPF_FUNC_skb_pull_data:
8056                 return &sk_skb_pull_data_proto;
8057         case BPF_FUNC_skb_change_tail:
8058                 return &sk_skb_change_tail_proto;
8059         case BPF_FUNC_skb_change_head:
8060                 return &sk_skb_change_head_proto;
8061         case BPF_FUNC_skb_adjust_room:
8062                 return &sk_skb_adjust_room_proto;
8063         case BPF_FUNC_get_socket_cookie:
8064                 return &bpf_get_socket_cookie_proto;
8065         case BPF_FUNC_get_socket_uid:
8066                 return &bpf_get_socket_uid_proto;
8067         case BPF_FUNC_sk_redirect_map:
8068                 return &bpf_sk_redirect_map_proto;
8069         case BPF_FUNC_sk_redirect_hash:
8070                 return &bpf_sk_redirect_hash_proto;
8071         case BPF_FUNC_perf_event_output:
8072                 return &bpf_skb_event_output_proto;
8073 #ifdef CONFIG_INET
8074         case BPF_FUNC_sk_lookup_tcp:
8075                 return &bpf_sk_lookup_tcp_proto;
8076         case BPF_FUNC_sk_lookup_udp:
8077                 return &bpf_sk_lookup_udp_proto;
8078         case BPF_FUNC_sk_release:
8079                 return &bpf_sk_release_proto;
8080         case BPF_FUNC_skc_lookup_tcp:
8081                 return &bpf_skc_lookup_tcp_proto;
8082 #endif
8083         default:
8084                 return bpf_sk_base_func_proto(func_id);
8085         }
8086 }
8087
8088 static const struct bpf_func_proto *
8089 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8090 {
8091         switch (func_id) {
8092         case BPF_FUNC_skb_load_bytes:
8093                 return &bpf_flow_dissector_load_bytes_proto;
8094         default:
8095                 return bpf_sk_base_func_proto(func_id);
8096         }
8097 }
8098
8099 static const struct bpf_func_proto *
8100 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8101 {
8102         switch (func_id) {
8103         case BPF_FUNC_skb_load_bytes:
8104                 return &bpf_skb_load_bytes_proto;
8105         case BPF_FUNC_skb_pull_data:
8106                 return &bpf_skb_pull_data_proto;
8107         case BPF_FUNC_csum_diff:
8108                 return &bpf_csum_diff_proto;
8109         case BPF_FUNC_get_cgroup_classid:
8110                 return &bpf_get_cgroup_classid_proto;
8111         case BPF_FUNC_get_route_realm:
8112                 return &bpf_get_route_realm_proto;
8113         case BPF_FUNC_get_hash_recalc:
8114                 return &bpf_get_hash_recalc_proto;
8115         case BPF_FUNC_perf_event_output:
8116                 return &bpf_skb_event_output_proto;
8117         case BPF_FUNC_get_smp_processor_id:
8118                 return &bpf_get_smp_processor_id_proto;
8119         case BPF_FUNC_skb_under_cgroup:
8120                 return &bpf_skb_under_cgroup_proto;
8121         default:
8122                 return bpf_sk_base_func_proto(func_id);
8123         }
8124 }
8125
8126 static const struct bpf_func_proto *
8127 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8128 {
8129         switch (func_id) {
8130         case BPF_FUNC_lwt_push_encap:
8131                 return &bpf_lwt_in_push_encap_proto;
8132         default:
8133                 return lwt_out_func_proto(func_id, prog);
8134         }
8135 }
8136
8137 static const struct bpf_func_proto *
8138 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8139 {
8140         switch (func_id) {
8141         case BPF_FUNC_skb_get_tunnel_key:
8142                 return &bpf_skb_get_tunnel_key_proto;
8143         case BPF_FUNC_skb_set_tunnel_key:
8144                 return bpf_get_skb_set_tunnel_proto(func_id);
8145         case BPF_FUNC_skb_get_tunnel_opt:
8146                 return &bpf_skb_get_tunnel_opt_proto;
8147         case BPF_FUNC_skb_set_tunnel_opt:
8148                 return bpf_get_skb_set_tunnel_proto(func_id);
8149         case BPF_FUNC_redirect:
8150                 return &bpf_redirect_proto;
8151         case BPF_FUNC_clone_redirect:
8152                 return &bpf_clone_redirect_proto;
8153         case BPF_FUNC_skb_change_tail:
8154                 return &bpf_skb_change_tail_proto;
8155         case BPF_FUNC_skb_change_head:
8156                 return &bpf_skb_change_head_proto;
8157         case BPF_FUNC_skb_store_bytes:
8158                 return &bpf_skb_store_bytes_proto;
8159         case BPF_FUNC_csum_update:
8160                 return &bpf_csum_update_proto;
8161         case BPF_FUNC_csum_level:
8162                 return &bpf_csum_level_proto;
8163         case BPF_FUNC_l3_csum_replace:
8164                 return &bpf_l3_csum_replace_proto;
8165         case BPF_FUNC_l4_csum_replace:
8166                 return &bpf_l4_csum_replace_proto;
8167         case BPF_FUNC_set_hash_invalid:
8168                 return &bpf_set_hash_invalid_proto;
8169         case BPF_FUNC_lwt_push_encap:
8170                 return &bpf_lwt_xmit_push_encap_proto;
8171         default:
8172                 return lwt_out_func_proto(func_id, prog);
8173         }
8174 }
8175
8176 static const struct bpf_func_proto *
8177 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
8178 {
8179         switch (func_id) {
8180 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
8181         case BPF_FUNC_lwt_seg6_store_bytes:
8182                 return &bpf_lwt_seg6_store_bytes_proto;
8183         case BPF_FUNC_lwt_seg6_action:
8184                 return &bpf_lwt_seg6_action_proto;
8185         case BPF_FUNC_lwt_seg6_adjust_srh:
8186                 return &bpf_lwt_seg6_adjust_srh_proto;
8187 #endif
8188         default:
8189                 return lwt_out_func_proto(func_id, prog);
8190         }
8191 }
8192
8193 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
8194                                     const struct bpf_prog *prog,
8195                                     struct bpf_insn_access_aux *info)
8196 {
8197         const int size_default = sizeof(__u32);
8198
8199         if (off < 0 || off >= sizeof(struct __sk_buff))
8200                 return false;
8201
8202         /* The verifier guarantees that size > 0. */
8203         if (off % size != 0)
8204                 return false;
8205
8206         switch (off) {
8207         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8208                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
8209                         return false;
8210                 break;
8211         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
8212         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
8213         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
8214         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
8215         case bpf_ctx_range(struct __sk_buff, data):
8216         case bpf_ctx_range(struct __sk_buff, data_meta):
8217         case bpf_ctx_range(struct __sk_buff, data_end):
8218                 if (size != size_default)
8219                         return false;
8220                 break;
8221         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8222                 return false;
8223         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8224                 if (type == BPF_WRITE || size != sizeof(__u64))
8225                         return false;
8226                 break;
8227         case bpf_ctx_range(struct __sk_buff, tstamp):
8228                 if (size != sizeof(__u64))
8229                         return false;
8230                 break;
8231         case offsetof(struct __sk_buff, sk):
8232                 if (type == BPF_WRITE || size != sizeof(__u64))
8233                         return false;
8234                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
8235                 break;
8236         case offsetof(struct __sk_buff, tstamp_type):
8237                 return false;
8238         case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1:
8239                 /* Explicitly prohibit access to padding in __sk_buff. */
8240                 return false;
8241         default:
8242                 /* Only narrow read access allowed for now. */
8243                 if (type == BPF_WRITE) {
8244                         if (size != size_default)
8245                                 return false;
8246                 } else {
8247                         bpf_ctx_record_field_size(info, size_default);
8248                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8249                                 return false;
8250                 }
8251         }
8252
8253         return true;
8254 }
8255
8256 static bool sk_filter_is_valid_access(int off, int size,
8257                                       enum bpf_access_type type,
8258                                       const struct bpf_prog *prog,
8259                                       struct bpf_insn_access_aux *info)
8260 {
8261         switch (off) {
8262         case bpf_ctx_range(struct __sk_buff, tc_classid):
8263         case bpf_ctx_range(struct __sk_buff, data):
8264         case bpf_ctx_range(struct __sk_buff, data_meta):
8265         case bpf_ctx_range(struct __sk_buff, data_end):
8266         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8267         case bpf_ctx_range(struct __sk_buff, tstamp):
8268         case bpf_ctx_range(struct __sk_buff, wire_len):
8269         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8270                 return false;
8271         }
8272
8273         if (type == BPF_WRITE) {
8274                 switch (off) {
8275                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8276                         break;
8277                 default:
8278                         return false;
8279                 }
8280         }
8281
8282         return bpf_skb_is_valid_access(off, size, type, prog, info);
8283 }
8284
8285 static bool cg_skb_is_valid_access(int off, int size,
8286                                    enum bpf_access_type type,
8287                                    const struct bpf_prog *prog,
8288                                    struct bpf_insn_access_aux *info)
8289 {
8290         switch (off) {
8291         case bpf_ctx_range(struct __sk_buff, tc_classid):
8292         case bpf_ctx_range(struct __sk_buff, data_meta):
8293         case bpf_ctx_range(struct __sk_buff, wire_len):
8294                 return false;
8295         case bpf_ctx_range(struct __sk_buff, data):
8296         case bpf_ctx_range(struct __sk_buff, data_end):
8297                 if (!bpf_capable())
8298                         return false;
8299                 break;
8300         }
8301
8302         if (type == BPF_WRITE) {
8303                 switch (off) {
8304                 case bpf_ctx_range(struct __sk_buff, mark):
8305                 case bpf_ctx_range(struct __sk_buff, priority):
8306                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8307                         break;
8308                 case bpf_ctx_range(struct __sk_buff, tstamp):
8309                         if (!bpf_capable())
8310                                 return false;
8311                         break;
8312                 default:
8313                         return false;
8314                 }
8315         }
8316
8317         switch (off) {
8318         case bpf_ctx_range(struct __sk_buff, data):
8319                 info->reg_type = PTR_TO_PACKET;
8320                 break;
8321         case bpf_ctx_range(struct __sk_buff, data_end):
8322                 info->reg_type = PTR_TO_PACKET_END;
8323                 break;
8324         }
8325
8326         return bpf_skb_is_valid_access(off, size, type, prog, info);
8327 }
8328
8329 static bool lwt_is_valid_access(int off, int size,
8330                                 enum bpf_access_type type,
8331                                 const struct bpf_prog *prog,
8332                                 struct bpf_insn_access_aux *info)
8333 {
8334         switch (off) {
8335         case bpf_ctx_range(struct __sk_buff, tc_classid):
8336         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8337         case bpf_ctx_range(struct __sk_buff, data_meta):
8338         case bpf_ctx_range(struct __sk_buff, tstamp):
8339         case bpf_ctx_range(struct __sk_buff, wire_len):
8340         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8341                 return false;
8342         }
8343
8344         if (type == BPF_WRITE) {
8345                 switch (off) {
8346                 case bpf_ctx_range(struct __sk_buff, mark):
8347                 case bpf_ctx_range(struct __sk_buff, priority):
8348                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8349                         break;
8350                 default:
8351                         return false;
8352                 }
8353         }
8354
8355         switch (off) {
8356         case bpf_ctx_range(struct __sk_buff, data):
8357                 info->reg_type = PTR_TO_PACKET;
8358                 break;
8359         case bpf_ctx_range(struct __sk_buff, data_end):
8360                 info->reg_type = PTR_TO_PACKET_END;
8361                 break;
8362         }
8363
8364         return bpf_skb_is_valid_access(off, size, type, prog, info);
8365 }
8366
8367 /* Attach type specific accesses */
8368 static bool __sock_filter_check_attach_type(int off,
8369                                             enum bpf_access_type access_type,
8370                                             enum bpf_attach_type attach_type)
8371 {
8372         switch (off) {
8373         case offsetof(struct bpf_sock, bound_dev_if):
8374         case offsetof(struct bpf_sock, mark):
8375         case offsetof(struct bpf_sock, priority):
8376                 switch (attach_type) {
8377                 case BPF_CGROUP_INET_SOCK_CREATE:
8378                 case BPF_CGROUP_INET_SOCK_RELEASE:
8379                         goto full_access;
8380                 default:
8381                         return false;
8382                 }
8383         case bpf_ctx_range(struct bpf_sock, src_ip4):
8384                 switch (attach_type) {
8385                 case BPF_CGROUP_INET4_POST_BIND:
8386                         goto read_only;
8387                 default:
8388                         return false;
8389                 }
8390         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8391                 switch (attach_type) {
8392                 case BPF_CGROUP_INET6_POST_BIND:
8393                         goto read_only;
8394                 default:
8395                         return false;
8396                 }
8397         case bpf_ctx_range(struct bpf_sock, src_port):
8398                 switch (attach_type) {
8399                 case BPF_CGROUP_INET4_POST_BIND:
8400                 case BPF_CGROUP_INET6_POST_BIND:
8401                         goto read_only;
8402                 default:
8403                         return false;
8404                 }
8405         }
8406 read_only:
8407         return access_type == BPF_READ;
8408 full_access:
8409         return true;
8410 }
8411
8412 bool bpf_sock_common_is_valid_access(int off, int size,
8413                                      enum bpf_access_type type,
8414                                      struct bpf_insn_access_aux *info)
8415 {
8416         switch (off) {
8417         case bpf_ctx_range_till(struct bpf_sock, type, priority):
8418                 return false;
8419         default:
8420                 return bpf_sock_is_valid_access(off, size, type, info);
8421         }
8422 }
8423
8424 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
8425                               struct bpf_insn_access_aux *info)
8426 {
8427         const int size_default = sizeof(__u32);
8428         int field_size;
8429
8430         if (off < 0 || off >= sizeof(struct bpf_sock))
8431                 return false;
8432         if (off % size != 0)
8433                 return false;
8434
8435         switch (off) {
8436         case offsetof(struct bpf_sock, state):
8437         case offsetof(struct bpf_sock, family):
8438         case offsetof(struct bpf_sock, type):
8439         case offsetof(struct bpf_sock, protocol):
8440         case offsetof(struct bpf_sock, src_port):
8441         case offsetof(struct bpf_sock, rx_queue_mapping):
8442         case bpf_ctx_range(struct bpf_sock, src_ip4):
8443         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
8444         case bpf_ctx_range(struct bpf_sock, dst_ip4):
8445         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
8446                 bpf_ctx_record_field_size(info, size_default);
8447                 return bpf_ctx_narrow_access_ok(off, size, size_default);
8448         case bpf_ctx_range(struct bpf_sock, dst_port):
8449                 field_size = size == size_default ?
