net/core/filter.c

   1 /*
   2  * Linux Socket Filter - Kernel level socket filtering
   3  *
   4  * Based on the design of the Berkeley Packet Filter. The new
   5  * internal format has been designed by PLUMgrid:
   6  *
   7  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   8  *
   9  * Authors:
  10  *
  11  *      Jay Schulist <jschlst@samba.org>
  12  *      Alexei Starovoitov <ast@plumgrid.com>
  13  *      Daniel Borkmann <dborkman@redhat.com>
  14  *
  15  * This program is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU General Public License
  17  * as published by the Free Software Foundation; either version
  18  * 2 of the License, or (at your option) any later version.
  19  *
  20  * Andi Kleen - Fix a few bad bugs and races.
  21  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  22  */
  23
  24 #include <linux/module.h>
  25 #include <linux/types.h>
  26 #include <linux/mm.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/socket.h>
  29 #include <linux/sock_diag.h>
  30 #include <linux/in.h>
  31 #include <linux/inet.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/if_packet.h>
  34 #include <linux/if_arp.h>
  35 #include <linux/gfp.h>
  36 #include <net/inet_common.h>
  37 #include <net/ip.h>
  38 #include <net/protocol.h>
  39 #include <net/netlink.h>
  40 #include <linux/skbuff.h>
  41 #include <linux/skmsg.h>
  42 #include <net/sock.h>
  43 #include <net/flow_dissector.h>
  44 #include <linux/errno.h>
  45 #include <linux/timer.h>
  46 #include <linux/uaccess.h>
  47 #include <asm/unaligned.h>
  48 #include <asm/cmpxchg.h>
  49 #include <linux/filter.h>
  50 #include <linux/ratelimit.h>
  51 #include <linux/seccomp.h>
  52 #include <linux/if_vlan.h>
  53 #include <linux/bpf.h>
  54 #include <net/sch_generic.h>
  55 #include <net/cls_cgroup.h>
  56 #include <net/dst_metadata.h>
  57 #include <net/dst.h>
  58 #include <net/sock_reuseport.h>
  59 #include <net/busy_poll.h>
  60 #include <net/tcp.h>
  61 #include <net/xfrm.h>
  62 #include <net/udp.h>
  63 #include <linux/bpf_trace.h>
  64 #include <net/xdp_sock.h>
  65 #include <linux/inetdevice.h>
  66 #include <net/inet_hashtables.h>
  67 #include <net/inet6_hashtables.h>
  68 #include <net/ip_fib.h>
  69 #include <net/flow.h>
  70 #include <net/arp.h>
  71 #include <net/ipv6.h>
  72 #include <net/net_namespace.h>
  73 #include <linux/seg6_local.h>
  74 #include <net/seg6.h>
  75 #include <net/seg6_local.h>
  76
  77 /**
  78  *      sk_filter_trim_cap - run a packet through a socket filter
  79  *      @sk: sock associated with &sk_buff
  80  *      @skb: buffer to filter
  81  *      @cap: limit on how short the eBPF program may trim the packet
  82  *
  83  * Run the eBPF program and then cut skb->data to correct size returned by
  84  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
  85  * than pkt_len we keep whole skb->data. This is the socket level
  86  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
  87  * be accepted or -EPERM if the packet should be tossed.
  88  *
  89  */
  90 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
  91 {
  92         int err;
  93         struct sk_filter *filter;
  94
  95         /*
  96          * If the skb was allocated from pfmemalloc reserves, only
  97          * allow SOCK_MEMALLOC sockets to use it as this socket is
  98          * helping free memory
  99          */
 100         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
 101                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
 102                 return -ENOMEM;
 103         }
 104         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 105         if (err)
 106                 return err;
 107
 108         err = security_sock_rcv_skb(sk, skb);
 109         if (err)
 110                 return err;
 111
 112         rcu_read_lock();
 113         filter = rcu_dereference(sk->sk_filter);
 114         if (filter) {
 115                 struct sock *save_sk = skb->sk;
 116                 unsigned int pkt_len;
 117
 118                 skb->sk = sk;
 119                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 120                 skb->sk = save_sk;
 121                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 122         }
 123         rcu_read_unlock();
 124
 125         return err;
 126 }
 127 EXPORT_SYMBOL(sk_filter_trim_cap);
 128
 129 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 130 {
 131         return skb_get_poff(skb);
 132 }
 133
 134 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 135 {
 136         struct nlattr *nla;
 137
 138         if (skb_is_nonlinear(skb))
 139                 return 0;
 140
 141         if (skb->len < sizeof(struct nlattr))
 142                 return 0;
 143
 144         if (a > skb->len - sizeof(struct nlattr))
 145                 return 0;
 146
 147         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 148         if (nla)
 149                 return (void *) nla - (void *) skb->data;
 150
 151         return 0;
 152 }
 153
 154 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 155 {
 156         struct nlattr *nla;
 157
 158         if (skb_is_nonlinear(skb))
 159                 return 0;
 160
 161         if (skb->len < sizeof(struct nlattr))
 162                 return 0;
 163
 164         if (a > skb->len - sizeof(struct nlattr))
 165                 return 0;
 166
 167         nla = (struct nlattr *) &skb->data[a];
 168         if (nla->nla_len > skb->len - a)
 169                 return 0;
 170
 171         nla = nla_find_nested(nla, x);
 172         if (nla)
 173                 return (void *) nla - (void *) skb->data;
 174
 175         return 0;
 176 }
 177
 178 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 179            data, int, headlen, int, offset)
 180 {
 181         u8 tmp, *ptr;
 182         const int len = sizeof(tmp);
 183
 184         if (offset >= 0) {
 185                 if (headlen - offset >= len)
 186                         return *(u8 *)(data + offset);
 187                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 188                         return tmp;
 189         } else {
 190                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 191                 if (likely(ptr))
 192                         return *(u8 *)ptr;
 193         }
 194
 195         return -EFAULT;
 196 }
 197
 198 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 199            int, offset)
 200 {
 201         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 202                                          offset);
 203 }
 204
 205 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 206            data, int, headlen, int, offset)
 207 {
 208         u16 tmp, *ptr;
 209         const int len = sizeof(tmp);
 210
 211         if (offset >= 0) {
 212                 if (headlen - offset >= len)
 213                         return get_unaligned_be16(data + offset);
 214                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 215                         return be16_to_cpu(tmp);
 216         } else {
 217                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 218                 if (likely(ptr))
 219                         return get_unaligned_be16(ptr);
 220         }
 221
 222         return -EFAULT;
 223 }
 224
 225 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 226            int, offset)
 227 {
 228         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 229                                           offset);
 230 }
 231
 232 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 233            data, int, headlen, int, offset)
 234 {
 235         u32 tmp, *ptr;
 236         const int len = sizeof(tmp);
 237
 238         if (likely(offset >= 0)) {
 239                 if (headlen - offset >= len)
 240                         return get_unaligned_be32(data + offset);
 241                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 242                         return be32_to_cpu(tmp);
 243         } else {
 244                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 245                 if (likely(ptr))
 246                         return get_unaligned_be32(ptr);
 247         }
 248
 249         return -EFAULT;
 250 }
 251
 252 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 253            int, offset)
 254 {
 255         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 256                                           offset);
 257 }
 258
 259 BPF_CALL_0(bpf_get_raw_cpu_id)
 260 {
 261         return raw_smp_processor_id();
 262 }
 263
 264 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 265         .func           = bpf_get_raw_cpu_id,
 266         .gpl_only       = false,
 267         .ret_type       = RET_INTEGER,
 268 };
 269
 270 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 271                               struct bpf_insn *insn_buf)
 272 {
 273         struct bpf_insn *insn = insn_buf;
 274
 275         switch (skb_field) {
 276         case SKF_AD_MARK:
 277                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
 278
 279                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 280                                       offsetof(struct sk_buff, mark));
 281                 break;
 282
 283         case SKF_AD_PKTTYPE:
 284                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 285                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 286 #ifdef __BIG_ENDIAN_BITFIELD
 287                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 288 #endif
 289                 break;
 290
 291         case SKF_AD_QUEUE:
 292                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
 293
 294                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 295                                       offsetof(struct sk_buff, queue_mapping));
 296                 break;
 297
 298         case SKF_AD_VLAN_TAG:
 299         case SKF_AD_VLAN_TAG_PRESENT:
 300                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
 301                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
 302
 303                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 304                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 305                                       offsetof(struct sk_buff, vlan_tci));
 306                 if (skb_field == SKF_AD_VLAN_TAG) {
 307                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
 308                                                 ~VLAN_TAG_PRESENT);
 309                 } else {
 310                         /* dst_reg >>= 12 */
 311                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
 312                         /* dst_reg &= 1 */
 313                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 314                 }
 315                 break;
 316         }
 317
 318         return insn - insn_buf;
 319 }
 320
 321 static bool convert_bpf_extensions(struct sock_filter *fp,
 322                                    struct bpf_insn **insnp)
 323 {
 324         struct bpf_insn *insn = *insnp;
 325         u32 cnt;
 326
 327         switch (fp->k) {
 328         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 329                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
 330
 331                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 332                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 333                                       offsetof(struct sk_buff, protocol));
 334                 /* A = ntohs(A) [emitting a nop or swap16] */
 335                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 336                 break;
 337
 338         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 339                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 340                 insn += cnt - 1;
 341                 break;
 342
 343         case SKF_AD_OFF + SKF_AD_IFINDEX:
 344         case SKF_AD_OFF + SKF_AD_HATYPE:
 345                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
 346                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
 347
 348                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 349                                       BPF_REG_TMP, BPF_REG_CTX,
 350                                       offsetof(struct sk_buff, dev));
 351                 /* if (tmp != 0) goto pc + 1 */
 352                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 353                 *insn++ = BPF_EXIT_INSN();
 354                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 355                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 356                                             offsetof(struct net_device, ifindex));
 357                 else
 358                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 359                                             offsetof(struct net_device, type));
 360                 break;
 361
 362         case SKF_AD_OFF + SKF_AD_MARK:
 363                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 364                 insn += cnt - 1;
 365                 break;
 366
 367         case SKF_AD_OFF + SKF_AD_RXHASH:
 368                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
 369
 370                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 371                                     offsetof(struct sk_buff, hash));
 372                 break;
 373
 374         case SKF_AD_OFF + SKF_AD_QUEUE:
 375                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 376                 insn += cnt - 1;
 377                 break;
 378
 379         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 380                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 381                                          BPF_REG_A, BPF_REG_CTX, insn);
 382                 insn += cnt - 1;
 383                 break;
 384
 385         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 386                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 387                                          BPF_REG_A, BPF_REG_CTX, insn);
 388                 insn += cnt - 1;
 389                 break;
 390
 391         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 392                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
 393
 394                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 395                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 396                                       offsetof(struct sk_buff, vlan_proto));
 397                 /* A = ntohs(A) [emitting a nop or swap16] */
 398                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 399                 break;
 400
 401         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 402         case SKF_AD_OFF + SKF_AD_NLATTR:
 403         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 404         case SKF_AD_OFF + SKF_AD_CPU:
 405         case SKF_AD_OFF + SKF_AD_RANDOM:
 406                 /* arg1 = CTX */
 407                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 408                 /* arg2 = A */
 409                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 410                 /* arg3 = X */
 411                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 412                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 413                 switch (fp->k) {
 414                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 415                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 416                         break;
 417                 case SKF_AD_OFF + SKF_AD_NLATTR:
 418                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 419                         break;
 420                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 421                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 422                         break;
 423                 case SKF_AD_OFF + SKF_AD_CPU:
 424                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 425                         break;
 426                 case SKF_AD_OFF + SKF_AD_RANDOM:
 427                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 428                         bpf_user_rnd_init_once();
 429                         break;
 430                 }
 431                 break;
 432
 433         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 434                 /* A ^= X */
 435                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 436                 break;
 437
 438         default:
 439                 /* This is just a dummy call to avoid letting the compiler
 440                  * evict __bpf_call_base() as an optimization. Placed here
 441                  * where no-one bothers.