8450                         size_default : sizeof_field(struct bpf_sock, dst_port);
8451                 bpf_ctx_record_field_size(info, field_size);
8452                 return bpf_ctx_narrow_access_ok(off, size, field_size);
8453         case offsetofend(struct bpf_sock, dst_port) ...
8454              offsetof(struct bpf_sock, dst_ip4) - 1:
8455                 return false;
8456         }
8457
8458         return size == size_default;
8459 }
8460
8461 static bool sock_filter_is_valid_access(int off, int size,
8462                                         enum bpf_access_type type,
8463                                         const struct bpf_prog *prog,
8464                                         struct bpf_insn_access_aux *info)
8465 {
8466         if (!bpf_sock_is_valid_access(off, size, type, info))
8467                 return false;
8468         return __sock_filter_check_attach_type(off, type,
8469                                                prog->expected_attach_type);
8470 }
8471
8472 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
8473                              const struct bpf_prog *prog)
8474 {
8475         /* Neither direct read nor direct write requires any preliminary
8476          * action.
8477          */
8478         return 0;
8479 }
8480
8481 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
8482                                 const struct bpf_prog *prog, int drop_verdict)
8483 {
8484         struct bpf_insn *insn = insn_buf;
8485
8486         if (!direct_write)
8487                 return 0;
8488
8489         /* if (!skb->cloned)
8490          *       goto start;
8491          *
8492          * (Fast-path, otherwise approximation that we might be
8493          *  a clone, do the rest in helper.)
8494          */
8495         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET);
8496         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
8497         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
8498
8499         /* ret = bpf_skb_pull_data(skb, 0); */
8500         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
8501         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
8502         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
8503                                BPF_FUNC_skb_pull_data);
8504         /* if (!ret)
8505          *      goto restore;
8506          * return TC_ACT_SHOT;
8507          */
8508         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
8509         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
8510         *insn++ = BPF_EXIT_INSN();
8511
8512         /* restore: */
8513         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
8514         /* start: */
8515         *insn++ = prog->insnsi[0];
8516
8517         return insn - insn_buf;
8518 }
8519
8520 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
8521                           struct bpf_insn *insn_buf)
8522 {
8523         bool indirect = BPF_MODE(orig->code) == BPF_IND;
8524         struct bpf_insn *insn = insn_buf;
8525
8526         if (!indirect) {
8527                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
8528         } else {
8529                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
8530                 if (orig->imm)
8531                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
8532         }
8533         /* We're guaranteed here that CTX is in R6. */
8534         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
8535
8536         switch (BPF_SIZE(orig->code)) {
8537         case BPF_B:
8538                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
8539                 break;
8540         case BPF_H:
8541                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
8542                 break;
8543         case BPF_W:
8544                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
8545                 break;
8546         }
8547
8548         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
8549         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
8550         *insn++ = BPF_EXIT_INSN();
8551
8552         return insn - insn_buf;
8553 }
8554
8555 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
8556                                const struct bpf_prog *prog)
8557 {
8558         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
8559 }
8560
8561 static bool tc_cls_act_is_valid_access(int off, int size,
8562                                        enum bpf_access_type type,
8563                                        const struct bpf_prog *prog,
8564                                        struct bpf_insn_access_aux *info)
8565 {
8566         if (type == BPF_WRITE) {
8567                 switch (off) {
8568                 case bpf_ctx_range(struct __sk_buff, mark):
8569                 case bpf_ctx_range(struct __sk_buff, tc_index):
8570                 case bpf_ctx_range(struct __sk_buff, priority):
8571                 case bpf_ctx_range(struct __sk_buff, tc_classid):
8572                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
8573                 case bpf_ctx_range(struct __sk_buff, tstamp):
8574                 case bpf_ctx_range(struct __sk_buff, queue_mapping):
8575                         break;
8576                 default:
8577                         return false;
8578                 }
8579         }
8580
8581         switch (off) {
8582         case bpf_ctx_range(struct __sk_buff, data):
8583                 info->reg_type = PTR_TO_PACKET;
8584                 break;
8585         case bpf_ctx_range(struct __sk_buff, data_meta):
8586                 info->reg_type = PTR_TO_PACKET_META;
8587                 break;
8588         case bpf_ctx_range(struct __sk_buff, data_end):
8589                 info->reg_type = PTR_TO_PACKET_END;
8590                 break;
8591         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
8592                 return false;
8593         case offsetof(struct __sk_buff, tstamp_type):
8594                 /* The convert_ctx_access() on reading and writing
8595                  * __sk_buff->tstamp depends on whether the bpf prog
8596                  * has used __sk_buff->tstamp_type or not.
8597                  * Thus, we need to set prog->tstamp_type_access
8598                  * earlier during is_valid_access() here.
8599                  */
8600                 ((struct bpf_prog *)prog)->tstamp_type_access = 1;
8601                 return size == sizeof(__u8);
8602         }
8603
8604         return bpf_skb_is_valid_access(off, size, type, prog, info);
8605 }
8606
8607 static bool __is_valid_xdp_access(int off, int size)
8608 {
8609         if (off < 0 || off >= sizeof(struct xdp_md))
8610                 return false;
8611         if (off % size != 0)
8612                 return false;
8613         if (size != sizeof(__u32))
8614                 return false;
8615
8616         return true;
8617 }
8618
8619 static bool xdp_is_valid_access(int off, int size,
8620                                 enum bpf_access_type type,
8621                                 const struct bpf_prog *prog,
8622                                 struct bpf_insn_access_aux *info)
8623 {
8624         if (prog->expected_attach_type != BPF_XDP_DEVMAP) {
8625                 switch (off) {
8626                 case offsetof(struct xdp_md, egress_ifindex):
8627                         return false;
8628                 }
8629         }
8630
8631         if (type == BPF_WRITE) {
8632                 if (bpf_prog_is_dev_bound(prog->aux)) {
8633                         switch (off) {
8634                         case offsetof(struct xdp_md, rx_queue_index):
8635                                 return __is_valid_xdp_access(off, size);
8636                         }
8637                 }
8638                 return false;
8639         }
8640
8641         switch (off) {
8642         case offsetof(struct xdp_md, data):
8643                 info->reg_type = PTR_TO_PACKET;
8644                 break;
8645         case offsetof(struct xdp_md, data_meta):
8646                 info->reg_type = PTR_TO_PACKET_META;
8647                 break;
8648         case offsetof(struct xdp_md, data_end):
8649                 info->reg_type = PTR_TO_PACKET_END;
8650                 break;
8651         }
8652
8653         return __is_valid_xdp_access(off, size);
8654 }
8655
8656 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act)
8657 {
8658         const u32 act_max = XDP_REDIRECT;
8659
8660         pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n",
8661                      act > act_max ? "Illegal" : "Driver unsupported",
8662                      act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A");
8663 }
8664 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
8665
8666 static bool sock_addr_is_valid_access(int off, int size,
8667                                       enum bpf_access_type type,
8668                                       const struct bpf_prog *prog,
8669                                       struct bpf_insn_access_aux *info)
8670 {
8671         const int size_default = sizeof(__u32);
8672
8673         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
8674                 return false;
8675         if (off % size != 0)
8676                 return false;
8677
8678         /* Disallow access to IPv6 fields from IPv4 contex and vise
8679          * versa.
8680          */
8681         switch (off) {
8682         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8683                 switch (prog->expected_attach_type) {
8684                 case BPF_CGROUP_INET4_BIND:
8685                 case BPF_CGROUP_INET4_CONNECT:
8686                 case BPF_CGROUP_INET4_GETPEERNAME:
8687                 case BPF_CGROUP_INET4_GETSOCKNAME:
8688                 case BPF_CGROUP_UDP4_SENDMSG:
8689                 case BPF_CGROUP_UDP4_RECVMSG:
8690                         break;
8691                 default:
8692                         return false;
8693                 }
8694                 break;
8695         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8696                 switch (prog->expected_attach_type) {
8697                 case BPF_CGROUP_INET6_BIND:
8698                 case BPF_CGROUP_INET6_CONNECT:
8699                 case BPF_CGROUP_INET6_GETPEERNAME:
8700                 case BPF_CGROUP_INET6_GETSOCKNAME:
8701                 case BPF_CGROUP_UDP6_SENDMSG:
8702                 case BPF_CGROUP_UDP6_RECVMSG:
8703                         break;
8704                 default:
8705                         return false;
8706                 }
8707                 break;
8708         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8709                 switch (prog->expected_attach_type) {
8710                 case BPF_CGROUP_UDP4_SENDMSG:
8711                         break;
8712                 default:
8713                         return false;
8714                 }
8715                 break;
8716         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8717                                 msg_src_ip6[3]):
8718                 switch (prog->expected_attach_type) {
8719                 case BPF_CGROUP_UDP6_SENDMSG:
8720                         break;
8721                 default:
8722                         return false;
8723                 }
8724                 break;
8725         }
8726
8727         switch (off) {
8728         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
8729         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
8730         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
8731         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
8732                                 msg_src_ip6[3]):
8733         case bpf_ctx_range(struct bpf_sock_addr, user_port):
8734                 if (type == BPF_READ) {
8735                         bpf_ctx_record_field_size(info, size_default);
8736
8737                         if (bpf_ctx_wide_access_ok(off, size,
8738                                                    struct bpf_sock_addr,
8739                                                    user_ip6))
8740                                 return true;
8741
8742                         if (bpf_ctx_wide_access_ok(off, size,
8743                                                    struct bpf_sock_addr,
8744                                                    msg_src_ip6))
8745                                 return true;
8746
8747                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
8748                                 return false;
8749                 } else {
8750                         if (bpf_ctx_wide_access_ok(off, size,
8751                                                    struct bpf_sock_addr,
8752                                                    user_ip6))
8753                                 return true;
8754
8755                         if (bpf_ctx_wide_access_ok(off, size,
8756                                                    struct bpf_sock_addr,
8757                                                    msg_src_ip6))
8758                                 return true;
8759
8760                         if (size != size_default)
8761                                 return false;
8762                 }
8763                 break;
8764         case offsetof(struct bpf_sock_addr, sk):
8765                 if (type != BPF_READ)
8766                         return false;
8767                 if (size != sizeof(__u64))
8768                         return false;
8769                 info->reg_type = PTR_TO_SOCKET;
8770                 break;
8771         default:
8772                 if (type == BPF_READ) {
8773                         if (size != size_default)
8774                                 return false;
8775                 } else {
8776                         return false;
8777                 }
8778         }
8779
8780         return true;
8781 }
8782
8783 static bool sock_ops_is_valid_access(int off, int size,
8784                                      enum bpf_access_type type,
8785                                      const struct bpf_prog *prog,
8786                                      struct bpf_insn_access_aux *info)
8787 {
8788         const int size_default = sizeof(__u32);
8789
8790         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
8791                 return false;
8792
8793         /* The verifier guarantees that size > 0. */
8794         if (off % size != 0)
8795                 return false;
8796
8797         if (type == BPF_WRITE) {
8798                 switch (off) {
8799                 case offsetof(struct bpf_sock_ops, reply):
8800                 case offsetof(struct bpf_sock_ops, sk_txhash):
8801                         if (size != size_default)
8802                                 return false;
8803                         break;
8804                 default:
8805                         return false;
8806                 }
8807         } else {
8808                 switch (off) {
8809                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
8810                                         bytes_acked):
8811                         if (size != sizeof(__u64))
8812                                 return false;
8813                         break;
8814                 case offsetof(struct bpf_sock_ops, sk):
8815                         if (size != sizeof(__u64))
8816                                 return false;
8817                         info->reg_type = PTR_TO_SOCKET_OR_NULL;
8818                         break;
8819                 case offsetof(struct bpf_sock_ops, skb_data):
8820                         if (size != sizeof(__u64))
8821                                 return false;
8822                         info->reg_type = PTR_TO_PACKET;
8823                         break;
8824                 case offsetof(struct bpf_sock_ops, skb_data_end):
8825                         if (size != sizeof(__u64))
8826                                 return false;
8827                         info->reg_type = PTR_TO_PACKET_END;
8828                         break;
8829                 case offsetof(struct bpf_sock_ops, skb_tcp_flags):
8830                         bpf_ctx_record_field_size(info, size_default);
8831                         return bpf_ctx_narrow_access_ok(off, size,
8832                                                         size_default);
8833                 default:
8834                         if (size != size_default)
8835                                 return false;
8836                         break;
8837                 }
8838         }
8839
8840         return true;
8841 }
8842
8843 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
8844                            const struct bpf_prog *prog)
8845 {
8846         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
8847 }
8848
8849 static bool sk_skb_is_valid_access(int off, int size,
8850                                    enum bpf_access_type type,
8851                                    const struct bpf_prog *prog,
8852                                    struct bpf_insn_access_aux *info)
8853 {
8854         switch (off) {
8855         case bpf_ctx_range(struct __sk_buff, tc_classid):
8856         case bpf_ctx_range(struct __sk_buff, data_meta):
8857         case bpf_ctx_range(struct __sk_buff, tstamp):
8858         case bpf_ctx_range(struct __sk_buff, wire_len):
8859         case bpf_ctx_range(struct __sk_buff, hwtstamp):
8860                 return false;
8861         }
8862
8863         if (type == BPF_WRITE) {
8864                 switch (off) {
8865                 case bpf_ctx_range(struct __sk_buff, tc_index):
8866                 case bpf_ctx_range(struct __sk_buff, priority):
8867                         break;
8868                 default:
8869                         return false;
8870                 }
8871         }
8872
8873         switch (off) {
8874         case bpf_ctx_range(struct __sk_buff, mark):
8875                 return false;
8876         case bpf_ctx_range(struct __sk_buff, data):
8877                 info->reg_type = PTR_TO_PACKET;
8878                 break;
8879         case bpf_ctx_range(struct __sk_buff, data_end):
8880                 info->reg_type = PTR_TO_PACKET_END;
8881                 break;
8882         }
8883
8884         return bpf_skb_is_valid_access(off, size, type, prog, info);
8885 }
8886
8887 static bool sk_msg_is_valid_access(int off, int size,
8888                                    enum bpf_access_type type,
8889                                    const struct bpf_prog *prog,
8890                                    struct bpf_insn_access_aux *info)
8891 {
8892         if (type == BPF_WRITE)
8893                 return false;
8894
8895         if (off % size != 0)
8896                 return false;
8897
8898         switch (off) {
8899         case offsetof(struct sk_msg_md, data):
8900                 info->reg_type = PTR_TO_PACKET;
8901                 if (size != sizeof(__u64))
8902                         return false;
8903                 break;
8904         case offsetof(struct sk_msg_md, data_end):
8905                 info->reg_type = PTR_TO_PACKET_END;
8906                 if (size != sizeof(__u64))
8907                         return false;
8908                 break;
8909         case offsetof(struct sk_msg_md, sk):
8910                 if (size != sizeof(__u64))
8911                         return false;
8912                 info->reg_type = PTR_TO_SOCKET;
8913                 break;
8914         case bpf_ctx_range(struct sk_msg_md, family):
8915         case bpf_ctx_range(struct sk_msg_md, remote_ip4):
8916         case bpf_ctx_range(struct sk_msg_md, local_ip4):
8917         case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]):
8918         case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]):
8919         case bpf_ctx_range(struct sk_msg_md, remote_port):
8920         case bpf_ctx_range(struct sk_msg_md, local_port):
8921         case bpf_ctx_range(struct sk_msg_md, size):
8922                 if (size != sizeof(__u32))
8923                         return false;
8924                 break;
8925         default:
8926                 return false;
8927         }
8928         return true;
8929 }
8930
8931 static bool flow_dissector_is_valid_access(int off, int size,
8932                                            enum bpf_access_type type,
8933                                            const struct bpf_prog *prog,
8934                                            struct bpf_insn_access_aux *info)
8935 {
8936         const int size_default = sizeof(__u32);
8937
8938         if (off < 0 || off >= sizeof(struct __sk_buff))
8939                 return false;
8940
8941         if (type == BPF_WRITE)
8942                 return false;
8943
8944         switch (off) {
8945         case bpf_ctx_range(struct __sk_buff, data):
8946                 if (size != size_default)
8947                         return false;
8948                 info->reg_type = PTR_TO_PACKET;
8949                 return true;
8950         case bpf_ctx_range(struct __sk_buff, data_end):
8951                 if (size != size_default)
8952                         return false;
8953                 info->reg_type = PTR_TO_PACKET_END;
8954                 return true;
8955         case bpf_ctx_range_ptr(struct __sk_buff, flow_keys):
8956                 if (size != sizeof(__u64))
8957                         return false;
8958                 info->reg_type = PTR_TO_FLOW_KEYS;
8959                 return true;
8960         default:
8961                 return false;
8962         }
8963 }
8964
8965 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type,
8966                                              const struct bpf_insn *si,
8967                                              struct bpf_insn *insn_buf,
8968                                              struct bpf_prog *prog,
8969                                              u32 *target_size)
8970
8971 {
8972         struct bpf_insn *insn = insn_buf;
8973
8974         switch (si->off) {
8975         case offsetof(struct __sk_buff, data):
8976                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data),
8977                                       si->dst_reg, si->src_reg,
8978                                       offsetof(struct bpf_flow_dissector, data));
8979                 break;
8980
8981         case offsetof(struct __sk_buff, data_end):
8982                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end),
8983                                       si->dst_reg, si->src_reg,
8984                                       offsetof(struct bpf_flow_dissector, data_end));
8985                 break;
8986
8987         case offsetof(struct __sk_buff, flow_keys):
8988                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys),
8989                                       si->dst_reg, si->src_reg,
8990                                       offsetof(struct bpf_flow_dissector, flow_keys));
8991                 break;
8992         }
8993
8994         return insn - insn_buf;
8995 }
8996
8997 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si,
8998                                                      struct bpf_insn *insn)
8999 {
9000         __u8 value_reg = si->dst_reg;
9001         __u8 skb_reg = si->src_reg;
9002         /* AX is needed because src_reg and dst_reg could be the same */
9003         __u8 tmp_reg = BPF_REG_AX;
9004
9005         *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg,
9006                               PKT_VLAN_PRESENT_OFFSET);
9007         *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg,
9008                                 SKB_MONO_DELIVERY_TIME_MASK, 2);
9009         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC);
9010         *insn++ = BPF_JMP_A(1);
9011         *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO);
9012
9013         return insn;
9014 }
9015
9016 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si,
9017                                                   struct bpf_insn *insn)
9018 {
9019         /* si->dst_reg = skb_shinfo(SKB); */
9020 #ifdef NET_SKBUFF_DATA_USES_OFFSET
9021         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9022                               BPF_REG_AX, si->src_reg,
9023                               offsetof(struct sk_buff, end));
9024         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head),
9025                               si->dst_reg, si->src_reg,
9026                               offsetof(struct sk_buff, head));
9027         *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX);
9028 #else
9029         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end),
9030                               si->dst_reg, si->src_reg,
9031                               offsetof(struct sk_buff, end));
9032 #endif
9033
9034         return insn;
9035 }
9036
9037 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog,
9038                                                 const struct bpf_insn *si,
9039                                                 struct bpf_insn *insn)
9040 {
9041         __u8 value_reg = si->dst_reg;
9042         __u8 skb_reg = si->src_reg;
9043
9044 #ifdef CONFIG_NET_CLS_ACT
9045         /* If the tstamp_type is read,
9046          * the bpf prog is aware the tstamp could have delivery time.
9047          * Thus, read skb->tstamp as is if tstamp_type_access is true.
9048          */
9049         if (!prog->tstamp_type_access) {
9050                 /* AX is needed because src_reg and dst_reg could be the same */
9051                 __u8 tmp_reg = BPF_REG_AX;
9052
9053                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9054                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg,
9055                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK);
9056                 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg,
9057                                         TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2);
9058                 /* skb->tc_at_ingress && skb->mono_delivery_time,
9059                  * read 0 as the (rcv) timestamp.
9060                  */
9061                 *insn++ = BPF_MOV64_IMM(value_reg, 0);
9062                 *insn++ = BPF_JMP_A(1);
9063         }
9064 #endif
9065
9066         *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg,
9067                               offsetof(struct sk_buff, tstamp));
9068         return insn;
9069 }
9070
9071 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog,
9072                                                  const struct bpf_insn *si,
9073                                                  struct bpf_insn *insn)
9074 {
9075         __u8 value_reg = si->src_reg;
9076         __u8 skb_reg = si->dst_reg;
9077
9078 #ifdef CONFIG_NET_CLS_ACT
9079         /* If the tstamp_type is read,
9080          * the bpf prog is aware the tstamp could have delivery time.
9081          * Thus, write skb->tstamp as is if tstamp_type_access is true.
9082          * Otherwise, writing at ingress will have to clear the
9083          * mono_delivery_time bit also.
9084          */
9085         if (!prog->tstamp_type_access) {
9086                 __u8 tmp_reg = BPF_REG_AX;
9087
9088                 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET);
9089                 /* Writing __sk_buff->tstamp as ingress, goto <clear> */
9090                 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1);
9091                 /* goto <store> */
9092                 *insn++ = BPF_JMP_A(2);
9093                 /* <clear>: mono_delivery_time */
9094                 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK);
9095                 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET);
9096         }
9097 #endif
9098
9099         /* <store>: skb->tstamp = tstamp */
9100         *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg,
9101                               offsetof(struct sk_buff, tstamp));
9102         return insn;
9103 }
9104
9105 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
9106                                   const struct bpf_insn *si,
9107                                   struct bpf_insn *insn_buf,
9108                                   struct bpf_prog *prog, u32 *target_size)
9109 {
9110         struct bpf_insn *insn = insn_buf;
9111         int off;
9112
9113         switch (si->off) {
9114         case offsetof(struct __sk_buff, len):
9115                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9116                                       bpf_target_off(struct sk_buff, len, 4,
9117                                                      target_size));
9118                 break;
9119
9120         case offsetof(struct __sk_buff, protocol):
9121                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9122                                       bpf_target_off(struct sk_buff, protocol, 2,
9123                                                      target_size));
9124                 break;
9125
9126         case offsetof(struct __sk_buff, vlan_proto):
9127                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9128                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
9129                                                      target_size));
9130                 break;
9131
9132         case offsetof(struct __sk_buff, priority):
9133                 if (type == BPF_WRITE)
9134                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9135                                               bpf_target_off(struct sk_buff, priority, 4,
9136                                                              target_size));
9137                 else
9138                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9139                                               bpf_target_off(struct sk_buff, priority, 4,
9140                                                              target_size));
9141                 break;
9142
9143         case offsetof(struct __sk_buff, ingress_ifindex):
9144                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9145                                       bpf_target_off(struct sk_buff, skb_iif, 4,
9146                                                      target_size));
9147                 break;
9148
9149         case offsetof(struct __sk_buff, ifindex):
9150                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9151                                       si->dst_reg, si->src_reg,
9152                                       offsetof(struct sk_buff, dev));
9153                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
9154                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9155                                       bpf_target_off(struct net_device, ifindex, 4,
9156                                                      target_size));
9157                 break;
9158
9159         case offsetof(struct __sk_buff, hash):
9160                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9161                                       bpf_target_off(struct sk_buff, hash, 4,
9162                                                      target_size));
9163                 break;
9164
9165         case offsetof(struct __sk_buff, mark):
9166                 if (type == BPF_WRITE)
9167                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9168                                               bpf_target_off(struct sk_buff, mark, 4,
9169                                                              target_size));
9170                 else
9171                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9172                                               bpf_target_off(struct sk_buff, mark, 4,
9173                                                              target_size));
9174                 break;
9175
9176         case offsetof(struct __sk_buff, pkt_type):
9177                 *target_size = 1;
9178                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9179                                       PKT_TYPE_OFFSET);
9180                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
9181 #ifdef __BIG_ENDIAN_BITFIELD
9182                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
9183 #endif
9184                 break;
9185
9186         case offsetof(struct __sk_buff, queue_mapping):
9187                 if (type == BPF_WRITE) {
9188                         *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1);
9189                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9190                                               bpf_target_off(struct sk_buff,
9191                                                              queue_mapping,
9192                                                              2, target_size));
9193                 } else {
9194                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9195                                               bpf_target_off(struct sk_buff,
9196                                                              queue_mapping,
9197                                                              2, target_size));
9198                 }
9199                 break;
9200
9201         case offsetof(struct __sk_buff, vlan_present):
9202                 *target_size = 1;
9203                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
9204                                       PKT_VLAN_PRESENT_OFFSET);
9205                 if (PKT_VLAN_PRESENT_BIT)
9206                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT);
9207                 if (PKT_VLAN_PRESENT_BIT < 7)
9208                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
9209                 break;
9210
9211         case offsetof(struct __sk_buff, vlan_tci):
9212                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9213                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
9214                                                      target_size));
9215                 break;
9216
9217         case offsetof(struct __sk_buff, cb[0]) ...
9218              offsetofend(struct __sk_buff, cb[4]) - 1:
9219                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20);
9220                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
9221                               offsetof(struct qdisc_skb_cb, data)) %
9222                              sizeof(__u64));
9223
9224                 prog->cb_access = 1;
9225                 off  = si->off;
9226                 off -= offsetof(struct __sk_buff, cb[0]);
9227                 off += offsetof(struct sk_buff, cb);
9228                 off += offsetof(struct qdisc_skb_cb, data);
9229                 if (type == BPF_WRITE)
9230                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
9231                                               si->src_reg, off);
9232                 else
9233                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
9234                                               si->src_reg, off);
9235                 break;
9236
9237         case offsetof(struct __sk_buff, tc_classid):
9238                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2);
9239
9240                 off  = si->off;
9241                 off -= offsetof(struct __sk_buff, tc_classid);
9242                 off += offsetof(struct sk_buff, cb);
9243                 off += offsetof(struct qdisc_skb_cb, tc_classid);
9244                 *target_size = 2;
9245                 if (type == BPF_WRITE)
9246                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
9247                                               si->src_reg, off);
9248                 else
9249                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
9250                                               si->src_reg, off);
9251                 break;
9252
9253         case offsetof(struct __sk_buff, data):
9254                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
9255                                       si->dst_reg, si->src_reg,
9256                                       offsetof(struct sk_buff, data));
9257                 break;
9258
9259         case offsetof(struct __sk_buff, data_meta):
9260                 off  = si->off;
9261                 off -= offsetof(struct __sk_buff, data_meta);
9262                 off += offsetof(struct sk_buff, cb);
9263                 off += offsetof(struct bpf_skb_data_end, data_meta);
9264                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9265                                       si->src_reg, off);
9266                 break;
9267
9268         case offsetof(struct __sk_buff, data_end):
9269                 off  = si->off;
9270                 off -= offsetof(struct __sk_buff, data_end);
9271                 off += offsetof(struct sk_buff, cb);
9272                 off += offsetof(struct bpf_skb_data_end, data_end);
9273                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
9274                                       si->src_reg, off);
9275                 break;
9276
9277         case offsetof(struct __sk_buff, tc_index):
9278 #ifdef CONFIG_NET_SCHED
9279                 if (type == BPF_WRITE)
9280                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
9281                                               bpf_target_off(struct sk_buff, tc_index, 2,
9282                                                              target_size));
9283                 else
9284                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
9285                                               bpf_target_off(struct sk_buff, tc_index, 2,
9286                                                              target_size));
9287 #else
9288                 *target_size = 2;
9289                 if (type == BPF_WRITE)
9290                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
9291                 else
9292                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9293 #endif
9294                 break;
9295
9296         case offsetof(struct __sk_buff, napi_id):
9297 #if defined(CONFIG_NET_RX_BUSY_POLL)
9298                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9299                                       bpf_target_off(struct sk_buff, napi_id, 4,
9300                                                      target_size));
9301                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
9302                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9303 #else
9304                 *target_size = 4;
9305                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
9306 #endif
9307                 break;
9308         case offsetof(struct __sk_buff, family):
9309                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
9310
9311                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9312                                       si->dst_reg, si->src_reg,
9313                                       offsetof(struct sk_buff, sk));
9314                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9315                                       bpf_target_off(struct sock_common,
9316                                                      skc_family,
9317                                                      2, target_size));
9318                 break;
9319         case offsetof(struct __sk_buff, remote_ip4):
9320                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
9321
9322                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9323                                       si->dst_reg, si->src_reg,
9324                                       offsetof(struct sk_buff, sk));
9325                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9326                                       bpf_target_off(struct sock_common,
9327                                                      skc_daddr,
9328                                                      4, target_size));
9329                 break;
9330         case offsetof(struct __sk_buff, local_ip4):
9331                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9332                                           skc_rcv_saddr) != 4);
9333
9334                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9335                                       si->dst_reg, si->src_reg,
9336                                       offsetof(struct sk_buff, sk));
9337                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9338                                       bpf_target_off(struct sock_common,
9339                                                      skc_rcv_saddr,
9340                                                      4, target_size));
9341                 break;
9342         case offsetof(struct __sk_buff, remote_ip6[0]) ...