 442                  */
 443                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 444                 return false;
 445         }
 446
 447         *insnp = insn;
 448         return true;
 449 }
 450
 451 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 452 {
 453         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 454         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 455         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 456                       BPF_SIZE(fp->code) == BPF_W;
 457         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 458         const int ip_align = NET_IP_ALIGN;
 459         struct bpf_insn *insn = *insnp;
 460         int offset = fp->k;
 461
 462         if (!indirect &&
 463             ((unaligned_ok && offset >= 0) ||
 464              (!unaligned_ok && offset >= 0 &&
 465               offset + ip_align >= 0 &&
 466               offset + ip_align % size == 0))) {
 467                 bool ldx_off_ok = offset <= S16_MAX;
 468
 469                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 470                 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 471                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 472                                       size, 2 + endian + (!ldx_off_ok * 2));
 473                 if (ldx_off_ok) {
 474                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 475                                               BPF_REG_D, offset);
 476                 } else {
 477                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 478                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 479                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 480                                               BPF_REG_TMP, 0);
 481                 }
 482                 if (endian)
 483                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 484                 *insn++ = BPF_JMP_A(8);
 485         }
 486
 487         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 488         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 489         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 490         if (!indirect) {
 491                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 492         } else {
 493                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 494                 if (fp->k)
 495                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 496         }
 497
 498         switch (BPF_SIZE(fp->code)) {
 499         case BPF_B:
 500                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 501                 break;
 502         case BPF_H:
 503                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 504                 break;
 505         case BPF_W:
 506                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 507                 break;
 508         default:
 509                 return false;
 510         }
 511
 512         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 513         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 514         *insn   = BPF_EXIT_INSN();
 515
 516         *insnp = insn;
 517         return true;
 518 }
 519
 520 /**
 521  *      bpf_convert_filter - convert filter program
 522  *      @prog: the user passed filter program
 523  *      @len: the length of the user passed filter program
 524  *      @new_prog: allocated 'struct bpf_prog' or NULL
 525  *      @new_len: pointer to store length of converted program
 526  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 527  *
 528  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 529  * style extended BPF (eBPF).
 530  * Conversion workflow:
 531  *
 532  * 1) First pass for calculating the new program length:
 533  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 534  *
 535  * 2) 2nd pass to remap in two passes: 1st pass finds new
 536  *    jump offsets, 2nd pass remapping:
 537  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 538  */
 539 static int bpf_convert_filter(struct sock_filter *prog, int len,
 540                               struct bpf_prog *new_prog, int *new_len,
 541                               bool *seen_ld_abs)
 542 {
 543         int new_flen = 0, pass = 0, target, i, stack_off;
 544         struct bpf_insn *new_insn, *first_insn = NULL;
 545         struct sock_filter *fp;
 546         int *addrs = NULL;
 547         u8 bpf_src;
 548
 549         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 550         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 551
 552         if (len <= 0 || len > BPF_MAXINSNS)
 553                 return -EINVAL;
 554
 555         if (new_prog) {
 556                 first_insn = new_prog->insnsi;
 557                 addrs = kcalloc(len, sizeof(*addrs),
 558                                 GFP_KERNEL | __GFP_NOWARN);
 559                 if (!addrs)
 560                         return -ENOMEM;
 561         }
 562
 563 do_pass:
 564         new_insn = first_insn;
 565         fp = prog;
 566
 567         /* Classic BPF related prologue emission. */
 568         if (new_prog) {
 569                 /* Classic BPF expects A and X to be reset first. These need
 570                  * to be guaranteed to be the first two instructions.
 571                  */
 572                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 573                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 574
 575                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 576                  * In eBPF case it's done by the compiler, here we need to
 577                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 578                  */
 579                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 580                 if (*seen_ld_abs) {
 581                         /* For packet access in classic BPF, cache skb->data
 582                          * in callee-saved BPF R8 and skb->len - skb->data_len
 583                          * (headlen) in BPF R9. Since classic BPF is read-only
 584                          * on CTX, we only need to cache it once.
 585                          */
 586                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 587                                                   BPF_REG_D, BPF_REG_CTX,
 588                                                   offsetof(struct sk_buff, data));
 589                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 590                                                   offsetof(struct sk_buff, len));
 591                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 592                                                   offsetof(struct sk_buff, data_len));
 593                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 594                 }
 595         } else {
 596                 new_insn += 3;
 597         }
 598
 599         for (i = 0; i < len; fp++, i++) {
 600                 struct bpf_insn tmp_insns[32] = { };
 601                 struct bpf_insn *insn = tmp_insns;
 602
 603                 if (addrs)
 604                         addrs[i] = new_insn - first_insn;
 605
 606                 switch (fp->code) {
 607                 /* All arithmetic insns and skb loads map as-is. */
 608                 case BPF_ALU | BPF_ADD | BPF_X:
 609                 case BPF_ALU | BPF_ADD | BPF_K:
 610                 case BPF_ALU | BPF_SUB | BPF_X:
 611                 case BPF_ALU | BPF_SUB | BPF_K:
 612                 case BPF_ALU | BPF_AND | BPF_X:
 613                 case BPF_ALU | BPF_AND | BPF_K:
 614                 case BPF_ALU | BPF_OR | BPF_X:
 615                 case BPF_ALU | BPF_OR | BPF_K:
 616                 case BPF_ALU | BPF_LSH | BPF_X:
 617                 case BPF_ALU | BPF_LSH | BPF_K:
 618                 case BPF_ALU | BPF_RSH | BPF_X:
 619                 case BPF_ALU | BPF_RSH | BPF_K:
 620                 case BPF_ALU | BPF_XOR | BPF_X:
 621                 case BPF_ALU | BPF_XOR | BPF_K:
 622                 case BPF_ALU | BPF_MUL | BPF_X:
 623                 case BPF_ALU | BPF_MUL | BPF_K:
 624                 case BPF_ALU | BPF_DIV | BPF_X:
 625                 case BPF_ALU | BPF_DIV | BPF_K:
 626                 case BPF_ALU | BPF_MOD | BPF_X:
 627                 case BPF_ALU | BPF_MOD | BPF_K:
 628                 case BPF_ALU | BPF_NEG:
 629                 case BPF_LD | BPF_ABS | BPF_W:
 630                 case BPF_LD | BPF_ABS | BPF_H:
 631                 case BPF_LD | BPF_ABS | BPF_B:
 632                 case BPF_LD | BPF_IND | BPF_W:
 633                 case BPF_LD | BPF_IND | BPF_H:
 634                 case BPF_LD | BPF_IND | BPF_B:
 635                         /* Check for overloaded BPF extension and
 636                          * directly convert it if found, otherwise
 637                          * just move on with mapping.
 638                          */
 639                         if (BPF_CLASS(fp->code) == BPF_LD &&
 640                             BPF_MODE(fp->code) == BPF_ABS &&
 641                             convert_bpf_extensions(fp, &insn))
 642                                 break;
 643                         if (BPF_CLASS(fp->code) == BPF_LD &&
 644                             convert_bpf_ld_abs(fp, &insn)) {
 645                                 *seen_ld_abs = true;
 646                                 break;
 647                         }
 648
 649                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 650                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 651                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 652                                 /* Error with exception code on div/mod by 0.
 653                                  * For cBPF programs, this was always return 0.
 654                                  */
 655                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 656                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 657                                 *insn++ = BPF_EXIT_INSN();
 658                         }
 659
 660                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 661                         break;
 662
 663                 /* Jump transformation cannot use BPF block macros
 664                  * everywhere as offset calculation and target updates
 665                  * require a bit more work than the rest, i.e. jump
 666                  * opcodes map as-is, but offsets need adjustment.
 667                  */
 668
 669 #define BPF_EMIT_JMP                                                    \
 670         do {                                                            \
 671                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 672                 s32 off;                                                \
 673                                                                         \
 674                 if (target >= len || target < 0)                        \
 675                         goto err;                                       \
 676                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 677                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 678                 off -= insn - tmp_insns;                                \
 679                 /* Reject anything not fitting into insn->off. */       \
 680                 if (off < off_min || off > off_max)                     \
 681                         goto err;                                       \
 682                 insn->off = off;                                        \
 683         } while (0)
 684
 685                 case BPF_JMP | BPF_JA:
 686                         target = i + fp->k + 1;
 687                         insn->code = fp->code;
 688                         BPF_EMIT_JMP;
 689                         break;
 690
 691                 case BPF_JMP | BPF_JEQ | BPF_K:
 692                 case BPF_JMP | BPF_JEQ | BPF_X:
 693                 case BPF_JMP | BPF_JSET | BPF_K:
 694                 case BPF_JMP | BPF_JSET | BPF_X:
 695                 case BPF_JMP | BPF_JGT | BPF_K:
 696                 case BPF_JMP | BPF_JGT | BPF_X:
 697                 case BPF_JMP | BPF_JGE | BPF_K:
 698                 case BPF_JMP | BPF_JGE | BPF_X:
 699                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 700                                 /* BPF immediates are signed, zero extend
 701                                  * immediate into tmp register and use it
 702                                  * in compare insn.