9343              offsetof(struct __sk_buff, remote_ip6[3]):
9344 #if IS_ENABLED(CONFIG_IPV6)
9345                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9346                                           skc_v6_daddr.s6_addr32[0]) != 4);
9347
9348                 off = si->off;
9349                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
9350
9351                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9352                                       si->dst_reg, si->src_reg,
9353                                       offsetof(struct sk_buff, sk));
9354                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9355                                       offsetof(struct sock_common,
9356                                                skc_v6_daddr.s6_addr32[0]) +
9357                                       off);
9358 #else
9359                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9360 #endif
9361                 break;
9362         case offsetof(struct __sk_buff, local_ip6[0]) ...
9363              offsetof(struct __sk_buff, local_ip6[3]):
9364 #if IS_ENABLED(CONFIG_IPV6)
9365                 BUILD_BUG_ON(sizeof_field(struct sock_common,
9366                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
9367
9368                 off = si->off;
9369                 off -= offsetof(struct __sk_buff, local_ip6[0]);
9370
9371                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9372                                       si->dst_reg, si->src_reg,
9373                                       offsetof(struct sk_buff, sk));
9374                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9375                                       offsetof(struct sock_common,
9376                                                skc_v6_rcv_saddr.s6_addr32[0]) +
9377                                       off);
9378 #else
9379                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9380 #endif
9381                 break;
9382
9383         case offsetof(struct __sk_buff, remote_port):
9384                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
9385
9386                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9387                                       si->dst_reg, si->src_reg,
9388                                       offsetof(struct sk_buff, sk));
9389                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9390                                       bpf_target_off(struct sock_common,
9391                                                      skc_dport,
9392                                                      2, target_size));
9393 #ifndef __BIG_ENDIAN_BITFIELD
9394                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
9395 #endif
9396                 break;
9397
9398         case offsetof(struct __sk_buff, local_port):
9399                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
9400
9401                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9402                                       si->dst_reg, si->src_reg,
9403                                       offsetof(struct sk_buff, sk));
9404                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
9405                                       bpf_target_off(struct sock_common,
9406                                                      skc_num, 2, target_size));
9407                 break;
9408
9409         case offsetof(struct __sk_buff, tstamp):
9410                 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8);
9411
9412                 if (type == BPF_WRITE)
9413                         insn = bpf_convert_tstamp_write(prog, si, insn);
9414                 else
9415                         insn = bpf_convert_tstamp_read(prog, si, insn);
9416                 break;
9417
9418         case offsetof(struct __sk_buff, tstamp_type):
9419                 insn = bpf_convert_tstamp_type_read(si, insn);
9420                 break;
9421
9422         case offsetof(struct __sk_buff, gso_segs):
9423                 insn = bpf_convert_shinfo_access(si, insn);
9424                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs),
9425                                       si->dst_reg, si->dst_reg,
9426                                       bpf_target_off(struct skb_shared_info,
9427                                                      gso_segs, 2,
9428                                                      target_size));
9429                 break;
9430         case offsetof(struct __sk_buff, gso_size):
9431                 insn = bpf_convert_shinfo_access(si, insn);
9432                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size),
9433                                       si->dst_reg, si->dst_reg,
9434                                       bpf_target_off(struct skb_shared_info,
9435                                                      gso_size, 2,
9436                                                      target_size));
9437                 break;
9438         case offsetof(struct __sk_buff, wire_len):
9439                 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4);
9440
9441                 off = si->off;
9442                 off -= offsetof(struct __sk_buff, wire_len);
9443                 off += offsetof(struct sk_buff, cb);
9444                 off += offsetof(struct qdisc_skb_cb, pkt_len);
9445                 *target_size = 4;
9446                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off);
9447                 break;
9448
9449         case offsetof(struct __sk_buff, sk):
9450                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
9451                                       si->dst_reg, si->src_reg,
9452                                       offsetof(struct sk_buff, sk));
9453                 break;
9454         case offsetof(struct __sk_buff, hwtstamp):
9455                 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8);
9456                 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0);
9457
9458                 insn = bpf_convert_shinfo_access(si, insn);
9459                 *insn++ = BPF_LDX_MEM(BPF_DW,
9460                                       si->dst_reg, si->dst_reg,
9461                                       bpf_target_off(struct skb_shared_info,
9462                                                      hwtstamps, 8,
9463                                                      target_size));
9464                 break;
9465         }
9466
9467         return insn - insn_buf;
9468 }
9469
9470 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
9471                                 const struct bpf_insn *si,
9472                                 struct bpf_insn *insn_buf,
9473                                 struct bpf_prog *prog, u32 *target_size)
9474 {
9475         struct bpf_insn *insn = insn_buf;
9476         int off;
9477
9478         switch (si->off) {
9479         case offsetof(struct bpf_sock, bound_dev_if):
9480                 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4);
9481
9482                 if (type == BPF_WRITE)
9483                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9484                                         offsetof(struct sock, sk_bound_dev_if));
9485                 else
9486                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9487                                       offsetof(struct sock, sk_bound_dev_if));
9488                 break;
9489
9490         case offsetof(struct bpf_sock, mark):
9491                 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4);
9492
9493                 if (type == BPF_WRITE)
9494                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9495                                         offsetof(struct sock, sk_mark));
9496                 else
9497                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9498                                       offsetof(struct sock, sk_mark));
9499                 break;
9500
9501         case offsetof(struct bpf_sock, priority):
9502                 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4);
9503
9504                 if (type == BPF_WRITE)
9505                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
9506                                         offsetof(struct sock, sk_priority));
9507                 else
9508                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
9509                                       offsetof(struct sock, sk_priority));
9510                 break;
9511
9512         case offsetof(struct bpf_sock, family):
9513                 *insn++ = BPF_LDX_MEM(
9514                         BPF_FIELD_SIZEOF(struct sock_common, skc_family),
9515                         si->dst_reg, si->src_reg,
9516                         bpf_target_off(struct sock_common,
9517                                        skc_family,
9518                                        sizeof_field(struct sock_common,
9519                                                     skc_family),
9520                                        target_size));
9521                 break;
9522
9523         case offsetof(struct bpf_sock, type):
9524                 *insn++ = BPF_LDX_MEM(
9525                         BPF_FIELD_SIZEOF(struct sock, sk_type),
9526                         si->dst_reg, si->src_reg,
9527                         bpf_target_off(struct sock, sk_type,
9528                                        sizeof_field(struct sock, sk_type),
9529                                        target_size));
9530                 break;
9531
9532         case offsetof(struct bpf_sock, protocol):
9533                 *insn++ = BPF_LDX_MEM(
9534                         BPF_FIELD_SIZEOF(struct sock, sk_protocol),
9535                         si->dst_reg, si->src_reg,
9536                         bpf_target_off(struct sock, sk_protocol,
9537                                        sizeof_field(struct sock, sk_protocol),
9538                                        target_size));
9539                 break;
9540
9541         case offsetof(struct bpf_sock, src_ip4):
9542                 *insn++ = BPF_LDX_MEM(
9543                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9544                         bpf_target_off(struct sock_common, skc_rcv_saddr,
9545                                        sizeof_field(struct sock_common,
9546                                                     skc_rcv_saddr),
9547                                        target_size));
9548                 break;
9549
9550         case offsetof(struct bpf_sock, dst_ip4):
9551                 *insn++ = BPF_LDX_MEM(
9552                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9553                         bpf_target_off(struct sock_common, skc_daddr,
9554                                        sizeof_field(struct sock_common,
9555                                                     skc_daddr),
9556                                        target_size));
9557                 break;
9558
9559         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
9560 #if IS_ENABLED(CONFIG_IPV6)
9561                 off = si->off;
9562                 off -= offsetof(struct bpf_sock, src_ip6[0]);
9563                 *insn++ = BPF_LDX_MEM(
9564                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9565                         bpf_target_off(
9566                                 struct sock_common,
9567                                 skc_v6_rcv_saddr.s6_addr32[0],
9568                                 sizeof_field(struct sock_common,
9569                                              skc_v6_rcv_saddr.s6_addr32[0]),
9570                                 target_size) + off);
9571 #else
9572                 (void)off;
9573                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9574 #endif
9575                 break;
9576
9577         case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]):
9578 #if IS_ENABLED(CONFIG_IPV6)
9579                 off = si->off;
9580                 off -= offsetof(struct bpf_sock, dst_ip6[0]);
9581                 *insn++ = BPF_LDX_MEM(
9582                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
9583                         bpf_target_off(struct sock_common,
9584                                        skc_v6_daddr.s6_addr32[0],
9585                                        sizeof_field(struct sock_common,
9586                                                     skc_v6_daddr.s6_addr32[0]),
9587                                        target_size) + off);
9588 #else
9589                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
9590                 *target_size = 4;
9591 #endif
9592                 break;
9593
9594         case offsetof(struct bpf_sock, src_port):
9595                 *insn++ = BPF_LDX_MEM(
9596                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
9597                         si->dst_reg, si->src_reg,
9598                         bpf_target_off(struct sock_common, skc_num,
9599                                        sizeof_field(struct sock_common,
9600                                                     skc_num),
9601                                        target_size));
9602                 break;
9603
9604         case offsetof(struct bpf_sock, dst_port):
9605                 *insn++ = BPF_LDX_MEM(
9606                         BPF_FIELD_SIZEOF(struct sock_common, skc_dport),
9607                         si->dst_reg, si->src_reg,
9608                         bpf_target_off(struct sock_common, skc_dport,
9609                                        sizeof_field(struct sock_common,
9610                                                     skc_dport),
9611                                        target_size));
9612                 break;
9613
9614         case offsetof(struct bpf_sock, state):
9615                 *insn++ = BPF_LDX_MEM(
9616                         BPF_FIELD_SIZEOF(struct sock_common, skc_state),
9617                         si->dst_reg, si->src_reg,
9618                         bpf_target_off(struct sock_common, skc_state,
9619                                        sizeof_field(struct sock_common,
9620                                                     skc_state),
9621                                        target_size));
9622                 break;
9623         case offsetof(struct bpf_sock, rx_queue_mapping):
9624 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
9625                 *insn++ = BPF_LDX_MEM(
9626                         BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping),
9627                         si->dst_reg, si->src_reg,
9628                         bpf_target_off(struct sock, sk_rx_queue_mapping,
9629                                        sizeof_field(struct sock,
9630                                                     sk_rx_queue_mapping),
9631                                        target_size));
9632                 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING,
9633                                       1);
9634                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9635 #else
9636                 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1);
9637                 *target_size = 2;
9638 #endif
9639                 break;
9640         }
9641
9642         return insn - insn_buf;
9643 }
9644
9645 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
9646                                          const struct bpf_insn *si,
9647                                          struct bpf_insn *insn_buf,
9648                                          struct bpf_prog *prog, u32 *target_size)
9649 {
9650         struct bpf_insn *insn = insn_buf;
9651
9652         switch (si->off) {
9653         case offsetof(struct __sk_buff, ifindex):
9654                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
9655                                       si->dst_reg, si->src_reg,
9656                                       offsetof(struct sk_buff, dev));
9657                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9658                                       bpf_target_off(struct net_device, ifindex, 4,
9659                                                      target_size));
9660                 break;
9661         default:
9662                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
9663                                               target_size);
9664         }
9665
9666         return insn - insn_buf;
9667 }
9668
9669 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
9670                                   const struct bpf_insn *si,
9671                                   struct bpf_insn *insn_buf,
9672                                   struct bpf_prog *prog, u32 *target_size)
9673 {
9674         struct bpf_insn *insn = insn_buf;
9675
9676         switch (si->off) {
9677         case offsetof(struct xdp_md, data):
9678                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
9679                                       si->dst_reg, si->src_reg,
9680                                       offsetof(struct xdp_buff, data));
9681                 break;
9682         case offsetof(struct xdp_md, data_meta):
9683                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
9684                                       si->dst_reg, si->src_reg,
9685                                       offsetof(struct xdp_buff, data_meta));
9686                 break;
9687         case offsetof(struct xdp_md, data_end):
9688                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
9689                                       si->dst_reg, si->src_reg,
9690                                       offsetof(struct xdp_buff, data_end));
9691                 break;
9692         case offsetof(struct xdp_md, ingress_ifindex):
9693                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9694                                       si->dst_reg, si->src_reg,
9695                                       offsetof(struct xdp_buff, rxq));
9696                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
9697                                       si->dst_reg, si->dst_reg,
9698                                       offsetof(struct xdp_rxq_info, dev));
9699                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9700                                       offsetof(struct net_device, ifindex));
9701                 break;
9702         case offsetof(struct xdp_md, rx_queue_index):
9703                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
9704                                       si->dst_reg, si->src_reg,
9705                                       offsetof(struct xdp_buff, rxq));
9706                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9707                                       offsetof(struct xdp_rxq_info,
9708                                                queue_index));
9709                 break;
9710         case offsetof(struct xdp_md, egress_ifindex):
9711                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq),
9712                                       si->dst_reg, si->src_reg,
9713                                       offsetof(struct xdp_buff, txq));
9714                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev),
9715                                       si->dst_reg, si->dst_reg,
9716                                       offsetof(struct xdp_txq_info, dev));
9717                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
9718                                       offsetof(struct net_device, ifindex));
9719                 break;
9720         }
9721
9722         return insn - insn_buf;
9723 }
9724
9725 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
9726  * context Structure, F is Field in context structure that contains a pointer
9727  * to Nested Structure of type NS that has the field NF.
9728  *
9729  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
9730  * sure that SIZE is not greater than actual size of S.F.NF.
9731  *
9732  * If offset OFF is provided, the load happens from that offset relative to
9733  * offset of NF.
9734  */
9735 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
9736         do {                                                                   \
9737                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
9738                                       si->src_reg, offsetof(S, F));            \
9739                 *insn++ = BPF_LDX_MEM(                                         \
9740                         SIZE, si->dst_reg, si->dst_reg,                        \
9741                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9742                                        target_size)                            \
9743                                 + OFF);                                        \
9744         } while (0)
9745
9746 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
9747         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
9748                                              BPF_FIELD_SIZEOF(NS, NF), 0)
9749
9750 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
9751  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
9752  *
9753  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
9754  * "register" since two registers available in convert_ctx_access are not
9755  * enough: we can't override neither SRC, since it contains value to store, nor
9756  * DST since it contains pointer to context that may be used by later
9757  * instructions. But we need a temporary place to save pointer to nested
9758  * structure whose field we want to store to.
9759  */
9760 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF)          \
9761         do {                                                                   \
9762                 int tmp_reg = BPF_REG_9;                                       \
9763                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9764                         --tmp_reg;                                             \
9765                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
9766                         --tmp_reg;                                             \
9767                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
9768                                       offsetof(S, TF));                        \
9769                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
9770                                       si->dst_reg, offsetof(S, F));            \
9771                 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg,              \
9772                         bpf_target_off(NS, NF, sizeof_field(NS, NF),           \
9773                                        target_size)                            \
9774                                 + OFF);                                        \
9775                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
9776                                       offsetof(S, TF));                        \
9777         } while (0)
9778
9779 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
9780                                                       TF)                      \
9781         do {                                                                   \
9782                 if (type == BPF_WRITE) {                                       \
9783                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE,   \
9784                                                          OFF, TF);             \
9785                 } else {                                                       \
9786                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
9787                                 S, NS, F, NF, SIZE, OFF);  \
9788                 }                                                              \
9789         } while (0)
9790
9791 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
9792         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
9793                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
9794
9795 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
9796                                         const struct bpf_insn *si,
9797                                         struct bpf_insn *insn_buf,
9798                                         struct bpf_prog *prog, u32 *target_size)
9799 {
9800         int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port);
9801         struct bpf_insn *insn = insn_buf;
9802
9803         switch (si->off) {
9804         case offsetof(struct bpf_sock_addr, user_family):
9805                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9806                                             struct sockaddr, uaddr, sa_family);
9807                 break;
9808
9809         case offsetof(struct bpf_sock_addr, user_ip4):
9810                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9811                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
9812                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
9813                 break;
9814
9815         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
9816                 off = si->off;
9817                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
9818                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9819                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9820                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
9821                         tmp_reg);
9822                 break;
9823
9824         case offsetof(struct bpf_sock_addr, user_port):
9825                 /* To get port we need to know sa_family first and then treat
9826                  * sockaddr as either sockaddr_in or sockaddr_in6.
9827                  * Though we can simplify since port field has same offset and
9828                  * size in both structures.
9829                  * Here we check this invariant and use just one of the
9830                  * structures if it's true.
9831                  */
9832                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
9833                              offsetof(struct sockaddr_in6, sin6_port));
9834                 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) !=
9835                              sizeof_field(struct sockaddr_in6, sin6_port));
9836                 /* Account for sin6_port being smaller than user_port. */
9837                 port_size = min(port_size, BPF_LDST_BYTES(si));
9838                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9839                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
9840                         sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg);
9841                 break;
9842
9843         case offsetof(struct bpf_sock_addr, family):
9844                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9845                                             struct sock, sk, sk_family);
9846                 break;
9847
9848         case offsetof(struct bpf_sock_addr, type):
9849                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9850                                             struct sock, sk, sk_type);
9851                 break;
9852
9853         case offsetof(struct bpf_sock_addr, protocol):
9854                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
9855                                             struct sock, sk, sk_protocol);
9856                 break;
9857
9858         case offsetof(struct bpf_sock_addr, msg_src_ip4):
9859                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
9860                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9861                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
9862                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
9863                 break;
9864
9865         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
9866                                 msg_src_ip6[3]):
9867                 off = si->off;
9868                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
9869                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
9870                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
9871                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
9872                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
9873                 break;
9874         case offsetof(struct bpf_sock_addr, sk):
9875                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk),
9876                                       si->dst_reg, si->src_reg,
9877                                       offsetof(struct bpf_sock_addr_kern, sk));
9878                 break;
9879         }
9880
9881         return insn - insn_buf;
9882 }
9883
9884 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
9885                                        const struct bpf_insn *si,
9886                                        struct bpf_insn *insn_buf,
9887                                        struct bpf_prog *prog,
9888                                        u32 *target_size)
9889 {
9890         struct bpf_insn *insn = insn_buf;
9891         int off;
9892
9893 /* Helper macro for adding read access to tcp_sock or sock fields. */
9894 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9895         do {                                                                  \
9896                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2;     \
9897                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9898                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9899                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9900                         reg--;                                                \
9901                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9902                         reg--;                                                \
9903                 if (si->dst_reg == si->src_reg) {                             \
9904                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9905                                           offsetof(struct bpf_sock_ops_kern,  \
9906                                           temp));                             \
9907                         fullsock_reg = reg;                                   \
9908                         jmp += 2;                                             \
9909                 }                                                             \
9910                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9911                                                 struct bpf_sock_ops_kern,     \
9912                                                 is_fullsock),                 \
9913                                       fullsock_reg, si->src_reg,              \
9914                                       offsetof(struct bpf_sock_ops_kern,      \
9915                                                is_fullsock));                 \
9916                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9917                 if (si->dst_reg == si->src_reg)                               \
9918                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9919                                       offsetof(struct bpf_sock_ops_kern,      \
9920                                       temp));                                 \
9921                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9922                                                 struct bpf_sock_ops_kern, sk),\
9923                                       si->dst_reg, si->src_reg,               \
9924                                       offsetof(struct bpf_sock_ops_kern, sk));\
9925                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
9926                                                        OBJ_FIELD),            \
9927                                       si->dst_reg, si->dst_reg,               \
9928                                       offsetof(OBJ, OBJ_FIELD));              \
9929                 if (si->dst_reg == si->src_reg) {                             \
9930                         *insn++ = BPF_JMP_A(1);                               \
9931                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9932                                       offsetof(struct bpf_sock_ops_kern,      \
9933                                       temp));                                 \
9934                 }                                                             \
9935         } while (0)
9936
9937 #define SOCK_OPS_GET_SK()                                                             \
9938         do {                                                                  \
9939                 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1;     \
9940                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9941                         reg--;                                                \
9942                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9943                         reg--;                                                \
9944                 if (si->dst_reg == si->src_reg) {                             \
9945                         *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg,       \
9946                                           offsetof(struct bpf_sock_ops_kern,  \
9947                                           temp));                             \
9948                         fullsock_reg = reg;                                   \
9949                         jmp += 2;                                             \
9950                 }                                                             \
9951                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9952                                                 struct bpf_sock_ops_kern,     \
9953                                                 is_fullsock),                 \
9954                                       fullsock_reg, si->src_reg,              \
9955                                       offsetof(struct bpf_sock_ops_kern,      \
9956                                                is_fullsock));                 \
9957                 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp);         \
9958                 if (si->dst_reg == si->src_reg)                               \
9959                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9960                                       offsetof(struct bpf_sock_ops_kern,      \
9961                                       temp));                                 \
9962                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9963                                                 struct bpf_sock_ops_kern, sk),\
9964                                       si->dst_reg, si->src_reg,               \
9965                                       offsetof(struct bpf_sock_ops_kern, sk));\
9966                 if (si->dst_reg == si->src_reg) {                             \
9967                         *insn++ = BPF_JMP_A(1);                               \
9968                         *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg,       \
9969                                       offsetof(struct bpf_sock_ops_kern,      \
9970                                       temp));                                 \
9971                 }                                                             \
9972         } while (0)
9973
9974 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \
9975                 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock)
9976
9977 /* Helper macro for adding write access to tcp_sock or sock fields.
9978  * The macro is called with two registers, dst_reg which contains a pointer
9979  * to ctx (context) and src_reg which contains the value that should be
9980  * stored. However, we need an additional register since we cannot overwrite
9981  * dst_reg because it may be used later in the program.
9982  * Instead we "borrow" one of the other register. We first save its value
9983  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
9984  * it at the end of the macro.
9985  */
9986 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
9987         do {                                                                  \
9988                 int reg = BPF_REG_9;                                          \
9989                 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) >                   \
9990                              sizeof_field(struct bpf_sock_ops, BPF_FIELD));   \
9991                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9992                         reg--;                                                \
9993                 if (si->dst_reg == reg || si->src_reg == reg)                 \
9994                         reg--;                                                \
9995                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
9996                                       offsetof(struct bpf_sock_ops_kern,      \
9997                                                temp));                        \
9998                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
9999                                                 struct bpf_sock_ops_kern,     \
10000                                                 is_fullsock),                 \
10001                                       reg, si->dst_reg,                       \
10002                                       offsetof(struct bpf_sock_ops_kern,      \
10003                                                is_fullsock));                 \
10004                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
10005                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
10006                                                 struct bpf_sock_ops_kern, sk),\
10007                                       reg, si->dst_reg,                       \
10008                                       offsetof(struct bpf_sock_ops_kern, sk));\
10009                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
10010                                       reg, si->src_reg,                       \
10011                                       offsetof(OBJ, OBJ_FIELD));              \
10012                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
10013                                       offsetof(struct bpf_sock_ops_kern,      \
10014                                                temp));                        \
10015         } while (0)
10016
10017 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
10018         do {                                                                  \
10019                 if (TYPE == BPF_WRITE)                                        \
10020                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10021                 else                                                          \
10022                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
10023         } while (0)
10024
10025         if (insn > insn_buf)
10026                 return insn - insn_buf;
10027
10028         switch (si->off) {
10029         case offsetof(struct bpf_sock_ops, op):
10030                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10031                                                        op),
10032                                       si->dst_reg, si->src_reg,
10033                                       offsetof(struct bpf_sock_ops_kern, op));
10034                 break;
10035
10036         case offsetof(struct bpf_sock_ops, replylong[0]) ...
10037              offsetof(struct bpf_sock_ops, replylong[3]):
10038                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) !=
10039                              sizeof_field(struct bpf_sock_ops_kern, reply));
10040                 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) !=
10041                              sizeof_field(struct bpf_sock_ops_kern, replylong));
10042                 off = si->off;
10043                 off -= offsetof(struct bpf_sock_ops, replylong[0]);
10044                 off += offsetof(struct bpf_sock_ops_kern, replylong[0]);
10045                 if (type == BPF_WRITE)
10046                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
10047                                               off);
10048                 else
10049                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
10050                                               off);
10051                 break;
10052
10053         case offsetof(struct bpf_sock_ops, family):
10054                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10055
10056                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10057                                               struct bpf_sock_ops_kern, sk),
10058                                       si->dst_reg, si->src_reg,
10059                                       offsetof(struct bpf_sock_ops_kern, sk));
10060                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10061                                       offsetof(struct sock_common, skc_family));
10062                 break;
10063
10064         case offsetof(struct bpf_sock_ops, remote_ip4):
10065                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10066
10067                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10068                                                 struct bpf_sock_ops_kern, sk),
10069                                       si->dst_reg, si->src_reg,
10070                                       offsetof(struct bpf_sock_ops_kern, sk));
10071                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10072                                       offsetof(struct sock_common, skc_daddr));
10073                 break;
10074
10075         case offsetof(struct bpf_sock_ops, local_ip4):
10076                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10077                                           skc_rcv_saddr) != 4);
10078
10079                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10080                                               struct bpf_sock_ops_kern, sk),
10081                                       si->dst_reg, si->src_reg,
10082                                       offsetof(struct bpf_sock_ops_kern, sk));
10083                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10084                                       offsetof(struct sock_common,
10085                                                skc_rcv_saddr));
10086                 break;
10087
10088         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
10089              offsetof(struct bpf_sock_ops, remote_ip6[3]):
10090 #if IS_ENABLED(CONFIG_IPV6)
10091                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10092                                           skc_v6_daddr.s6_addr32[0]) != 4);
10093
10094                 off = si->off;
10095                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
10096                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10097                                                 struct bpf_sock_ops_kern, sk),
10098                                       si->dst_reg, si->src_reg,
10099                                       offsetof(struct bpf_sock_ops_kern, sk));
10100                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10101                                       offsetof(struct sock_common,
10102                                                skc_v6_daddr.s6_addr32[0]) +