 703                                  */
 704                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 705
 706                                 insn->dst_reg = BPF_REG_A;
 707                                 insn->src_reg = BPF_REG_TMP;
 708                                 bpf_src = BPF_X;
 709                         } else {
 710                                 insn->dst_reg = BPF_REG_A;
 711                                 insn->imm = fp->k;
 712                                 bpf_src = BPF_SRC(fp->code);
 713                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 714                         }
 715
 716                         /* Common case where 'jump_false' is next insn. */
 717                         if (fp->jf == 0) {
 718                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 719                                 target = i + fp->jt + 1;
 720                                 BPF_EMIT_JMP;
 721                                 break;
 722                         }
 723
 724                         /* Convert some jumps when 'jump_true' is next insn. */
 725                         if (fp->jt == 0) {
 726                                 switch (BPF_OP(fp->code)) {
 727                                 case BPF_JEQ:
 728                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 729                                         break;
 730                                 case BPF_JGT:
 731                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 732                                         break;
 733                                 case BPF_JGE:
 734                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 735                                         break;
 736                                 default:
 737                                         goto jmp_rest;
 738                                 }
 739
 740                                 target = i + fp->jf + 1;
 741                                 BPF_EMIT_JMP;
 742                                 break;
 743                         }
 744 jmp_rest:
 745                         /* Other jumps are mapped into two insns: Jxx and JA. */
 746                         target = i + fp->jt + 1;
 747                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 748                         BPF_EMIT_JMP;
 749                         insn++;
 750
 751                         insn->code = BPF_JMP | BPF_JA;
 752                         target = i + fp->jf + 1;
 753                         BPF_EMIT_JMP;
 754                         break;
 755
 756                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 757                 case BPF_LDX | BPF_MSH | BPF_B: {
 758                         struct sock_filter tmp = {
 759                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 760                                 .k      = fp->k,
 761                         };
 762
 763                         *seen_ld_abs = true;
 764
 765                         /* X = A */
 766                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 767                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 768                         convert_bpf_ld_abs(&tmp, &insn);
 769                         insn++;
 770                         /* A &= 0xf */
 771                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 772                         /* A <<= 2 */
 773                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 774                         /* tmp = X */
 775                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 776                         /* X = A */
 777                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 778                         /* A = tmp */
 779                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 780                         break;
 781                 }
 782                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 783                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 784                  */
 785                 case BPF_RET | BPF_A:
 786                 case BPF_RET | BPF_K:
 787                         if (BPF_RVAL(fp->code) == BPF_K)
 788                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 789                                                         0, fp->k);
 790                         *insn = BPF_EXIT_INSN();
 791                         break;
 792
 793                 /* Store to stack. */
 794                 case BPF_ST:
 795                 case BPF_STX:
 796                         stack_off = fp->k * 4  + 4;
 797                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 798                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 799                                             -stack_off);
 800                         /* check_load_and_stores() verifies that classic BPF can
 801                          * load from stack only after write, so tracking
 802                          * stack_depth for ST|STX insns is enough
 803                          */
 804                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 805                                 new_prog->aux->stack_depth = stack_off;
 806                         break;
 807
 808                 /* Load from stack. */
 809                 case BPF_LD | BPF_MEM:
 810                 case BPF_LDX | BPF_MEM:
 811                         stack_off = fp->k * 4  + 4;
 812                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 813                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 814                                             -stack_off);
 815                         break;
 816
 817                 /* A = K or X = K */
 818                 case BPF_LD | BPF_IMM:
 819                 case BPF_LDX | BPF_IMM:
 820                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 821                                               BPF_REG_A : BPF_REG_X, fp->k);
 822                         break;
 823
 824                 /* X = A */
 825                 case BPF_MISC | BPF_TAX:
 826                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 827                         break;
 828
 829                 /* A = X */
 830                 case BPF_MISC | BPF_TXA:
 831                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 832                         break;
 833
 834                 /* A = skb->len or X = skb->len */
 835                 case BPF_LD | BPF_W | BPF_LEN:
 836                 case BPF_LDX | BPF_W | BPF_LEN:
 837                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 838                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 839                                             offsetof(struct sk_buff, len));
 840                         break;
 841
 842                 /* Access seccomp_data fields. */
 843                 case BPF_LDX | BPF_ABS | BPF_W:
 844                         /* A = *(u32 *) (ctx + K) */
 845                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 846                         break;
 847
 848                 /* Unknown instruction. */
 849                 default:
 850                         goto err;
 851                 }
 852
 853                 insn++;
 854                 if (new_prog)
 855                         memcpy(new_insn, tmp_insns,
 856                                sizeof(*insn) * (insn - tmp_insns));
 857                 new_insn += insn - tmp_insns;
 858         }
 859
 860         if (!new_prog) {
 861                 /* Only calculating new length. */
 862                 *new_len = new_insn - first_insn;
 863                 if (*seen_ld_abs)
 864                         *new_len += 4; /* Prologue bits. */
 865                 return 0;
 866         }
 867
 868         pass++;
 869         if (new_flen != new_insn - first_insn) {
 870                 new_flen = new_insn - first_insn;
 871                 if (pass > 2)
 872                         goto err;
 873                 goto do_pass;
 874         }
 875
 876         kfree(addrs);
 877         BUG_ON(*new_len != new_flen);
 878         return 0;
 879 err:
 880         kfree(addrs);
 881         return -EINVAL;
 882 }
 883
 884 /* Security:
 885  *
 886  * As we dont want to clear mem[] array for each packet going through
 887  * __bpf_prog_run(), we check that filter loaded by user never try to read
 888  * a cell if not previously written, and we check all branches to be sure
 889  * a malicious user doesn't try to abuse us.
 890  */
 891 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 892 {
 893         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 894         int pc, ret = 0;
 895
 896         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 897
 898         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 899         if (!masks)
 900                 return -ENOMEM;
 901
 902         memset(masks, 0xff, flen * sizeof(*masks));
 903
 904         for (pc = 0; pc < flen; pc++) {
 905                 memvalid &= masks[pc];
 906
 907                 switch (filter[pc].code) {
 908                 case BPF_ST:
 909                 case BPF_STX:
 910                         memvalid |= (1 << filter[pc].k);
 911                         break;
 912                 case BPF_LD | BPF_MEM:
 913                 case BPF_LDX | BPF_MEM:
 914                         if (!(memvalid & (1 << filter[pc].k))) {
 915                                 ret = -EINVAL;
 916                                 goto error;
 917                         }
 918                         break;
 919                 case BPF_JMP | BPF_JA:
 920                         /* A jump must set masks on target */
 921                         masks[pc + 1 + filter[pc].k] &= memvalid;
 922                         memvalid = ~0;
 923                         break;
 924                 case BPF_JMP | BPF_JEQ | BPF_K:
 925                 case BPF_JMP | BPF_JEQ | BPF_X:
 926                 case BPF_JMP | BPF_JGE | BPF_K:
 927                 case BPF_JMP | BPF_JGE | BPF_X:
 928                 case BPF_JMP | BPF_JGT | BPF_K:
 929                 case BPF_JMP | BPF_JGT | BPF_X:
 930                 case BPF_JMP | BPF_JSET | BPF_K:
 931                 case BPF_JMP | BPF_JSET | BPF_X:
 932                         /* A jump must set masks on targets */
 933                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 934                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 935                         memvalid = ~0;
 936                         break;
 937                 }
 938         }
 939 error:
 940         kfree(masks);
 941         return ret;
 942 }
 943
 944 static bool chk_code_allowed(u16 code_to_probe)
 945 {
 946         static const bool codes[] = {
 947                 /* 32 bit ALU operations */
 948                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 949                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 950                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 951                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 952                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 953                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 954                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 955                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 956                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 957                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 958                 [BPF_ALU | BPF_AND | BPF_K] = true,
 959                 [BPF_ALU | BPF_AND | BPF_X] = true,
 960                 [BPF_ALU | BPF_OR | BPF_K] = true,
 961                 [BPF_ALU | BPF_OR | BPF_X] = true,
 962                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 963                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 964                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 965                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 966                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 967                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 968                 [BPF_ALU | BPF_NEG] = true,
 969                 /* Load instructions */
 970                 [BPF_LD | BPF_W | BPF_ABS] = true,
 971                 [BPF_LD | BPF_H | BPF_ABS] = true,
 972                 [BPF_LD | BPF_B | BPF_ABS] = true,
 973                 [BPF_LD | BPF_W | BPF_LEN] = true,
 974                 [BPF_LD | BPF_W | BPF_IND] = true,
 975                 [BPF_LD | BPF_H | BPF_IND] = true,
 976                 [BPF_LD | BPF_B | BPF_IND] = true,
 977                 [BPF_LD | BPF_IMM] = true,
 978                 [BPF_LD | BPF_MEM] = true,
 979                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 980                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 981                 [BPF_LDX | BPF_IMM] = true,
 982                 [BPF_LDX | BPF_MEM] = true,
 983                 /* Store instructions */
 984                 [BPF_ST] = true,
 985                 [BPF_STX] = true,
 986                 /* Misc instructions */
 987                 [BPF_MISC | BPF_TAX] = true,
 988                 [BPF_MISC | BPF_TXA] = true,
 989                 /* Return instructions */
 990                 [BPF_RET | BPF_K] = true,
 991                 [BPF_RET | BPF_A] = true,
 992                 /* Jump instructions */
 993                 [BPF_JMP | BPF_JA] = true,
 994                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
 995                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
 996                 [BPF_JMP | BPF_JGE | BPF_K] = true,
 997                 [BPF_JMP | BPF_JGE | BPF_X] = true,
 998                 [BPF_JMP | BPF_JGT | BPF_K] = true,
 999                 [BPF_JMP | BPF_JGT | BPF_X] = true,
1000                 [BPF_JMP | BPF_JSET | BPF_K] = true,
1001                 [BPF_JMP | BPF_JSET | BPF_X] = true,
1002         };
1003
1004         if (code_to_probe >= ARRAY_SIZE(codes))
1005                 return false;
1006
1007         return codes[code_to_probe];
1008 }
1009
1010 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1011                                 unsigned int flen)
1012 {
1013         if (filter == NULL)
1014                 return false;
1015         if (flen == 0 || flen > BPF_MAXINSNS)
1016                 return false;
1017
1018         return true;
1019 }
1020
1021 /**
1022  *      bpf_check_classic - verify socket filter code
1023  *      @filter: filter to verify
1024  *      @flen: length of filter
1025  *
1026  * Check the user's filter code. If we let some ugly
1027  * filter code slip through kaboom! The filter must contain
1028  * no references or jumps that are out of range, no illegal
1029  * instructions, and must end with a RET instruction.
1030  *
1031  * All jumps are forward as they are not signed.
1032  *
1033  * Returns 0 if the rule set is legal or -EINVAL if not.
1034  */
1035 static int bpf_check_classic(const struct sock_filter *filter,
1036                              unsigned int flen)
1037 {
1038         bool anc_found;
1039         int pc;
1040
1041         /* Check the filter code now */
1042         for (pc = 0; pc < flen; pc++) {
1043                 const struct sock_filter *ftest = &filter[pc];
1044
1045                 /* May we actually operate on this code? */
1046                 if (!chk_code_allowed(ftest->code))
1047                         return -EINVAL;
1048
1049                 /* Some instructions need special checks */
1050                 switch (ftest->code) {
1051                 case BPF_ALU | BPF_DIV | BPF_K:
1052                 case BPF_ALU | BPF_MOD | BPF_K:
1053                         /* Check for division by zero */
1054                         if (ftest->k == 0)
1055                                 return -EINVAL;
1056                         break;
1057                 case BPF_ALU | BPF_LSH | BPF_K:
1058                 case BPF_ALU | BPF_RSH | BPF_K:
1059                         if (ftest->k >= 32)
1060                                 return -EINVAL;
1061                         break;
1062                 case BPF_LD | BPF_MEM:
1063                 case BPF_LDX | BPF_MEM:
1064                 case BPF_ST:
1065                 case BPF_STX:
1066                         /* Check for invalid memory addresses */
1067                         if (ftest->k >= BPF_MEMWORDS)
1068                                 return -EINVAL;
1069                         break;
1070                 case BPF_JMP | BPF_JA:
1071                         /* Note, the large ftest->k might cause loops.