10103                                       off);
10104 #else
10105                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10106 #endif
10107                 break;
10108
10109         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
10110              offsetof(struct bpf_sock_ops, local_ip6[3]):
10111 #if IS_ENABLED(CONFIG_IPV6)
10112                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10113                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10114
10115                 off = si->off;
10116                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
10117                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10118                                                 struct bpf_sock_ops_kern, sk),
10119                                       si->dst_reg, si->src_reg,
10120                                       offsetof(struct bpf_sock_ops_kern, sk));
10121                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10122                                       offsetof(struct sock_common,
10123                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10124                                       off);
10125 #else
10126                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10127 #endif
10128                 break;
10129
10130         case offsetof(struct bpf_sock_ops, remote_port):
10131                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10132
10133                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10134                                                 struct bpf_sock_ops_kern, sk),
10135                                       si->dst_reg, si->src_reg,
10136                                       offsetof(struct bpf_sock_ops_kern, sk));
10137                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10138                                       offsetof(struct sock_common, skc_dport));
10139 #ifndef __BIG_ENDIAN_BITFIELD
10140                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10141 #endif
10142                 break;
10143
10144         case offsetof(struct bpf_sock_ops, local_port):
10145                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10146
10147                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10148                                                 struct bpf_sock_ops_kern, sk),
10149                                       si->dst_reg, si->src_reg,
10150                                       offsetof(struct bpf_sock_ops_kern, sk));
10151                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10152                                       offsetof(struct sock_common, skc_num));
10153                 break;
10154
10155         case offsetof(struct bpf_sock_ops, is_fullsock):
10156                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10157                                                 struct bpf_sock_ops_kern,
10158                                                 is_fullsock),
10159                                       si->dst_reg, si->src_reg,
10160                                       offsetof(struct bpf_sock_ops_kern,
10161                                                is_fullsock));
10162                 break;
10163
10164         case offsetof(struct bpf_sock_ops, state):
10165                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
10166
10167                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10168                                                 struct bpf_sock_ops_kern, sk),
10169                                       si->dst_reg, si->src_reg,
10170                                       offsetof(struct bpf_sock_ops_kern, sk));
10171                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
10172                                       offsetof(struct sock_common, skc_state));
10173                 break;
10174
10175         case offsetof(struct bpf_sock_ops, rtt_min):
10176                 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) !=
10177                              sizeof(struct minmax));
10178                 BUILD_BUG_ON(sizeof(struct minmax) <
10179                              sizeof(struct minmax_sample));
10180
10181                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10182                                                 struct bpf_sock_ops_kern, sk),
10183                                       si->dst_reg, si->src_reg,
10184                                       offsetof(struct bpf_sock_ops_kern, sk));
10185                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10186                                       offsetof(struct tcp_sock, rtt_min) +
10187                                       sizeof_field(struct minmax_sample, t));
10188                 break;
10189
10190         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
10191                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
10192                                    struct tcp_sock);
10193                 break;
10194
10195         case offsetof(struct bpf_sock_ops, sk_txhash):
10196                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
10197                                           struct sock, type);
10198                 break;
10199         case offsetof(struct bpf_sock_ops, snd_cwnd):
10200                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd);
10201                 break;
10202         case offsetof(struct bpf_sock_ops, srtt_us):
10203                 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us);
10204                 break;
10205         case offsetof(struct bpf_sock_ops, snd_ssthresh):
10206                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh);
10207                 break;
10208         case offsetof(struct bpf_sock_ops, rcv_nxt):
10209                 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt);
10210                 break;
10211         case offsetof(struct bpf_sock_ops, snd_nxt):
10212                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt);
10213                 break;
10214         case offsetof(struct bpf_sock_ops, snd_una):
10215                 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una);
10216                 break;
10217         case offsetof(struct bpf_sock_ops, mss_cache):
10218                 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache);
10219                 break;
10220         case offsetof(struct bpf_sock_ops, ecn_flags):
10221                 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags);
10222                 break;
10223         case offsetof(struct bpf_sock_ops, rate_delivered):
10224                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered);
10225                 break;
10226         case offsetof(struct bpf_sock_ops, rate_interval_us):
10227                 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us);
10228                 break;
10229         case offsetof(struct bpf_sock_ops, packets_out):
10230                 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out);
10231                 break;
10232         case offsetof(struct bpf_sock_ops, retrans_out):
10233                 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out);
10234                 break;
10235         case offsetof(struct bpf_sock_ops, total_retrans):
10236                 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans);
10237                 break;
10238         case offsetof(struct bpf_sock_ops, segs_in):
10239                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in);
10240                 break;
10241         case offsetof(struct bpf_sock_ops, data_segs_in):
10242                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in);
10243                 break;
10244         case offsetof(struct bpf_sock_ops, segs_out):
10245                 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out);
10246                 break;
10247         case offsetof(struct bpf_sock_ops, data_segs_out):
10248                 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out);
10249                 break;
10250         case offsetof(struct bpf_sock_ops, lost_out):
10251                 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out);
10252                 break;
10253         case offsetof(struct bpf_sock_ops, sacked_out):
10254                 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out);
10255                 break;
10256         case offsetof(struct bpf_sock_ops, bytes_received):
10257                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received);
10258                 break;
10259         case offsetof(struct bpf_sock_ops, bytes_acked):
10260                 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked);
10261                 break;
10262         case offsetof(struct bpf_sock_ops, sk):
10263                 SOCK_OPS_GET_SK();
10264                 break;
10265         case offsetof(struct bpf_sock_ops, skb_data_end):
10266                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10267                                                        skb_data_end),
10268                                       si->dst_reg, si->src_reg,
10269                                       offsetof(struct bpf_sock_ops_kern,
10270                                                skb_data_end));
10271                 break;
10272         case offsetof(struct bpf_sock_ops, skb_data):
10273                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10274                                                        skb),
10275                                       si->dst_reg, si->src_reg,
10276                                       offsetof(struct bpf_sock_ops_kern,
10277                                                skb));
10278                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10279                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10280                                       si->dst_reg, si->dst_reg,
10281                                       offsetof(struct sk_buff, data));
10282                 break;
10283         case offsetof(struct bpf_sock_ops, skb_len):
10284                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10285                                                        skb),
10286                                       si->dst_reg, si->src_reg,
10287                                       offsetof(struct bpf_sock_ops_kern,
10288                                                skb));
10289                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10290                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10291                                       si->dst_reg, si->dst_reg,
10292                                       offsetof(struct sk_buff, len));
10293                 break;
10294         case offsetof(struct bpf_sock_ops, skb_tcp_flags):
10295                 off = offsetof(struct sk_buff, cb);
10296                 off += offsetof(struct tcp_skb_cb, tcp_flags);
10297                 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags);
10298                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern,
10299                                                        skb),
10300                                       si->dst_reg, si->src_reg,
10301                                       offsetof(struct bpf_sock_ops_kern,
10302                                                skb));
10303                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
10304                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb,
10305                                                        tcp_flags),
10306                                       si->dst_reg, si->dst_reg, off);
10307                 break;
10308         }
10309         return insn - insn_buf;
10310 }
10311
10312 /* data_end = skb->data + skb_headlen() */
10313 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si,
10314                                                     struct bpf_insn *insn)
10315 {
10316         int reg;
10317         int temp_reg_off = offsetof(struct sk_buff, cb) +
10318                            offsetof(struct sk_skb_cb, temp_reg);
10319
10320         if (si->src_reg == si->dst_reg) {
10321                 /* We need an extra register, choose and save a register. */
10322                 reg = BPF_REG_9;
10323                 if (si->src_reg == reg || si->dst_reg == reg)
10324                         reg--;
10325                 if (si->src_reg == reg || si->dst_reg == reg)
10326                         reg--;
10327                 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off);
10328         } else {
10329                 reg = si->dst_reg;
10330         }
10331
10332         /* reg = skb->data */
10333         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
10334                               reg, si->src_reg,
10335                               offsetof(struct sk_buff, data));
10336         /* AX = skb->len */
10337         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len),
10338                               BPF_REG_AX, si->src_reg,
10339                               offsetof(struct sk_buff, len));
10340         /* reg = skb->data + skb->len */
10341         *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX);
10342         /* AX = skb->data_len */
10343         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len),
10344                               BPF_REG_AX, si->src_reg,
10345                               offsetof(struct sk_buff, data_len));
10346
10347         /* reg = skb->data + skb->len - skb->data_len */
10348         *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX);
10349
10350         if (si->src_reg == si->dst_reg) {
10351                 /* Restore the saved register */
10352                 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg);
10353                 *insn++ = BPF_MOV64_REG(si->dst_reg, reg);
10354                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off);
10355         }
10356
10357         return insn;
10358 }
10359
10360 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
10361                                      const struct bpf_insn *si,
10362                                      struct bpf_insn *insn_buf,
10363                                      struct bpf_prog *prog, u32 *target_size)
10364 {
10365         struct bpf_insn *insn = insn_buf;
10366         int off;
10367
10368         switch (si->off) {
10369         case offsetof(struct __sk_buff, data_end):
10370                 insn = bpf_convert_data_end_access(si, insn);
10371                 break;
10372         case offsetof(struct __sk_buff, cb[0]) ...
10373              offsetofend(struct __sk_buff, cb[4]) - 1:
10374                 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20);
10375                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
10376                               offsetof(struct sk_skb_cb, data)) %
10377                              sizeof(__u64));
10378
10379                 prog->cb_access = 1;
10380                 off  = si->off;
10381                 off -= offsetof(struct __sk_buff, cb[0]);
10382                 off += offsetof(struct sk_buff, cb);
10383                 off += offsetof(struct sk_skb_cb, data);
10384                 if (type == BPF_WRITE)
10385                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
10386                                               si->src_reg, off);
10387                 else
10388                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
10389                                               si->src_reg, off);
10390                 break;
10391
10392
10393         default:
10394                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
10395                                               target_size);
10396         }
10397
10398         return insn - insn_buf;
10399 }
10400
10401 static u32 sk_msg_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, u32 *target_size)
10405 {
10406         struct bpf_insn *insn = insn_buf;
10407 #if IS_ENABLED(CONFIG_IPV6)
10408         int off;
10409 #endif
10410
10411         /* convert ctx uses the fact sg element is first in struct */
10412         BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0);
10413
10414         switch (si->off) {
10415         case offsetof(struct sk_msg_md, data):
10416                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
10417                                       si->dst_reg, si->src_reg,
10418                                       offsetof(struct sk_msg, data));
10419                 break;
10420         case offsetof(struct sk_msg_md, data_end):
10421                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
10422                                       si->dst_reg, si->src_reg,
10423                                       offsetof(struct sk_msg, data_end));
10424                 break;
10425         case offsetof(struct sk_msg_md, family):
10426                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2);
10427
10428                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10429                                               struct sk_msg, sk),
10430                                       si->dst_reg, si->src_reg,
10431                                       offsetof(struct sk_msg, sk));
10432                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10433                                       offsetof(struct sock_common, skc_family));
10434                 break;
10435
10436         case offsetof(struct sk_msg_md, remote_ip4):
10437                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4);
10438
10439                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10440                                                 struct sk_msg, sk),
10441                                       si->dst_reg, si->src_reg,
10442                                       offsetof(struct sk_msg, sk));
10443                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10444                                       offsetof(struct sock_common, skc_daddr));
10445                 break;
10446
10447         case offsetof(struct sk_msg_md, local_ip4):
10448                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10449                                           skc_rcv_saddr) != 4);
10450
10451                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10452                                               struct sk_msg, sk),
10453                                       si->dst_reg, si->src_reg,
10454                                       offsetof(struct sk_msg, sk));
10455                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10456                                       offsetof(struct sock_common,
10457                                                skc_rcv_saddr));
10458                 break;
10459
10460         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
10461              offsetof(struct sk_msg_md, remote_ip6[3]):
10462 #if IS_ENABLED(CONFIG_IPV6)
10463                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10464                                           skc_v6_daddr.s6_addr32[0]) != 4);
10465
10466                 off = si->off;
10467                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
10468                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10469                                                 struct sk_msg, sk),
10470                                       si->dst_reg, si->src_reg,
10471                                       offsetof(struct sk_msg, sk));
10472                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10473                                       offsetof(struct sock_common,
10474                                                skc_v6_daddr.s6_addr32[0]) +
10475                                       off);
10476 #else
10477                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10478 #endif
10479                 break;
10480
10481         case offsetof(struct sk_msg_md, local_ip6[0]) ...
10482              offsetof(struct sk_msg_md, local_ip6[3]):
10483 #if IS_ENABLED(CONFIG_IPV6)
10484                 BUILD_BUG_ON(sizeof_field(struct sock_common,
10485                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
10486
10487                 off = si->off;
10488                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
10489                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10490                                                 struct sk_msg, sk),
10491                                       si->dst_reg, si->src_reg,
10492                                       offsetof(struct sk_msg, sk));
10493                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
10494                                       offsetof(struct sock_common,
10495                                                skc_v6_rcv_saddr.s6_addr32[0]) +
10496                                       off);
10497 #else
10498                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
10499 #endif
10500                 break;
10501
10502         case offsetof(struct sk_msg_md, remote_port):
10503                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2);
10504
10505                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10506                                                 struct sk_msg, sk),
10507                                       si->dst_reg, si->src_reg,
10508                                       offsetof(struct sk_msg, sk));
10509                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10510                                       offsetof(struct sock_common, skc_dport));
10511 #ifndef __BIG_ENDIAN_BITFIELD
10512                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
10513 #endif
10514                 break;
10515
10516         case offsetof(struct sk_msg_md, local_port):
10517                 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2);
10518
10519                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
10520                                                 struct sk_msg, sk),
10521                                       si->dst_reg, si->src_reg,
10522                                       offsetof(struct sk_msg, sk));
10523                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
10524                                       offsetof(struct sock_common, skc_num));
10525                 break;
10526
10527         case offsetof(struct sk_msg_md, size):
10528                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size),
10529                                       si->dst_reg, si->src_reg,
10530                                       offsetof(struct sk_msg_sg, size));
10531                 break;
10532
10533         case offsetof(struct sk_msg_md, sk):
10534                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk),
10535                                       si->dst_reg, si->src_reg,
10536                                       offsetof(struct sk_msg, sk));
10537                 break;
10538         }
10539
10540         return insn - insn_buf;
10541 }
10542
10543 const struct bpf_verifier_ops sk_filter_verifier_ops = {
10544         .get_func_proto         = sk_filter_func_proto,
10545         .is_valid_access        = sk_filter_is_valid_access,
10546         .convert_ctx_access     = bpf_convert_ctx_access,
10547         .gen_ld_abs             = bpf_gen_ld_abs,
10548 };
10549
10550 const struct bpf_prog_ops sk_filter_prog_ops = {
10551         .test_run               = bpf_prog_test_run_skb,
10552 };
10553
10554 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
10555         .get_func_proto         = tc_cls_act_func_proto,
10556         .is_valid_access        = tc_cls_act_is_valid_access,
10557         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
10558         .gen_prologue           = tc_cls_act_prologue,
10559         .gen_ld_abs             = bpf_gen_ld_abs,
10560 };
10561
10562 const struct bpf_prog_ops tc_cls_act_prog_ops = {
10563         .test_run               = bpf_prog_test_run_skb,
10564 };
10565
10566 const struct bpf_verifier_ops xdp_verifier_ops = {
10567         .get_func_proto         = xdp_func_proto,
10568         .is_valid_access        = xdp_is_valid_access,
10569         .convert_ctx_access     = xdp_convert_ctx_access,
10570         .gen_prologue           = bpf_noop_prologue,
10571 };
10572
10573 const struct bpf_prog_ops xdp_prog_ops = {
10574         .test_run               = bpf_prog_test_run_xdp,
10575 };
10576
10577 const struct bpf_verifier_ops cg_skb_verifier_ops = {
10578         .get_func_proto         = cg_skb_func_proto,
10579         .is_valid_access        = cg_skb_is_valid_access,
10580         .convert_ctx_access     = bpf_convert_ctx_access,
10581 };
10582
10583 const struct bpf_prog_ops cg_skb_prog_ops = {
10584         .test_run               = bpf_prog_test_run_skb,
10585 };
10586
10587 const struct bpf_verifier_ops lwt_in_verifier_ops = {
10588         .get_func_proto         = lwt_in_func_proto,
10589         .is_valid_access        = lwt_is_valid_access,
10590         .convert_ctx_access     = bpf_convert_ctx_access,
10591 };
10592
10593 const struct bpf_prog_ops lwt_in_prog_ops = {
10594         .test_run               = bpf_prog_test_run_skb,
10595 };
10596
10597 const struct bpf_verifier_ops lwt_out_verifier_ops = {
10598         .get_func_proto         = lwt_out_func_proto,
10599         .is_valid_access        = lwt_is_valid_access,
10600         .convert_ctx_access     = bpf_convert_ctx_access,
10601 };
10602
10603 const struct bpf_prog_ops lwt_out_prog_ops = {
10604         .test_run               = bpf_prog_test_run_skb,
10605 };
10606
10607 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
10608         .get_func_proto         = lwt_xmit_func_proto,
10609         .is_valid_access        = lwt_is_valid_access,
10610         .convert_ctx_access     = bpf_convert_ctx_access,
10611         .gen_prologue           = tc_cls_act_prologue,
10612 };
10613
10614 const struct bpf_prog_ops lwt_xmit_prog_ops = {
10615         .test_run               = bpf_prog_test_run_skb,
10616 };
10617
10618 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
10619         .get_func_proto         = lwt_seg6local_func_proto,
10620         .is_valid_access        = lwt_is_valid_access,
10621         .convert_ctx_access     = bpf_convert_ctx_access,
10622 };
10623
10624 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
10625         .test_run               = bpf_prog_test_run_skb,
10626 };
10627
10628 const struct bpf_verifier_ops cg_sock_verifier_ops = {
10629         .get_func_proto         = sock_filter_func_proto,
10630         .is_valid_access        = sock_filter_is_valid_access,
10631         .convert_ctx_access     = bpf_sock_convert_ctx_access,
10632 };
10633
10634 const struct bpf_prog_ops cg_sock_prog_ops = {
10635 };
10636
10637 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
10638         .get_func_proto         = sock_addr_func_proto,
10639         .is_valid_access        = sock_addr_is_valid_access,
10640         .convert_ctx_access     = sock_addr_convert_ctx_access,
10641 };
10642
10643 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
10644 };
10645
10646 const struct bpf_verifier_ops sock_ops_verifier_ops = {
10647         .get_func_proto         = sock_ops_func_proto,
10648         .is_valid_access        = sock_ops_is_valid_access,
10649         .convert_ctx_access     = sock_ops_convert_ctx_access,
10650 };
10651
10652 const struct bpf_prog_ops sock_ops_prog_ops = {
10653 };
10654
10655 const struct bpf_verifier_ops sk_skb_verifier_ops = {
10656         .get_func_proto         = sk_skb_func_proto,
10657         .is_valid_access        = sk_skb_is_valid_access,
10658         .convert_ctx_access     = sk_skb_convert_ctx_access,
10659         .gen_prologue           = sk_skb_prologue,
10660 };
10661
10662 const struct bpf_prog_ops sk_skb_prog_ops = {
10663 };
10664
10665 const struct bpf_verifier_ops sk_msg_verifier_ops = {
10666         .get_func_proto         = sk_msg_func_proto,
10667         .is_valid_access        = sk_msg_is_valid_access,
10668         .convert_ctx_access     = sk_msg_convert_ctx_access,
10669         .gen_prologue           = bpf_noop_prologue,
10670 };
10671
10672 const struct bpf_prog_ops sk_msg_prog_ops = {
10673 };
10674
10675 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
10676         .get_func_proto         = flow_dissector_func_proto,
10677         .is_valid_access        = flow_dissector_is_valid_access,
10678         .convert_ctx_access     = flow_dissector_convert_ctx_access,
10679 };
10680
10681 const struct bpf_prog_ops flow_dissector_prog_ops = {
10682         .test_run               = bpf_prog_test_run_flow_dissector,
10683 };
10684
10685 int sk_detach_filter(struct sock *sk)
10686 {
10687         int ret = -ENOENT;
10688         struct sk_filter *filter;
10689
10690         if (sock_flag(sk, SOCK_FILTER_LOCKED))
10691                 return -EPERM;
10692
10693         filter = rcu_dereference_protected(sk->sk_filter,
10694                                            lockdep_sock_is_held(sk));
10695         if (filter) {
10696                 RCU_INIT_POINTER(sk->sk_filter, NULL);
10697                 sk_filter_uncharge(sk, filter);
10698                 ret = 0;
10699         }
10700
10701         return ret;
10702 }
10703 EXPORT_SYMBOL_GPL(sk_detach_filter);
10704
10705 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len)
10706 {
10707         struct sock_fprog_kern *fprog;
10708         struct sk_filter *filter;
10709         int ret = 0;
10710
10711         sockopt_lock_sock(sk);
10712         filter = rcu_dereference_protected(sk->sk_filter,
10713                                            lockdep_sock_is_held(sk));
10714         if (!filter)
10715                 goto out;
10716
10717         /* We're copying the filter that has been originally attached,
10718          * so no conversion/decode needed anymore. eBPF programs that
10719          * have no original program cannot be dumped through this.
10720          */
10721         ret = -EACCES;
10722         fprog = filter->prog->orig_prog;
10723         if (!fprog)
10724                 goto out;
10725
10726         ret = fprog->len;
10727         if (!len)
10728                 /* User space only enquires number of filter blocks. */
10729                 goto out;
10730
10731         ret = -EINVAL;
10732         if (len < fprog->len)
10733                 goto out;
10734
10735         ret = -EFAULT;
10736         if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog)))
10737                 goto out;
10738
10739         /* Instead of bytes, the API requests to return the number
10740          * of filter blocks.
10741          */
10742         ret = fprog->len;
10743 out:
10744         sockopt_release_sock(sk);
10745         return ret;
10746 }
10747
10748 #ifdef CONFIG_INET
10749 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
10750                                     struct sock_reuseport *reuse,
10751                                     struct sock *sk, struct sk_buff *skb,
10752                                     struct sock *migrating_sk,
10753                                     u32 hash)
10754 {
10755         reuse_kern->skb = skb;
10756         reuse_kern->sk = sk;
10757         reuse_kern->selected_sk = NULL;
10758         reuse_kern->migrating_sk = migrating_sk;
10759         reuse_kern->data_end = skb->data + skb_headlen(skb);
10760         reuse_kern->hash = hash;
10761         reuse_kern->reuseport_id = reuse->reuseport_id;
10762         reuse_kern->bind_inany = reuse->bind_inany;
10763 }
10764
10765 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
10766                                   struct bpf_prog *prog, struct sk_buff *skb,
10767                                   struct sock *migrating_sk,
10768                                   u32 hash)
10769 {
10770         struct sk_reuseport_kern reuse_kern;
10771         enum sk_action action;
10772
10773         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash);
10774         action = bpf_prog_run(prog, &reuse_kern);
10775
10776         if (action == SK_PASS)
10777                 return reuse_kern.selected_sk;
10778         else
10779                 return ERR_PTR(-ECONNREFUSED);
10780 }
10781
10782 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
10783            struct bpf_map *, map, void *, key, u32, flags)
10784 {
10785         bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY;
10786         struct sock_reuseport *reuse;
10787         struct sock *selected_sk;
10788
10789         selected_sk = map->ops->map_lookup_elem(map, key);
10790         if (!selected_sk)
10791                 return -ENOENT;
10792
10793         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
10794         if (!reuse) {
10795                 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */
10796                 if (sk_is_refcounted(selected_sk))
10797                         sock_put(selected_sk);
10798
10799                 /* reuseport_array has only sk with non NULL sk_reuseport_cb.
10800                  * The only (!reuse) case here is - the sk has already been
10801                  * unhashed (e.g. by close()), so treat it as -ENOENT.
10802                  *
10803                  * Other maps (e.g. sock_map) do not provide this guarantee and
10804                  * the sk may never be in the reuseport group to begin with.
10805                  */
10806                 return is_sockarray ? -ENOENT : -EINVAL;
10807         }
10808
10809         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
10810                 struct sock *sk = reuse_kern->sk;
10811
10812                 if (sk->sk_protocol != selected_sk->sk_protocol)
10813                         return -EPROTOTYPE;
10814                 else if (sk->sk_family != selected_sk->sk_family)
10815                         return -EAFNOSUPPORT;
10816
10817                 /* Catch all. Likely bound to a different sockaddr. */
10818                 return -EBADFD;
10819         }
10820
10821         reuse_kern->selected_sk = selected_sk;
10822
10823         return 0;
10824 }
10825
10826 static const struct bpf_func_proto sk_select_reuseport_proto = {
10827         .func           = sk_select_reuseport,
10828         .gpl_only       = false,
10829         .ret_type       = RET_INTEGER,
10830         .arg1_type      = ARG_PTR_TO_CTX,
10831         .arg2_type      = ARG_CONST_MAP_PTR,
10832         .arg3_type      = ARG_PTR_TO_MAP_KEY,
10833         .arg4_type      = ARG_ANYTHING,
10834 };
10835
10836 BPF_CALL_4(sk_reuseport_load_bytes,
10837            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10838            void *, to, u32, len)
10839 {
10840         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
10841 }
10842
10843 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
10844         .func           = sk_reuseport_load_bytes,
10845         .gpl_only       = false,
10846         .ret_type       = RET_INTEGER,
10847         .arg1_type      = ARG_PTR_TO_CTX,
10848         .arg2_type      = ARG_ANYTHING,
10849         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10850         .arg4_type      = ARG_CONST_SIZE,
10851 };
10852
10853 BPF_CALL_5(sk_reuseport_load_bytes_relative,
10854            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
10855            void *, to, u32, len, u32, start_header)
10856 {
10857         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
10858                                                len, start_header);
10859 }
10860
10861 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
10862         .func           = sk_reuseport_load_bytes_relative,
10863         .gpl_only       = false,
10864         .ret_type       = RET_INTEGER,
10865         .arg1_type      = ARG_PTR_TO_CTX,
10866         .arg2_type      = ARG_ANYTHING,
10867         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
10868         .arg4_type      = ARG_CONST_SIZE,
10869         .arg5_type      = ARG_ANYTHING,
10870 };
10871
10872 static const struct bpf_func_proto *
10873 sk_reuseport_func_proto(enum bpf_func_id func_id,
10874                         const struct bpf_prog *prog)
10875 {
10876         switch (func_id) {
10877         case BPF_FUNC_sk_select_reuseport:
10878                 return &sk_select_reuseport_proto;
10879         case BPF_FUNC_skb_load_bytes:
10880                 return &sk_reuseport_load_bytes_proto;
10881         case BPF_FUNC_skb_load_bytes_relative:
10882                 return &sk_reuseport_load_bytes_relative_proto;
10883         case BPF_FUNC_get_socket_cookie:
10884                 return &bpf_get_socket_ptr_cookie_proto;
10885         case BPF_FUNC_ktime_get_coarse_ns:
10886                 return &bpf_ktime_get_coarse_ns_proto;
10887         default:
10888                 return bpf_base_func_proto(func_id);
10889         }
10890 }
10891
10892 static bool
10893 sk_reuseport_is_valid_access(int off, int size,
10894                              enum bpf_access_type type,
10895                              const struct bpf_prog *prog,
10896                              struct bpf_insn_access_aux *info)
10897 {
10898         const u32 size_default = sizeof(__u32);
10899
10900         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
10901             off % size || type != BPF_READ)
10902                 return false;
10903
10904         switch (off) {
10905         case offsetof(struct sk_reuseport_md, data):
10906                 info->reg_type = PTR_TO_PACKET;
10907                 return size == sizeof(__u64);
10908
10909         case offsetof(struct sk_reuseport_md, data_end):
10910                 info->reg_type = PTR_TO_PACKET_END;
10911                 return size == sizeof(__u64);
10912
10913         case offsetof(struct sk_reuseport_md, hash):
10914                 return size == size_default;
10915
10916         case offsetof(struct sk_reuseport_md, sk):
10917                 info->reg_type = PTR_TO_SOCKET;
10918                 return size == sizeof(__u64);
10919
10920         case offsetof(struct sk_reuseport_md, migrating_sk):
10921                 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL;
10922                 return size == sizeof(__u64);
10923
10924         /* Fields that allow narrowing */
10925         case bpf_ctx_range(struct sk_reuseport_md, eth_protocol):
10926                 if (size < sizeof_field(struct sk_buff, protocol))
10927                         return false;
10928                 fallthrough;
10929         case bpf_ctx_range(struct sk_reuseport_md, ip_protocol):
10930         case bpf_ctx_range(struct sk_reuseport_md, bind_inany):
10931         case bpf_ctx_range(struct sk_reuseport_md, len):
10932                 bpf_ctx_record_field_size(info, size_default);
10933                 return bpf_ctx_narrow_access_ok(off, size, size_default);
10934
10935         default:
10936                 return false;
10937         }
10938 }
10939
10940 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
10941         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
10942                               si->dst_reg, si->src_reg,                 \
10943                               bpf_target_off(struct sk_reuseport_kern, F, \
10944                                              sizeof_field(struct sk_reuseport_kern, F), \
10945                                              target_size));             \
10946         })
10947
10948 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
10949         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10950                                     struct sk_buff,                     \
10951                                     skb,                                \
10952                                     SKB_FIELD)
10953
10954 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD)                            \
10955         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
10956                                     struct sock,                        \
10957                                     sk,                                 \
10958                                     SK_FIELD)
10959
10960 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
10961                                            const struct bpf_insn *si,
10962                                            struct bpf_insn *insn_buf,
10963                                            struct bpf_prog *prog,
10964                                            u32 *target_size)
10965 {
10966         struct bpf_insn *insn = insn_buf;
10967
10968         switch (si->off) {
10969         case offsetof(struct sk_reuseport_md, data):
10970                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
10971                 break;
10972
10973         case offsetof(struct sk_reuseport_md, len):
10974                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
10975                 break;
10976
10977         case offsetof(struct sk_reuseport_md, eth_protocol):
10978                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
10979                 break;
10980
10981         case offsetof(struct sk_reuseport_md, ip_protocol):
10982                 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol);
10983                 break;
10984
10985         case offsetof(struct sk_reuseport_md, data_end):
10986                 SK_REUSEPORT_LOAD_FIELD(data_end);
10987                 break;
10988
10989         case offsetof(struct sk_reuseport_md, hash):
10990                 SK_REUSEPORT_LOAD_FIELD(hash);
10991                 break;
10992
10993         case offsetof(struct sk_reuseport_md, bind_inany):
10994                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
10995                 break;
10996
10997         case offsetof(struct sk_reuseport_md, sk):
10998                 SK_REUSEPORT_LOAD_FIELD(sk);
10999                 break;
11000
11001         case offsetof(struct sk_reuseport_md, migrating_sk):
11002                 SK_REUSEPORT_LOAD_FIELD(migrating_sk);
11003                 break;
11004         }
11005
11006         return insn - insn_buf;
11007 }
11008
11009 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
11010         .get_func_proto         = sk_reuseport_func_proto,
11011         .is_valid_access        = sk_reuseport_is_valid_access,
11012         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
11013 };
11014
11015 const struct bpf_prog_ops sk_reuseport_prog_ops = {
11016 };
11017
11018 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled);
11019 EXPORT_SYMBOL(bpf_sk_lookup_enabled);
11020
11021 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx,
11022            struct sock *, sk, u64, flags)
11023 {
11024         if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE |
11025                                BPF_SK_LOOKUP_F_NO_REUSEPORT)))
11026                 return -EINVAL;
11027         if (unlikely(sk && sk_is_refcounted(sk)))
11028                 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */
11029         if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN))
11030                 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */
11031         if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE))
11032                 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */
11033
11034         /* Check if socket is suitable for packet L3/L4 protocol */
11035         if (sk && sk->sk_protocol != ctx->protocol)
11036                 return -EPROTOTYPE;
11037         if (sk && sk->sk_family != ctx->family &&
11038             (sk->sk_family == AF_INET || ipv6_only_sock(sk)))
11039                 return -EAFNOSUPPORT;
11040
11041         if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE))
11042                 return -EEXIST;
11043
11044         /* Select socket as lookup result */
11045         ctx->selected_sk = sk;
11046         ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT;
11047         return 0;
11048 }
11049
11050 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = {
11051         .func           = bpf_sk_lookup_assign,
11052         .gpl_only       = false,
11053         .ret_type       = RET_INTEGER,
11054         .arg1_type      = ARG_PTR_TO_CTX,
11055         .arg2_type      = ARG_PTR_TO_SOCKET_OR_NULL,
11056         .arg3_type      = ARG_ANYTHING,
11057 };
11058
11059 static const struct bpf_func_proto *
11060 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
11061 {
11062         switch (func_id) {
11063         case BPF_FUNC_perf_event_output:
11064                 return &bpf_event_output_data_proto;
11065         case BPF_FUNC_sk_assign:
11066                 return &bpf_sk_lookup_assign_proto;
11067         case BPF_FUNC_sk_release:
11068                 return &bpf_sk_release_proto;
11069         default:
11070                 return bpf_sk_base_func_proto(func_id);
11071         }
11072 }
11073
11074 static bool sk_lookup_is_valid_access(int off, int size,
11075                                       enum bpf_access_type type,
11076                                       const struct bpf_prog *prog,
11077                                       struct bpf_insn_access_aux *info)
11078 {
11079         if (off < 0 || off >= sizeof(struct bpf_sk_lookup))
11080                 return false;
11081         if (off % size != 0)
11082                 return false;
11083         if (type != BPF_READ)
11084                 return false;
11085
11086         switch (off) {
11087         case offsetof(struct bpf_sk_lookup, sk):
11088                 info->reg_type = PTR_TO_SOCKET_OR_NULL;
11089                 return size == sizeof(__u64);
11090
11091         case bpf_ctx_range(struct bpf_sk_lookup, family):
11092         case bpf_ctx_range(struct bpf_sk_lookup, protocol):
11093         case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4):
11094         case bpf_ctx_range(struct bpf_sk_lookup, local_ip4):
11095         case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]):
11096         case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]):
11097         case bpf_ctx_range(struct bpf_sk_lookup, local_port):
11098         case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex):
11099                 bpf_ctx_record_field_size(info, sizeof(__u32));
11100                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32));
11101
11102         case bpf_ctx_range(struct bpf_sk_lookup, remote_port):
11103                 /* Allow 4-byte access to 2-byte field for backward compatibility */
11104                 if (size == sizeof(__u32))
11105                         return true;
11106                 bpf_ctx_record_field_size(info, sizeof(__be16));
11107                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16));
11108
11109         case offsetofend(struct bpf_sk_lookup, remote_port) ...