1072                          * Compare this with conditional jumps below,
1073                          * where offsets are limited. --ANK (981016)
1074                          */
1075                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1076                                 return -EINVAL;
1077                         break;
1078                 case BPF_JMP | BPF_JEQ | BPF_K:
1079                 case BPF_JMP | BPF_JEQ | BPF_X:
1080                 case BPF_JMP | BPF_JGE | BPF_K:
1081                 case BPF_JMP | BPF_JGE | BPF_X:
1082                 case BPF_JMP | BPF_JGT | BPF_K:
1083                 case BPF_JMP | BPF_JGT | BPF_X:
1084                 case BPF_JMP | BPF_JSET | BPF_K:
1085                 case BPF_JMP | BPF_JSET | BPF_X:
1086                         /* Both conditionals must be safe */
1087                         if (pc + ftest->jt + 1 >= flen ||
1088                             pc + ftest->jf + 1 >= flen)
1089                                 return -EINVAL;
1090                         break;
1091                 case BPF_LD | BPF_W | BPF_ABS:
1092                 case BPF_LD | BPF_H | BPF_ABS:
1093                 case BPF_LD | BPF_B | BPF_ABS:
1094                         anc_found = false;
1095                         if (bpf_anc_helper(ftest) & BPF_ANC)
1096                                 anc_found = true;
1097                         /* Ancillary operation unknown or unsupported */
1098                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1099                                 return -EINVAL;
1100                 }
1101         }
1102
1103         /* Last instruction must be a RET code */
1104         switch (filter[flen - 1].code) {
1105         case BPF_RET | BPF_K:
1106         case BPF_RET | BPF_A:
1107                 return check_load_and_stores(filter, flen);
1108         }
1109
1110         return -EINVAL;
1111 }
1112
1113 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1114                                       const struct sock_fprog *fprog)
1115 {
1116         unsigned int fsize = bpf_classic_proglen(fprog);
1117         struct sock_fprog_kern *fkprog;
1118
1119         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1120         if (!fp->orig_prog)
1121                 return -ENOMEM;
1122
1123         fkprog = fp->orig_prog;
1124         fkprog->len = fprog->len;
1125
1126         fkprog->filter = kmemdup(fp->insns, fsize,
1127                                  GFP_KERNEL | __GFP_NOWARN);
1128         if (!fkprog->filter) {
1129                 kfree(fp->orig_prog);
1130                 return -ENOMEM;
1131         }
1132
1133         return 0;
1134 }
1135
1136 static void bpf_release_orig_filter(struct bpf_prog *fp)
1137 {
1138         struct sock_fprog_kern *fprog = fp->orig_prog;
1139
1140         if (fprog) {
1141                 kfree(fprog->filter);
1142                 kfree(fprog);
1143         }
1144 }
1145
1146 static void __bpf_prog_release(struct bpf_prog *prog)
1147 {
1148         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1149                 bpf_prog_put(prog);
1150         } else {
1151                 bpf_release_orig_filter(prog);
1152                 bpf_prog_free(prog);
1153         }
1154 }
1155
1156 static void __sk_filter_release(struct sk_filter *fp)
1157 {
1158         __bpf_prog_release(fp->prog);
1159         kfree(fp);
1160 }
1161
1162 /**
1163  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1164  *      @rcu: rcu_head that contains the sk_filter to free
1165  */
1166 static void sk_filter_release_rcu(struct rcu_head *rcu)
1167 {
1168         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1169
1170         __sk_filter_release(fp);
1171 }
1172
1173 /**
1174  *      sk_filter_release - release a socket filter
1175  *      @fp: filter to remove
1176  *
1177  *      Remove a filter from a socket and release its resources.
1178  */
1179 static void sk_filter_release(struct sk_filter *fp)
1180 {
1181         if (refcount_dec_and_test(&fp->refcnt))
1182                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1183 }
1184
1185 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1186 {
1187         u32 filter_size = bpf_prog_size(fp->prog->len);
1188
1189         atomic_sub(filter_size, &sk->sk_omem_alloc);
1190         sk_filter_release(fp);
1191 }
1192
1193 /* try to charge the socket memory if there is space available
1194  * return true on success
1195  */
1196 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1197 {
1198         u32 filter_size = bpf_prog_size(fp->prog->len);
1199
1200         /* same check as in sock_kmalloc() */
1201         if (filter_size <= sysctl_optmem_max &&
1202             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1203                 atomic_add(filter_size, &sk->sk_omem_alloc);
1204                 return true;
1205         }
1206         return false;
1207 }
1208
1209 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1210 {
1211         if (!refcount_inc_not_zero(&fp->refcnt))
1212                 return false;
1213
1214         if (!__sk_filter_charge(sk, fp)) {
1215                 sk_filter_release(fp);
1216                 return false;
1217         }
1218         return true;
1219 }
1220
1221 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1222 {
1223         struct sock_filter *old_prog;
1224         struct bpf_prog *old_fp;
1225         int err, new_len, old_len = fp->len;
1226         bool seen_ld_abs = false;
1227
1228         /* We are free to overwrite insns et al right here as it
1229          * won't be used at this point in time anymore internally
1230          * after the migration to the internal BPF instruction
1231          * representation.
1232          */
1233         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1234                      sizeof(struct bpf_insn));
1235
1236         /* Conversion cannot happen on overlapping memory areas,
1237          * so we need to keep the user BPF around until the 2nd
1238          * pass. At this time, the user BPF is stored in fp->insns.
1239          */
1240         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1241                            GFP_KERNEL | __GFP_NOWARN);
1242         if (!old_prog) {
1243                 err = -ENOMEM;
1244                 goto out_err;
1245         }
1246
1247         /* 1st pass: calculate the new program length. */
1248         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1249                                  &seen_ld_abs);
1250         if (err)
1251                 goto out_err_free;
1252
1253         /* Expand fp for appending the new filter representation. */
1254         old_fp = fp;
1255         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1256         if (!fp) {
1257                 /* The old_fp is still around in case we couldn't
1258                  * allocate new memory, so uncharge on that one.
1259                  */
1260                 fp = old_fp;
1261                 err = -ENOMEM;
1262                 goto out_err_free;
1263         }
1264
1265         fp->len = new_len;
1266
1267         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1268         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1269                                  &seen_ld_abs);
1270         if (err)
1271                 /* 2nd bpf_convert_filter() can fail only if it fails
1272                  * to allocate memory, remapping must succeed. Note,
1273                  * that at this time old_fp has already been released
1274                  * by krealloc().
1275                  */
1276                 goto out_err_free;
1277
1278         fp = bpf_prog_select_runtime(fp, &err);
1279         if (err)
1280                 goto out_err_free;
1281
1282         kfree(old_prog);
1283         return fp;
1284
1285 out_err_free:
1286         kfree(old_prog);
1287 out_err:
1288         __bpf_prog_release(fp);
1289         return ERR_PTR(err);
1290 }
1291
1292 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1293                                            bpf_aux_classic_check_t trans)
1294 {
1295         int err;
1296
1297         fp->bpf_func = NULL;
1298         fp->jited = 0;
1299
1300         err = bpf_check_classic(fp->insns, fp->len);
1301         if (err) {
1302                 __bpf_prog_release(fp);
1303                 return ERR_PTR(err);
1304         }
1305
1306         /* There might be additional checks and transformations
1307          * needed on classic filters, f.e. in case of seccomp.
1308          */
1309         if (trans) {
1310                 err = trans(fp->insns, fp->len);
1311                 if (err) {
1312                         __bpf_prog_release(fp);
1313                         return ERR_PTR(err);
1314                 }
1315         }
1316
1317         /* Probe if we can JIT compile the filter and if so, do
1318          * the compilation of the filter.
1319          */
1320         bpf_jit_compile(fp);
1321
1322         /* JIT compiler couldn't process this filter, so do the
1323          * internal BPF translation for the optimized interpreter.
1324          */
1325         if (!fp->jited)
1326                 fp = bpf_migrate_filter(fp);
1327
1328         return fp;
1329 }
1330
1331 /**
1332  *      bpf_prog_create - create an unattached filter
1333  *      @pfp: the unattached filter that is created
1334  *      @fprog: the filter program
1335  *
1336  * Create a filter independent of any socket. We first run some
1337  * sanity checks on it to make sure it does not explode on us later.
1338  * If an error occurs or there is insufficient memory for the filter
1339  * a negative errno code is returned. On success the return is zero.
1340  */
1341 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1342 {
1343         unsigned int fsize = bpf_classic_proglen(fprog);
1344         struct bpf_prog *fp;
1345
1346         /* Make sure new filter is there and in the right amounts. */
1347         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1348                 return -EINVAL;
1349
1350         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1351         if (!fp)
1352                 return -ENOMEM;
1353
1354         memcpy(fp->insns, fprog->filter, fsize);
1355
1356         fp->len = fprog->len;
1357         /* Since unattached filters are not copied back to user
1358          * space through sk_get_filter(), we do not need to hold
1359          * a copy here, and can spare us the work.
1360          */
1361         fp->orig_prog = NULL;
1362
1363         /* bpf_prepare_filter() already takes care of freeing
1364          * memory in case something goes wrong.
1365          */
1366         fp = bpf_prepare_filter(fp, NULL);
1367         if (IS_ERR(fp))
1368                 return PTR_ERR(fp);
1369
1370         *pfp = fp;
1371         return 0;
1372 }
1373 EXPORT_SYMBOL_GPL(bpf_prog_create);
1374
1375 /**
1376  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1377  *      @pfp: the unattached filter that is created
1378  *      @fprog: the filter program
1379  *      @trans: post-classic verifier transformation handler
1380  *      @save_orig: save classic BPF program
1381  *
1382  * This function effectively does the same as bpf_prog_create(), only
1383  * that it builds up its insns buffer from user space provided buffer.
1384  * It also allows for passing a bpf_aux_classic_check_t handler.
1385  */
1386 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1387                               bpf_aux_classic_check_t trans, bool save_orig)
1388 {
1389         unsigned int fsize = bpf_classic_proglen(fprog);
1390         struct bpf_prog *fp;
1391         int err;
1392
1393         /* Make sure new filter is there and in the right amounts. */
1394         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1395                 return -EINVAL;
1396
1397         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1398         if (!fp)
1399                 return -ENOMEM;
1400
1401         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1402                 __bpf_prog_free(fp);
1403                 return -EFAULT;
1404         }
1405
1406         fp->len = fprog->len;
1407         fp->orig_prog = NULL;
1408
1409         if (save_orig) {
1410                 err = bpf_prog_store_orig_filter(fp, fprog);
1411                 if (err) {
1412                         __bpf_prog_free(fp);
1413                         return -ENOMEM;
1414                 }
1415         }
1416
1417         /* bpf_prepare_filter() already takes care of freeing
1418          * memory in case something goes wrong.
1419          */
1420         fp = bpf_prepare_filter(fp, trans);
1421         if (IS_ERR(fp))
1422                 return PTR_ERR(fp);
1423
1424         *pfp = fp;
1425         return 0;
1426 }
1427 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1428
1429 void bpf_prog_destroy(struct bpf_prog *fp)
1430 {
1431         __bpf_prog_release(fp);
1432 }
1433 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1434
1435 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1436 {
1437         struct sk_filter *fp, *old_fp;
1438
1439         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1440         if (!fp)
1441                 return -ENOMEM;
1442
1443         fp->prog = prog;
1444
1445         if (!__sk_filter_charge(sk, fp)) {
1446                 kfree(fp);
1447                 return -ENOMEM;
1448         }
1449         refcount_set(&fp->refcnt, 1);
1450
1451         old_fp = rcu_dereference_protected(sk->sk_filter,
1452                                            lockdep_sock_is_held(sk));
1453         rcu_assign_pointer(sk->sk_filter, fp);
1454
1455         if (old_fp)
1456                 sk_filter_uncharge(sk, old_fp);
1457
1458         return 0;
1459 }
1460
1461 static
1462 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1463 {
1464         unsigned int fsize = bpf_classic_proglen(fprog);
1465         struct bpf_prog *prog;
1466         int err;
1467
1468         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1469                 return ERR_PTR(-EPERM);
1470
1471         /* Make sure new filter is there and in the right amounts. */
1472         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1473                 return ERR_PTR(-EINVAL);
1474
1475         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1476         if (!prog)
1477                 return ERR_PTR(-ENOMEM);
1478
1479         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1480                 __bpf_prog_free(prog);
1481                 return ERR_PTR(-EFAULT);
1482         }
1483
1484         prog->len = fprog->len;
1485
1486         err = bpf_prog_store_orig_filter(prog, fprog);
1487         if (err) {
1488                 __bpf_prog_free(prog);
1489                 return ERR_PTR(-ENOMEM);
1490         }
1491
1492         /* bpf_prepare_filter() already takes care of freeing
1493          * memory in case something goes wrong.