11110              offsetof(struct bpf_sk_lookup, local_ip4) - 1:
11111                 /* Allow access to zero padding for backward compatibility */
11112                 bpf_ctx_record_field_size(info, sizeof(__u16));
11113                 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16));
11114
11115         default:
11116                 return false;
11117         }
11118 }
11119
11120 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type,
11121                                         const struct bpf_insn *si,
11122                                         struct bpf_insn *insn_buf,
11123                                         struct bpf_prog *prog,
11124                                         u32 *target_size)
11125 {
11126         struct bpf_insn *insn = insn_buf;
11127
11128         switch (si->off) {
11129         case offsetof(struct bpf_sk_lookup, sk):
11130                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11131                                       offsetof(struct bpf_sk_lookup_kern, selected_sk));
11132                 break;
11133
11134         case offsetof(struct bpf_sk_lookup, family):
11135                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11136                                       bpf_target_off(struct bpf_sk_lookup_kern,
11137                                                      family, 2, target_size));
11138                 break;
11139
11140         case offsetof(struct bpf_sk_lookup, protocol):
11141                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11142                                       bpf_target_off(struct bpf_sk_lookup_kern,
11143                                                      protocol, 2, target_size));
11144                 break;
11145
11146         case offsetof(struct bpf_sk_lookup, remote_ip4):
11147                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11148                                       bpf_target_off(struct bpf_sk_lookup_kern,
11149                                                      v4.saddr, 4, target_size));
11150                 break;
11151
11152         case offsetof(struct bpf_sk_lookup, local_ip4):
11153                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11154                                       bpf_target_off(struct bpf_sk_lookup_kern,
11155                                                      v4.daddr, 4, target_size));
11156                 break;
11157
11158         case bpf_ctx_range_till(struct bpf_sk_lookup,
11159                                 remote_ip6[0], remote_ip6[3]): {
11160 #if IS_ENABLED(CONFIG_IPV6)
11161                 int off = si->off;
11162
11163                 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]);
11164                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11165                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11166                                       offsetof(struct bpf_sk_lookup_kern, v6.saddr));
11167                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11168                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11169 #else
11170                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11171 #endif
11172                 break;
11173         }
11174         case bpf_ctx_range_till(struct bpf_sk_lookup,
11175                                 local_ip6[0], local_ip6[3]): {
11176 #if IS_ENABLED(CONFIG_IPV6)
11177                 int off = si->off;
11178
11179                 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]);
11180                 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size);
11181                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg,
11182                                       offsetof(struct bpf_sk_lookup_kern, v6.daddr));
11183                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
11184                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off);
11185 #else
11186                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11187 #endif
11188                 break;
11189         }
11190         case offsetof(struct bpf_sk_lookup, remote_port):
11191                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11192                                       bpf_target_off(struct bpf_sk_lookup_kern,
11193                                                      sport, 2, target_size));
11194                 break;
11195
11196         case offsetofend(struct bpf_sk_lookup, remote_port):
11197                 *target_size = 2;
11198                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
11199                 break;
11200
11201         case offsetof(struct bpf_sk_lookup, local_port):
11202                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
11203                                       bpf_target_off(struct bpf_sk_lookup_kern,
11204                                                      dport, 2, target_size));
11205                 break;
11206
11207         case offsetof(struct bpf_sk_lookup, ingress_ifindex):
11208                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
11209                                       bpf_target_off(struct bpf_sk_lookup_kern,
11210                                                      ingress_ifindex, 4, target_size));
11211                 break;
11212         }
11213
11214         return insn - insn_buf;
11215 }
11216
11217 const struct bpf_prog_ops sk_lookup_prog_ops = {
11218         .test_run = bpf_prog_test_run_sk_lookup,
11219 };
11220
11221 const struct bpf_verifier_ops sk_lookup_verifier_ops = {
11222         .get_func_proto         = sk_lookup_func_proto,
11223         .is_valid_access        = sk_lookup_is_valid_access,
11224         .convert_ctx_access     = sk_lookup_convert_ctx_access,
11225 };
11226
11227 #endif /* CONFIG_INET */
11228
11229 DEFINE_BPF_DISPATCHER(xdp)
11230
11231 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog)
11232 {
11233         bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog);
11234 }
11235
11236 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE)
11237 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type)
11238 BTF_SOCK_TYPE_xxx
11239 #undef BTF_SOCK_TYPE
11240
11241 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk)
11242 {
11243         /* tcp6_sock type is not generated in dwarf and hence btf,
11244          * trigger an explicit type generation here.
11245          */
11246         BTF_TYPE_EMIT(struct tcp6_sock);
11247         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP &&
11248             sk->sk_family == AF_INET6)
11249                 return (unsigned long)sk;
11250
11251         return (unsigned long)NULL;
11252 }
11253
11254 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = {
11255         .func                   = bpf_skc_to_tcp6_sock,
11256         .gpl_only               = false,
11257         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11258         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11259         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP6],
11260 };
11261
11262 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk)
11263 {
11264         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP)
11265                 return (unsigned long)sk;
11266
11267         return (unsigned long)NULL;
11268 }
11269
11270 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = {
11271         .func                   = bpf_skc_to_tcp_sock,
11272         .gpl_only               = false,
11273         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11274         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11275         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP],
11276 };
11277
11278 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk)
11279 {
11280         /* BTF types for tcp_timewait_sock and inet_timewait_sock are not
11281          * generated if CONFIG_INET=n. Trigger an explicit generation here.
11282          */
11283         BTF_TYPE_EMIT(struct inet_timewait_sock);
11284         BTF_TYPE_EMIT(struct tcp_timewait_sock);
11285
11286 #ifdef CONFIG_INET
11287         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT)
11288                 return (unsigned long)sk;
11289 #endif
11290
11291 #if IS_BUILTIN(CONFIG_IPV6)
11292         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT)
11293                 return (unsigned long)sk;
11294 #endif
11295
11296         return (unsigned long)NULL;
11297 }
11298
11299 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = {
11300         .func                   = bpf_skc_to_tcp_timewait_sock,
11301         .gpl_only               = false,
11302         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11303         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11304         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW],
11305 };
11306
11307 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk)
11308 {
11309 #ifdef CONFIG_INET
11310         if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11311                 return (unsigned long)sk;
11312 #endif
11313
11314 #if IS_BUILTIN(CONFIG_IPV6)
11315         if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV)
11316                 return (unsigned long)sk;
11317 #endif
11318
11319         return (unsigned long)NULL;
11320 }
11321
11322 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = {
11323         .func                   = bpf_skc_to_tcp_request_sock,
11324         .gpl_only               = false,
11325         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11326         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11327         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ],
11328 };
11329
11330 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk)
11331 {
11332         /* udp6_sock type is not generated in dwarf and hence btf,
11333          * trigger an explicit type generation here.
11334          */
11335         BTF_TYPE_EMIT(struct udp6_sock);
11336         if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP &&
11337             sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6)
11338                 return (unsigned long)sk;
11339
11340         return (unsigned long)NULL;
11341 }
11342
11343 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = {
11344         .func                   = bpf_skc_to_udp6_sock,
11345         .gpl_only               = false,
11346         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11347         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11348         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UDP6],
11349 };
11350
11351 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk)
11352 {
11353         /* unix_sock type is not generated in dwarf and hence btf,
11354          * trigger an explicit type generation here.
11355          */
11356         BTF_TYPE_EMIT(struct unix_sock);
11357         if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX)
11358                 return (unsigned long)sk;
11359
11360         return (unsigned long)NULL;
11361 }
11362
11363 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = {
11364         .func                   = bpf_skc_to_unix_sock,
11365         .gpl_only               = false,
11366         .ret_type               = RET_PTR_TO_BTF_ID_OR_NULL,
11367         .arg1_type              = ARG_PTR_TO_BTF_ID_SOCK_COMMON,
11368         .ret_btf_id             = &btf_sock_ids[BTF_SOCK_TYPE_UNIX],
11369 };
11370
11371 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk)
11372 {
11373         BTF_TYPE_EMIT(struct mptcp_sock);
11374         return (unsigned long)bpf_mptcp_sock_from_subflow(sk);
11375 }
11376
11377 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = {
11378         .func           = bpf_skc_to_mptcp_sock,
11379         .gpl_only       = false,
11380         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11381         .arg1_type      = ARG_PTR_TO_SOCK_COMMON,
11382         .ret_btf_id     = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP],
11383 };
11384
11385 BPF_CALL_1(bpf_sock_from_file, struct file *, file)
11386 {
11387         return (unsigned long)sock_from_file(file);
11388 }
11389
11390 BTF_ID_LIST(bpf_sock_from_file_btf_ids)
11391 BTF_ID(struct, socket)
11392 BTF_ID(struct, file)
11393
11394 const struct bpf_func_proto bpf_sock_from_file_proto = {
11395         .func           = bpf_sock_from_file,
11396         .gpl_only       = false,
11397         .ret_type       = RET_PTR_TO_BTF_ID_OR_NULL,
11398         .ret_btf_id     = &bpf_sock_from_file_btf_ids[0],
11399         .arg1_type      = ARG_PTR_TO_BTF_ID,
11400         .arg1_btf_id    = &bpf_sock_from_file_btf_ids[1],
11401 };
11402
11403 static const struct bpf_func_proto *
11404 bpf_sk_base_func_proto(enum bpf_func_id func_id)
11405 {
11406         const struct bpf_func_proto *func;
11407
11408         switch (func_id) {
11409         case BPF_FUNC_skc_to_tcp6_sock:
11410                 func = &bpf_skc_to_tcp6_sock_proto;
11411                 break;
11412         case BPF_FUNC_skc_to_tcp_sock:
11413                 func = &bpf_skc_to_tcp_sock_proto;
11414                 break;
11415         case BPF_FUNC_skc_to_tcp_timewait_sock:
11416                 func = &bpf_skc_to_tcp_timewait_sock_proto;
11417                 break;
11418         case BPF_FUNC_skc_to_tcp_request_sock:
11419                 func = &bpf_skc_to_tcp_request_sock_proto;
11420                 break;
11421         case BPF_FUNC_skc_to_udp6_sock:
11422                 func = &bpf_skc_to_udp6_sock_proto;
11423                 break;
11424         case BPF_FUNC_skc_to_unix_sock:
11425                 func = &bpf_skc_to_unix_sock_proto;
11426                 break;
11427         case BPF_FUNC_skc_to_mptcp_sock:
11428                 func = &bpf_skc_to_mptcp_sock_proto;
11429                 break;
11430         case BPF_FUNC_ktime_get_coarse_ns:
11431                 return &bpf_ktime_get_coarse_ns_proto;
11432         default:
11433                 return bpf_base_func_proto(func_id);
11434         }
11435
11436         if (!perfmon_capable())
11437                 return NULL;
11438
11439         return func;
11440 }