1494          */
1495         return bpf_prepare_filter(prog, NULL);
1496 }
1497
1498 /**
1499  *      sk_attach_filter - attach a socket filter
1500  *      @fprog: the filter program
1501  *      @sk: the socket to use
1502  *
1503  * Attach the user's filter code. We first run some sanity checks on
1504  * it to make sure it does not explode on us later. If an error
1505  * occurs or there is insufficient memory for the filter a negative
1506  * errno code is returned. On success the return is zero.
1507  */
1508 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1509 {
1510         struct bpf_prog *prog = __get_filter(fprog, sk);
1511         int err;
1512
1513         if (IS_ERR(prog))
1514                 return PTR_ERR(prog);
1515
1516         err = __sk_attach_prog(prog, sk);
1517         if (err < 0) {
1518                 __bpf_prog_release(prog);
1519                 return err;
1520         }
1521
1522         return 0;
1523 }
1524 EXPORT_SYMBOL_GPL(sk_attach_filter);
1525
1526 int sk_reuseport_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         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1535                 err = -ENOMEM;
1536         else
1537                 err = reuseport_attach_prog(sk, prog);
1538
1539         if (err)
1540                 __bpf_prog_release(prog);
1541
1542         return err;
1543 }
1544
1545 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1546 {
1547         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1548                 return ERR_PTR(-EPERM);
1549
1550         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1551 }
1552
1553 int sk_attach_bpf(u32 ufd, struct sock *sk)
1554 {
1555         struct bpf_prog *prog = __get_bpf(ufd, sk);
1556         int err;
1557
1558         if (IS_ERR(prog))
1559                 return PTR_ERR(prog);
1560
1561         err = __sk_attach_prog(prog, sk);
1562         if (err < 0) {
1563                 bpf_prog_put(prog);
1564                 return err;
1565         }
1566
1567         return 0;
1568 }
1569
1570 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1571 {
1572         struct bpf_prog *prog;
1573         int err;
1574
1575         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1576                 return -EPERM;
1577
1578         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1579         if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1580                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1581         if (IS_ERR(prog))
1582                 return PTR_ERR(prog);
1583
1584         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1585                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1586                  * bpf prog (e.g. sockmap).  It depends on the
1587                  * limitation imposed by bpf_prog_load().
1588                  * Hence, sysctl_optmem_max is not checked.
1589                  */
1590                 if ((sk->sk_type != SOCK_STREAM &&
1591                      sk->sk_type != SOCK_DGRAM) ||
1592                     (sk->sk_protocol != IPPROTO_UDP &&
1593                      sk->sk_protocol != IPPROTO_TCP) ||
1594                     (sk->sk_family != AF_INET &&
1595                      sk->sk_family != AF_INET6)) {
1596                         err = -ENOTSUPP;
1597                         goto err_prog_put;
1598                 }
1599         } else {
1600                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1601                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1602                         err = -ENOMEM;
1603                         goto err_prog_put;
1604                 }
1605         }
1606
1607         err = reuseport_attach_prog(sk, prog);
1608 err_prog_put:
1609         if (err)
1610                 bpf_prog_put(prog);
1611
1612         return err;
1613 }
1614
1615 void sk_reuseport_prog_free(struct bpf_prog *prog)
1616 {
1617         if (!prog)
1618                 return;
1619
1620         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1621                 bpf_prog_put(prog);
1622         else
1623                 bpf_prog_destroy(prog);
1624 }
1625
1626 struct bpf_scratchpad {
1627         union {
1628                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1629                 u8     buff[MAX_BPF_STACK];
1630         };
1631 };
1632
1633 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1634
1635 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1636                                           unsigned int write_len)
1637 {
1638         return skb_ensure_writable(skb, write_len);
1639 }
1640
1641 static inline int bpf_try_make_writable(struct sk_buff *skb,
1642                                         unsigned int write_len)
1643 {
1644         int err = __bpf_try_make_writable(skb, write_len);
1645
1646         bpf_compute_data_pointers(skb);
1647         return err;
1648 }
1649
1650 static int bpf_try_make_head_writable(struct sk_buff *skb)
1651 {
1652         return bpf_try_make_writable(skb, skb_headlen(skb));
1653 }
1654
1655 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1656 {
1657         if (skb_at_tc_ingress(skb))
1658                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1659 }
1660
1661 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1662 {
1663         if (skb_at_tc_ingress(skb))
1664                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1665 }
1666
1667 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1668            const void *, from, u32, len, u64, flags)
1669 {
1670         void *ptr;
1671
1672         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1673                 return -EINVAL;
1674         if (unlikely(offset > 0xffff))
1675                 return -EFAULT;
1676         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1677                 return -EFAULT;
1678
1679         ptr = skb->data + offset;
1680         if (flags & BPF_F_RECOMPUTE_CSUM)
1681                 __skb_postpull_rcsum(skb, ptr, len, offset);
1682
1683         memcpy(ptr, from, len);
1684
1685         if (flags & BPF_F_RECOMPUTE_CSUM)
1686                 __skb_postpush_rcsum(skb, ptr, len, offset);
1687         if (flags & BPF_F_INVALIDATE_HASH)
1688                 skb_clear_hash(skb);
1689
1690         return 0;
1691 }
1692
1693 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1694         .func           = bpf_skb_store_bytes,
1695         .gpl_only       = false,
1696         .ret_type       = RET_INTEGER,
1697         .arg1_type      = ARG_PTR_TO_CTX,
1698         .arg2_type      = ARG_ANYTHING,
1699         .arg3_type      = ARG_PTR_TO_MEM,
1700         .arg4_type      = ARG_CONST_SIZE,
1701         .arg5_type      = ARG_ANYTHING,
1702 };
1703
1704 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1705            void *, to, u32, len)
1706 {
1707         void *ptr;
1708
1709         if (unlikely(offset > 0xffff))
1710                 goto err_clear;
1711
1712         ptr = skb_header_pointer(skb, offset, len, to);
1713         if (unlikely(!ptr))
1714                 goto err_clear;
1715         if (ptr != to)
1716                 memcpy(to, ptr, len);
1717
1718         return 0;
1719 err_clear:
1720         memset(to, 0, len);
1721         return -EFAULT;
1722 }
1723
1724 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1725         .func           = bpf_skb_load_bytes,
1726         .gpl_only       = false,
1727         .ret_type       = RET_INTEGER,
1728         .arg1_type      = ARG_PTR_TO_CTX,
1729         .arg2_type      = ARG_ANYTHING,
1730         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1731         .arg4_type      = ARG_CONST_SIZE,
1732 };
1733
1734 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1735            u32, offset, void *, to, u32, len, u32, start_header)
1736 {
1737         u8 *end = skb_tail_pointer(skb);
1738         u8 *net = skb_network_header(skb);
1739         u8 *mac = skb_mac_header(skb);
1740         u8 *ptr;
1741
1742         if (unlikely(offset > 0xffff || len > (end - mac)))
1743                 goto err_clear;
1744
1745         switch (start_header) {
1746         case BPF_HDR_START_MAC:
1747                 ptr = mac + offset;
1748                 break;
1749         case BPF_HDR_START_NET:
1750                 ptr = net + offset;
1751                 break;
1752         default:
1753                 goto err_clear;
1754         }
1755
1756         if (likely(ptr >= mac && ptr + len <= end)) {
1757                 memcpy(to, ptr, len);
1758                 return 0;
1759         }
1760
1761 err_clear:
1762         memset(to, 0, len);
1763         return -EFAULT;
1764 }
1765
1766 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1767         .func           = bpf_skb_load_bytes_relative,
1768         .gpl_only       = false,
1769         .ret_type       = RET_INTEGER,
1770         .arg1_type      = ARG_PTR_TO_CTX,
1771         .arg2_type      = ARG_ANYTHING,
1772         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1773         .arg4_type      = ARG_CONST_SIZE,
1774         .arg5_type      = ARG_ANYTHING,
1775 };
1776
1777 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1778 {
1779         /* Idea is the following: should the needed direct read/write
1780          * test fail during runtime, we can pull in more data and redo
1781          * again, since implicitly, we invalidate previous checks here.
1782          *
1783          * Or, since we know how much we need to make read/writeable,
1784          * this can be done once at the program beginning for direct
1785          * access case. By this we overcome limitations of only current
1786          * headroom being accessible.
1787          */
1788         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1789 }
1790
1791 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1792         .func           = bpf_skb_pull_data,
1793         .gpl_only       = false,
1794         .ret_type       = RET_INTEGER,
1795         .arg1_type      = ARG_PTR_TO_CTX,
1796         .arg2_type      = ARG_ANYTHING,
1797 };
1798
1799 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1800                                            unsigned int write_len)
1801 {
1802         int err = __bpf_try_make_writable(skb, write_len);
1803
1804         bpf_compute_data_end_sk_skb(skb);
1805         return err;
1806 }
1807
1808 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1809 {
1810         /* Idea is the following: should the needed direct read/write
1811          * test fail during runtime, we can pull in more data and redo
1812          * again, since implicitly, we invalidate previous checks here.
1813          *
1814          * Or, since we know how much we need to make read/writeable,
1815          * this can be done once at the program beginning for direct
1816          * access case. By this we overcome limitations of only current
1817          * headroom being accessible.
1818          */
1819         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1820 }
1821
1822 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1823         .func           = sk_skb_pull_data,
1824         .gpl_only       = false,
1825         .ret_type       = RET_INTEGER,
1826         .arg1_type      = ARG_PTR_TO_CTX,
1827         .arg2_type      = ARG_ANYTHING,
1828 };
1829
1830 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1831            u64, from, u64, to, u64, flags)
1832 {
1833         __sum16 *ptr;
1834
1835         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1836                 return -EINVAL;
1837         if (unlikely(offset > 0xffff || offset & 1))
1838                 return -EFAULT;
1839         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1840                 return -EFAULT;
1841
1842         ptr = (__sum16 *)(skb->data + offset);
1843         switch (flags & BPF_F_HDR_FIELD_MASK) {
1844         case 0:
1845                 if (unlikely(from != 0))
1846                         return -EINVAL;
1847
1848                 csum_replace_by_diff(ptr, to);
1849                 break;
1850         case 2:
1851                 csum_replace2(ptr, from, to);
1852                 break;
1853         case 4:
1854                 csum_replace4(ptr, from, to);
1855                 break;
1856         default:
1857                 return -EINVAL;
1858         }
1859
1860         return 0;
1861 }
1862
1863 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1864         .func           = bpf_l3_csum_replace,
1865         .gpl_only       = false,
1866         .ret_type       = RET_INTEGER,
1867         .arg1_type      = ARG_PTR_TO_CTX,
1868         .arg2_type      = ARG_ANYTHING,
1869         .arg3_type      = ARG_ANYTHING,
1870         .arg4_type      = ARG_ANYTHING,
1871         .arg5_type      = ARG_ANYTHING,
1872 };
1873
1874 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1875            u64, from, u64, to, u64, flags)
1876 {
1877         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1878         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1879         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1880         __sum16 *ptr;
1881
1882         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1883                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1884                 return -EINVAL;
1885         if (unlikely(offset > 0xffff || offset & 1))
1886                 return -EFAULT;
1887         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1888                 return -EFAULT;
1889
1890         ptr = (__sum16 *)(skb->data + offset);
1891         if (is_mmzero && !do_mforce && !*ptr)
1892                 return 0;
1893
1894         switch (flags & BPF_F_HDR_FIELD_MASK) {
1895         case 0:
1896                 if (unlikely(from != 0))
1897                         return -EINVAL;
1898
1899                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1900                 break;
1901         case 2:
1902                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1903                 break;
1904         case 4:
1905                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1906                 break;
1907         default:
1908                 return -EINVAL;
1909         }
1910
1911         if (is_mmzero && !*ptr)
1912                 *ptr = CSUM_MANGLED_0;
1913         return 0;
1914 }
1915
1916 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1917         .func           = bpf_l4_csum_replace,
1918         .gpl_only       = false,
1919         .ret_type       = RET_INTEGER,
1920         .arg1_type      = ARG_PTR_TO_CTX,
1921         .arg2_type      = ARG_ANYTHING,
1922         .arg3_type      = ARG_ANYTHING,
1923         .arg4_type      = ARG_ANYTHING,
1924         .arg5_type      = ARG_ANYTHING,
1925 };
1926
1927 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1928            __be32 *, to, u32, to_size, __wsum, seed)
1929 {
1930         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1931         u32 diff_size = from_size + to_size;
1932         int i, j = 0;
1933
1934         /* This is quite flexible, some examples:
1935          *
1936          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1937          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1938          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1939          *
1940          * Even for diffing, from_size and to_size don't need to be equal.
1941          */
1942         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1943                      diff_size > sizeof(sp->diff)))
1944                 return -EINVAL;
1945
1946         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1947                 sp->diff[j] = ~from[i];
1948         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1949                 sp->diff[j] = to[i];
1950
1951         return csum_partial(sp->diff, diff_size, seed);
1952 }
1953
1954 static const struct bpf_func_proto bpf_csum_diff_proto = {
1955         .func           = bpf_csum_diff,
1956         .gpl_only       = false,
1957         .pkt_access     = true,
1958         .ret_type       = RET_INTEGER,
1959         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1960         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1961         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
1962         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1963         .arg5_type      = ARG_ANYTHING,
1964 };
1965
1966 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1967 {
1968         /* The interface is to be used in combination with bpf_csum_diff()
1969          * for direct packet writes. csum rotation for alignment as well
1970          * as emulating csum_sub() can be done from the eBPF program.
1971          */
1972         if (skb->ip_summed == CHECKSUM_COMPLETE)
1973                 return (skb->csum = csum_add(skb->csum, csum));
1974
1975         return -ENOTSUPP;
1976 }
1977
1978 static const struct bpf_func_proto bpf_csum_update_proto = {
1979         .func           = bpf_csum_update,
1980         .gpl_only       = false,
1981         .ret_type       = RET_INTEGER,
1982         .arg1_type      = ARG_PTR_TO_CTX,
1983         .arg2_type      = ARG_ANYTHING,
1984 };
1985
1986 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1987 {
1988         return dev_forward_skb(dev, skb);
1989 }
1990
1991 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1992                                       struct sk_buff *skb)
1993 {
1994         int ret = ____dev_forward_skb(dev, skb);
1995
1996         if (likely(!ret)) {
1997                 skb->dev = dev;
1998                 ret = netif_rx(skb);
1999         }
2000
2001         return ret;
2002 }
2003
2004 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2005 {
2006         int ret;
2007
2008         if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2009                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2010                 kfree_skb(skb);
2011                 return -ENETDOWN;
2012         }
2013
2014         skb->dev = dev;
2015
2016         __this_cpu_inc(xmit_recursion);
2017         ret = dev_queue_xmit(skb);
2018         __this_cpu_dec(xmit_recursion);
2019
2020         return ret;
2021 }
2022
2023 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2024                                  u32 flags)
2025 {
2026         /* skb->mac_len is not set on normal egress */
2027         unsigned int mlen = skb->network_header - skb->mac_header;
2028
2029         __skb_pull(skb, mlen);
2030
2031         /* At ingress, the mac header has already been pulled once.
2032          * At egress, skb_pospull_rcsum has to be done in case that
2033          * the skb is originated from ingress (i.e. a forwarded skb)
2034          * to ensure that rcsum starts at net header.
2035          */
2036         if (!skb_at_tc_ingress(skb))
2037                 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2038         skb_pop_mac_header(skb);
2039         skb_reset_mac_len(skb);
2040         return flags & BPF_F_INGRESS ?
2041                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2042 }
2043
2044 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2045                                  u32 flags)
2046 {
2047         /* Verify that a link layer header is carried */
2048         if (unlikely(skb->mac_header >= skb->network_header)) {
2049                 kfree_skb(skb);
2050                 return -ERANGE;
2051         }
2052
2053         bpf_push_mac_rcsum(skb);
2054         return flags & BPF_F_INGRESS ?
2055                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2056 }
2057
2058 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2059                           u32 flags)
2060 {
2061         if (dev_is_mac_header_xmit(dev))
2062                 return __bpf_redirect_common(skb, dev, flags);
2063         else
2064                 return __bpf_redirect_no_mac(skb, dev, flags);
2065 }
2066
2067 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2068 {
2069         struct net_device *dev;
2070         struct sk_buff *clone;
2071         int ret;
2072
2073         if (unlikely(flags & ~(BPF_F_INGRESS)))
2074                 return -EINVAL;
2075
2076         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2077         if (unlikely(!dev))
2078                 return -EINVAL;
2079
2080         clone = skb_clone(skb, GFP_ATOMIC);
2081         if (unlikely(!clone))
2082                 return -ENOMEM;
2083
2084         /* For direct write, we need to keep the invariant that the skbs
2085          * we're dealing with need to be uncloned. Should uncloning fail
2086          * here, we need to free the just generated clone to unclone once
2087          * again.
2088          */
2089         ret = bpf_try_make_head_writable(skb);
2090         if (unlikely(ret)) {
2091                 kfree_skb(clone);
2092                 return -ENOMEM;
2093         }
2094
2095         return __bpf_redirect(clone, dev, flags);
2096 }
2097
2098 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2099         .func           = bpf_clone_redirect,
2100         .gpl_only       = false,
2101         .ret_type       = RET_INTEGER,
2102         .arg1_type      = ARG_PTR_TO_CTX,
2103         .arg2_type      = ARG_ANYTHING,
2104         .arg3_type      = ARG_ANYTHING,
2105 };
2106
2107 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2108 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2109
2110 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2111 {
2112         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2113
2114         if (unlikely(flags & ~(BPF_F_INGRESS)))
2115                 return TC_ACT_SHOT;
2116
2117         ri->ifindex = ifindex;
2118         ri->flags = flags;
2119
2120         return TC_ACT_REDIRECT;
2121 }
2122
2123 int skb_do_redirect(struct sk_buff *skb)
2124 {
2125         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2126         struct net_device *dev;
2127
2128         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2129         ri->ifindex = 0;
2130         if (unlikely(!dev)) {
2131                 kfree_skb(skb);
2132                 return -EINVAL;
2133         }
2134
2135         return __bpf_redirect(skb, dev, ri->flags);
2136 }
2137
2138 static const struct bpf_func_proto bpf_redirect_proto = {
2139         .func           = bpf_redirect,
2140         .gpl_only       = false,
2141         .ret_type       = RET_INTEGER,
2142         .arg1_type      = ARG_ANYTHING,
2143         .arg2_type      = ARG_ANYTHING,
2144 };
2145
2146 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes)
2147 {
2148         msg->apply_bytes = bytes;
2149         return 0;
2150 }
2151
2152 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2153         .func           = bpf_msg_apply_bytes,
2154         .gpl_only       = false,
2155         .ret_type       = RET_INTEGER,
2156         .arg1_type      = ARG_PTR_TO_CTX,
2157         .arg2_type      = ARG_ANYTHING,
2158 };
2159
2160 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes)
2161 {
2162         msg->cork_bytes = bytes;
2163         return 0;
2164 }
2165
2166 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2167         .func           = bpf_msg_cork_bytes,
2168         .gpl_only       = false,
2169         .ret_type       = RET_INTEGER,
2170         .arg1_type      = ARG_PTR_TO_CTX,
2171         .arg2_type      = ARG_ANYTHING,
2172 };
2173
2174 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start,
2175            u32, end, u64, flags)
2176 {
2177         u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start;
2178         u32 first_sge, last_sge, i, shift, bytes_sg_total;
2179         struct scatterlist *sge;
2180         u8 *raw, *to, *from;
2181         struct page *page;
2182
2183         if (unlikely(flags || end <= start))
2184                 return -EINVAL;
2185
2186         /* First find the starting scatterlist element */
2187         i = msg->sg.start;
2188         do {
2189                 len = sk_msg_elem(msg, i)->length;
2190                 if (start < offset + len)
2191                         break;
2192                 offset += len;
2193                 sk_msg_iter_var_next(i);
2194         } while (i != msg->sg.end);
2195
2196         if (unlikely(start >= offset + len))
2197                 return -EINVAL;
2198
2199         first_sge = i;
2200         /* The start may point into the sg element so we need to also
2201          * account for the headroom.
2202          */
2203         bytes_sg_total = start - offset + bytes;
2204         if (!msg->sg.copy[i] && bytes_sg_total <= len)
2205                 goto out;
2206
2207         /* At this point we need to linearize multiple scatterlist
2208          * elements or a single shared page. Either way we need to
2209          * copy into a linear buffer exclusively owned by BPF. Then
2210          * place the buffer in the scatterlist and fixup the original
2211          * entries by removing the entries now in the linear buffer
2212          * and shifting the remaining entries. For now we do not try
2213          * to copy partial entries to avoid complexity of running out
2214          * of sg_entry slots. The downside is reading a single byte
2215          * will copy the entire sg entry.
2216          */
2217         do {
2218                 copy += sk_msg_elem(msg, i)->length;
2219                 sk_msg_iter_var_next(i);
2220                 if (bytes_sg_total <= copy)
2221                         break;
2222         } while (i != msg->sg.end);
2223         last_sge = i;
2224
2225         if (unlikely(bytes_sg_total > copy))
2226                 return -EINVAL;
2227
2228         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2229                            get_order(copy));
2230         if (unlikely(!page))
2231                 return -ENOMEM;
2232
2233         raw = page_address(page);
2234         i = first_sge;
2235         do {
2236                 sge = sk_msg_elem(msg, i);
2237                 from = sg_virt(sge);
2238                 len = sge->length;
2239                 to = raw + poffset;
2240
2241                 memcpy(to, from, len);
2242                 poffset += len;
2243                 sge->length = 0;
2244                 put_page(sg_page(sge));
2245
2246                 sk_msg_iter_var_next(i);
2247         } while (i != last_sge);
2248
2249         sg_set_page(&msg->sg.data[first_sge], page, copy, 0);
2250
2251         /* To repair sg ring we need to shift entries. If we only
2252          * had a single entry though we can just replace it and
2253          * be done. Otherwise walk the ring and shift the entries.
2254          */
2255         WARN_ON_ONCE(last_sge == first_sge);
2256         shift = last_sge > first_sge ?
2257                 last_sge - first_sge - 1 :
2258                 MAX_SKB_FRAGS - first_sge + last_sge - 1;
2259         if (!shift)
2260                 goto out;
2261
2262         i = first_sge;
2263         sk_msg_iter_var_next(i);
2264         do {
2265                 u32 move_from;
2266
2267                 if (i + shift >= MAX_MSG_FRAGS)
2268                         move_from = i + shift - MAX_MSG_FRAGS;
2269                 else
2270                         move_from = i + shift;
2271                 if (move_from == msg->sg.end)
2272                         break;
2273
2274                 msg->sg.data[i] = msg->sg.data[move_from];
2275                 msg->sg.data[move_from].length = 0;
2276                 msg->sg.data[move_from].page_link = 0;
2277                 msg->sg.data[move_from].offset = 0;
2278                 sk_msg_iter_var_next(i);
2279         } while (1);
2280
2281         msg->sg.end = msg->sg.end - shift > msg->sg.end ?
2282                       msg->sg.end - shift + MAX_MSG_FRAGS :
2283                       msg->sg.end - shift;
2284 out:
2285         msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset;
2286         msg->data_end = msg->data + bytes;
2287         return 0;
2288 }
2289
2290 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2291         .func           = bpf_msg_pull_data,
2292         .gpl_only       = false,
2293         .ret_type       = RET_INTEGER,
2294         .arg1_type      = ARG_PTR_TO_CTX,
2295         .arg2_type      = ARG_ANYTHING,
2296         .arg3_type      = ARG_ANYTHING,
2297         .arg4_type      = ARG_ANYTHING,
2298 };
2299
2300 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2301 {
2302         return task_get_classid(skb);
2303 }
2304
2305 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2306         .func           = bpf_get_cgroup_classid,
2307         .gpl_only       = false,
2308         .ret_type       = RET_INTEGER,
2309         .arg1_type      = ARG_PTR_TO_CTX,
2310 };
2311
2312 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2313 {
2314         return dst_tclassid(skb);
2315 }
2316
2317 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2318         .func           = bpf_get_route_realm,
2319         .gpl_only       = false,
2320         .ret_type       = RET_INTEGER,
2321         .arg1_type      = ARG_PTR_TO_CTX,
2322 };
2323
2324 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2325 {
2326         /* If skb_clear_hash() was called due to mangling, we can
2327          * trigger SW recalculation here. Later access to hash
2328          * can then use the inline skb->hash via context directly
2329          * instead of calling this helper again.
2330          */
2331         return skb_get_hash(skb);
2332 }
2333
2334 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2335         .func           = bpf_get_hash_recalc,
2336         .gpl_only       = false,
2337         .ret_type       = RET_INTEGER,
2338         .arg1_type      = ARG_PTR_TO_CTX,
2339 };
2340
2341 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2342 {
2343         /* After all direct packet write, this can be used once for
2344          * triggering a lazy recalc on next skb_get_hash() invocation.
2345          */
2346         skb_clear_hash(skb);
2347         return 0;
2348 }
2349
2350 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2351         .func           = bpf_set_hash_invalid,
2352         .gpl_only       = false,
2353         .ret_type       = RET_INTEGER,
2354         .arg1_type      = ARG_PTR_TO_CTX,
2355 };
2356
2357 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2358 {
2359         /* Set user specified hash as L4(+), so that it gets returned
2360          * on skb_get_hash() call unless BPF prog later on triggers a
2361          * skb_clear_hash().
2362          */
2363         __skb_set_sw_hash(skb, hash, true);
2364         return 0;
2365 }
2366
2367 static const struct bpf_func_proto bpf_set_hash_proto = {
2368         .func           = bpf_set_hash,
2369         .gpl_only       = false,
2370         .ret_type       = RET_INTEGER,
2371         .arg1_type      = ARG_PTR_TO_CTX,
2372         .arg2_type      = ARG_ANYTHING,
2373 };
2374
2375 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2376            u16, vlan_tci)
2377 {
2378         int ret;
2379
2380         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2381                      vlan_proto != htons(ETH_P_8021AD)))
2382                 vlan_proto = htons(ETH_P_8021Q);
2383
2384         bpf_push_mac_rcsum(skb);
2385         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2386         bpf_pull_mac_rcsum(skb);
2387
2388         bpf_compute_data_pointers(skb);
2389         return ret;
2390 }
2391
2392 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2393         .func           = bpf_skb_vlan_push,
2394         .gpl_only       = false,
2395         .ret_type       = RET_INTEGER,
2396         .arg1_type      = ARG_PTR_TO_CTX,
2397         .arg2_type      = ARG_ANYTHING,
2398         .arg3_type      = ARG_ANYTHING,
2399 };
2400
2401 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2402 {
2403         int ret;
2404
2405         bpf_push_mac_rcsum(skb);
2406         ret = skb_vlan_pop(skb);
2407         bpf_pull_mac_rcsum(skb);
2408
2409         bpf_compute_data_pointers(skb);
2410         return ret;
2411 }
2412
2413 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2414         .func           = bpf_skb_vlan_pop,
2415         .gpl_only       = false,
2416         .ret_type       = RET_INTEGER,
2417         .arg1_type      = ARG_PTR_TO_CTX,
2418 };
2419
2420 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2421 {
2422         /* Caller already did skb_cow() with len as headroom,
2423          * so no need to do it here.
2424          */
2425         skb_push(skb, len);
2426         memmove(skb->data, skb->data + len, off);
2427         memset(skb->data + off, 0, len);
2428
2429         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2430          * needed here as it does not change the skb->csum
2431          * result for checksum complete when summing over
2432          * zeroed blocks.
2433          */
2434         return 0;
2435 }
2436
2437 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2438 {
2439         /* skb_ensure_writable() is not needed here, as we're
2440          * already working on an uncloned skb.
2441          */
2442         if (unlikely(!pskb_may_pull(skb, off + len)))
2443                 return -ENOMEM;
2444
2445         skb_postpull_rcsum(skb, skb->data + off, len);
2446         memmove(skb->data + len, skb->data, off);
2447         __skb_pull(skb, len);
2448
2449         return 0;
2450 }
2451
2452 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2453 {
2454         bool trans_same = skb->transport_header == skb->network_header;
2455         int ret;
2456
2457         /* There's no need for __skb_push()/__skb_pull() pair to
2458          * get to the start of the mac header as we're guaranteed
2459          * to always start from here under eBPF.
2460          */
2461         ret = bpf_skb_generic_push(skb, off, len);
2462         if (likely(!ret)) {
2463                 skb->mac_header -= len;
2464                 skb->network_header -= len;
2465                 if (trans_same)
2466                         skb->transport_header = skb->network_header;
2467         }
2468
2469         return ret;
2470 }
2471
2472 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2473 {
2474         bool trans_same = skb->transport_header == skb->network_header;
2475         int ret;
2476
2477         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2478         ret = bpf_skb_generic_pop(skb, off, len);
2479         if (likely(!ret)) {
2480                 skb->mac_header += len;
2481                 skb->network_header += len;
2482                 if (trans_same)
2483                         skb->transport_header = skb->network_header;
2484         }
2485
2486         return ret;
2487 }
2488
2489 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2490 {
2491         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2492         u32 off = skb_mac_header_len(skb);
2493         int ret;
2494
2495         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2496         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2497                 return -ENOTSUPP;
2498
2499         ret = skb_cow(skb, len_diff);
2500         if (unlikely(ret < 0))
2501                 return ret;
2502
2503         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2504         if (unlikely(ret < 0))
2505                 return ret;
2506
2507         if (skb_is_gso(skb)) {
2508                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2509
2510                 /* SKB_GSO_TCPV4 needs to be changed into
2511                  * SKB_GSO_TCPV6.
2512                  */
2513                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2514                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2515                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2516                 }
2517
2518                 /* Due to IPv6 header, MSS needs to be downgraded. */
2519                 skb_decrease_gso_size(shinfo, len_diff);
2520                 /* Header must be checked, and gso_segs recomputed. */
2521                 shinfo->gso_type |= SKB_GSO_DODGY;
2522                 shinfo->gso_segs = 0;
2523         }
2524
2525         skb->protocol = htons(ETH_P_IPV6);
2526         skb_clear_hash(skb);
2527
2528         return 0;
2529 }
2530
2531 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2532 {
2533         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2534         u32 off = skb_mac_header_len(skb);
2535         int ret;
2536
2537         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2538         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2539                 return -ENOTSUPP;
2540
2541         ret = skb_unclone(skb, GFP_ATOMIC);
2542         if (unlikely(ret < 0))
2543                 return ret;
2544
2545         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2546         if (unlikely(ret < 0))
2547                 return ret;
2548
2549         if (skb_is_gso(skb)) {
2550                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2551
2552                 /* SKB_GSO_TCPV6 needs to be changed into
2553                  * SKB_GSO_TCPV4.
2554                  */
2555                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2556                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2557                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2558                 }
2559
2560                 /* Due to IPv4 header, MSS can be upgraded. */
2561                 skb_increase_gso_size(shinfo, len_diff);
2562                 /* Header must be checked, and gso_segs recomputed. */
2563                 shinfo->gso_type |= SKB_GSO_DODGY;
2564                 shinfo->gso_segs = 0;
2565         }
2566
2567         skb->protocol = htons(ETH_P_IP);
2568         skb_clear_hash(skb);
2569
2570         return 0;
2571 }
2572
2573 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2574 {
2575         __be16 from_proto = skb->protocol;
2576
2577         if (from_proto == htons(ETH_P_IP) &&
2578               to_proto == htons(ETH_P_IPV6))
2579                 return bpf_skb_proto_4_to_6(skb);
2580
2581         if (from_proto == htons(ETH_P_IPV6) &&
2582               to_proto == htons(ETH_P_IP))
2583                 return bpf_skb_proto_6_to_4(skb);
2584
2585         return -ENOTSUPP;
2586 }
2587
2588 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2589            u64, flags)
2590 {
2591         int ret;
2592
2593         if (unlikely(flags))
2594                 return -EINVAL;
2595
2596         /* General idea is that this helper does the basic groundwork
2597          * needed for changing the protocol, and eBPF program fills the
2598          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2599          * and other helpers, rather than passing a raw buffer here.
2600          *
2601          * The rationale is to keep this minimal and without a need to
2602          * deal with raw packet data. F.e. even if we would pass buffers
2603          * here, the program still needs to call the bpf_lX_csum_replace()
2604          * helpers anyway. Plus, this way we keep also separation of
2605          * concerns, since f.e. bpf_skb_store_bytes() should only take
2606          * care of stores.
2607          *
2608          * Currently, additional options and extension header space are
2609          * not supported, but flags register is reserved so we can adapt
2610          * that. For offloads, we mark packet as dodgy, so that headers
2611          * need to be verified first.
2612          */
2613         ret = bpf_skb_proto_xlat(skb, proto);
2614         bpf_compute_data_pointers(skb);
2615         return ret;
2616 }
2617
2618 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2619         .func           = bpf_skb_change_proto,
2620         .gpl_only       = false,
2621         .ret_type       = RET_INTEGER,
2622         .arg1_type      = ARG_PTR_TO_CTX,
2623         .arg2_type      = ARG_ANYTHING,
2624         .arg3_type      = ARG_ANYTHING,
2625 };
2626
2627 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2628 {
2629         /* We only allow a restricted subset to be changed for now. */
2630         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2631                      !skb_pkt_type_ok(pkt_type)))
2632                 return -EINVAL;
2633
2634         skb->pkt_type = pkt_type;
2635         return 0;
2636 }
2637
2638 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2639         .func           = bpf_skb_change_type,
2640         .gpl_only       = false,
2641         .ret_type       = RET_INTEGER,
2642         .arg1_type      = ARG_PTR_TO_CTX,
2643         .arg2_type      = ARG_ANYTHING,
2644 };
2645
2646 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2647 {
2648         switch (skb->protocol) {
2649         case htons(ETH_P_IP):
2650                 return sizeof(struct iphdr);
2651         case htons(ETH_P_IPV6):
2652                 return sizeof(struct ipv6hdr);
2653         default:
2654                 return ~0U;
2655         }
2656 }
2657
2658 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2659 {
2660         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2661         int ret;
2662
2663         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2664         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2665                 return -ENOTSUPP;
2666
2667         ret = skb_cow(skb, len_diff);
2668         if (unlikely(ret < 0))
2669                 return ret;
2670
2671         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2672         if (unlikely(ret < 0))
2673                 return ret;
2674
2675         if (skb_is_gso(skb)) {
2676                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2677
2678                 /* Due to header grow, MSS needs to be downgraded. */
2679                 skb_decrease_gso_size(shinfo, len_diff);
2680                 /* Header must be checked, and gso_segs recomputed. */
2681                 shinfo->gso_type |= SKB_GSO_DODGY;
2682                 shinfo->gso_segs = 0;
2683         }
2684
2685         return 0;
2686 }
2687
2688 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2689 {
2690         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2691         int ret;
2692
2693         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2694         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2695                 return -ENOTSUPP;
2696
2697         ret = skb_unclone(skb, GFP_ATOMIC);
2698         if (unlikely(ret < 0))
2699                 return ret;
2700
2701         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2702         if (unlikely(ret < 0))
2703                 return ret;
2704
2705         if (skb_is_gso(skb)) {
2706                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2707
2708                 /* Due to header shrink, MSS can be upgraded. */
2709                 skb_increase_gso_size(shinfo, len_diff);
2710                 /* Header must be checked, and gso_segs recomputed. */
2711                 shinfo->gso_type |= SKB_GSO_DODGY;
2712                 shinfo->gso_segs = 0;
2713         }
2714
2715         return 0;
2716 }
2717
2718 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2719 {
2720         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2721                           SKB_MAX_ALLOC;
2722 }
2723
2724 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2725 {
2726         bool trans_same = skb->transport_header == skb->network_header;
2727         u32 len_cur, len_diff_abs = abs(len_diff);
2728         u32 len_min = bpf_skb_net_base_len(skb);
2729         u32 len_max = __bpf_skb_max_len(skb);
2730         __be16 proto = skb->protocol;
2731         bool shrink = len_diff < 0;
2732         int ret;
2733
2734         if (unlikely(len_diff_abs > 0xfffU))
2735                 return -EFAULT;
2736         if (unlikely(proto != htons(ETH_P_IP) &&
2737                      proto != htons(ETH_P_IPV6)))
2738                 return -ENOTSUPP;
2739
2740         len_cur = skb->len - skb_network_offset(skb);
2741         if (skb_transport_header_was_set(skb) && !trans_same)
2742                 len_cur = skb_network_header_len(skb);
2743         if ((shrink && (len_diff_abs >= len_cur ||
2744                         len_cur - len_diff_abs < len_min)) ||
2745             (!shrink && (skb->len + len_diff_abs > len_max &&
2746                          !skb_is_gso(skb))))
2747                 return -ENOTSUPP;
2748
2749         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2750                        bpf_skb_net_grow(skb, len_diff_abs);
2751
2752         bpf_compute_data_pointers(skb);
2753         return ret;
2754 }
2755
2756 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2757            u32, mode, u64, flags)
2758 {
2759         if (unlikely(flags))
2760                 return -EINVAL;
2761         if (likely(mode == BPF_ADJ_ROOM_NET))
2762                 return bpf_skb_adjust_net(skb, len_diff);
2763
2764         return -ENOTSUPP;
2765 }
2766
2767 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2768         .func           = bpf_skb_adjust_room,
2769         .gpl_only       = false,
2770         .ret_type       = RET_INTEGER,
2771         .arg1_type      = ARG_PTR_TO_CTX,
2772         .arg2_type      = ARG_ANYTHING,
2773         .arg3_type      = ARG_ANYTHING,
2774         .arg4_type      = ARG_ANYTHING,
2775 };
2776
2777 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2778 {
2779         u32 min_len = skb_network_offset(skb);
2780
2781         if (skb_transport_header_was_set(skb))
2782                 min_len = skb_transport_offset(skb);
2783         if (skb->ip_summed == CHECKSUM_PARTIAL)
2784                 min_len = skb_checksum_start_offset(skb) +
2785                           skb->csum_offset + sizeof(__sum16);
2786         return min_len;
2787 }
2788
2789 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2790 {
2791         unsigned int old_len = skb->len;
2792         int ret;
2793
2794         ret = __skb_grow_rcsum(skb, new_len);
2795         if (!ret)
2796                 memset(skb->data + old_len, 0, new_len - old_len);
2797         return ret;
2798 }
2799
2800 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2801 {
2802         return __skb_trim_rcsum(skb, new_len);
2803 }
2804
2805 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2806                                         u64 flags)
2807 {
2808         u32 max_len = __bpf_skb_max_len(skb);
2809         u32 min_len = __bpf_skb_min_len(skb);
2810         int ret;
2811
2812         if (unlikely(flags || new_len > max_len || new_len < min_len))
2813                 return -EINVAL;
2814         if (skb->encapsulation)
2815                 return -ENOTSUPP;
2816
2817         /* The basic idea of this helper is that it's performing the
2818          * needed work to either grow or trim an skb, and eBPF program
2819          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2820          * bpf_lX_csum_replace() and others rather than passing a raw
2821          * buffer here. This one is a slow path helper and intended
2822          * for replies with control messages.
2823          *
2824          * Like in bpf_skb_change_proto(), we want to keep this rather
2825          * minimal and without protocol specifics so that we are able
2826          * to separate concerns as in bpf_skb_store_bytes() should only
2827          * be the one responsible for writing buffers.
2828          *
2829          * It's really expected to be a slow path operation here for
2830          * control message replies, so we're implicitly linearizing,
2831          * uncloning and drop offloads from the skb by this.
2832          */
2833         ret = __bpf_try_make_writable(skb, skb->len);
2834         if (!ret) {
2835                 if (new_len > skb->len)
2836                         ret = bpf_skb_grow_rcsum(skb, new_len);
2837                 else if (new_len < skb->len)
2838                         ret = bpf_skb_trim_rcsum(skb, new_len);
2839                 if (!ret && skb_is_gso(skb))
2840                         skb_gso_reset(skb);
2841         }
2842         return ret;
2843 }
2844
2845 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2846            u64, flags)
2847 {
2848         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2849
2850         bpf_compute_data_pointers(skb);
2851         return ret;
2852 }
2853
2854 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2855         .func           = bpf_skb_change_tail,
2856         .gpl_only       = false,
2857         .ret_type       = RET_INTEGER,
2858         .arg1_type      = ARG_PTR_TO_CTX,
2859         .arg2_type      = ARG_ANYTHING,
2860         .arg3_type      = ARG_ANYTHING,
2861 };
2862
2863 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2864            u64, flags)
2865 {
2866         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2867
2868         bpf_compute_data_end_sk_skb(skb);
2869         return ret;
2870 }
2871
2872 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2873         .func           = sk_skb_change_tail,
2874         .gpl_only       = false,
2875         .ret_type       = RET_INTEGER,
2876         .arg1_type      = ARG_PTR_TO_CTX,
2877         .arg2_type      = ARG_ANYTHING,
2878         .arg3_type      = ARG_ANYTHING,
2879 };
2880
2881 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
2882                                         u64 flags)
2883 {
2884         u32 max_len = __bpf_skb_max_len(skb);
2885         u32 new_len = skb->len + head_room;
2886         int ret;
2887
2888         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
2889                      new_len < skb->len))
2890                 return -EINVAL;
2891
2892         ret = skb_cow(skb, head_room);
2893         if (likely(!ret)) {
2894                 /* Idea for this helper is that we currently only
2895                  * allow to expand on mac header. This means that
2896                  * skb->protocol network header, etc, stay as is.
2897                  * Compared to bpf_skb_change_tail(), we're more
2898                  * flexible due to not needing to linearize or
2899                  * reset GSO. Intention for this helper is to be
2900                  * used by an L3 skb that needs to push mac header
2901                  * for redirection into L2 device.
2902                  */
2903                 __skb_push(skb, head_room);
2904                 memset(skb->data, 0, head_room);
2905                 skb_reset_mac_header(skb);
2906         }
2907
2908         return ret;
2909 }
2910
2911 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
2912            u64, flags)
2913 {
2914         int ret = __bpf_skb_change_head(skb, head_room, flags);
2915
2916         bpf_compute_data_pointers(skb);
2917         return ret;
2918 }
2919
2920 static const struct bpf_func_proto bpf_skb_change_head_proto = {
2921         .func           = bpf_skb_change_head,
2922         .gpl_only       = false,
2923         .ret_type       = RET_INTEGER,
2924         .arg1_type      = ARG_PTR_TO_CTX,
2925         .arg2_type      = ARG_ANYTHING,
2926         .arg3_type      = ARG_ANYTHING,
2927 };
2928
2929 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
2930            u64, flags)
2931 {
2932         int ret = __bpf_skb_change_head(skb, head_room, flags);
2933
2934         bpf_compute_data_end_sk_skb(skb);
2935         return ret;
2936 }
2937
2938 static const struct bpf_func_proto sk_skb_change_head_proto = {
2939         .func           = sk_skb_change_head,
2940         .gpl_only       = false,
2941         .ret_type       = RET_INTEGER,
2942         .arg1_type      = ARG_PTR_TO_CTX,
2943         .arg2_type      = ARG_ANYTHING,
2944         .arg3_type      = ARG_ANYTHING,
2945 };
2946 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
2947 {
2948         return xdp_data_meta_unsupported(xdp) ? 0 :
2949                xdp->data - xdp->data_meta;
2950 }
2951
2952 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
2953 {
2954         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
2955         unsigned long metalen = xdp_get_metalen(xdp);
2956         void *data_start = xdp_frame_end + metalen;
2957         void *data = xdp->data + offset;
2958
2959         if (unlikely(data < data_start ||
2960                      data > xdp->data_end - ETH_HLEN))
2961                 return -EINVAL;
2962
2963         if (metalen)
2964                 memmove(xdp->data_meta + offset,
2965                         xdp->data_meta, metalen);
2966         xdp->data_meta += offset;
2967         xdp->data = data;
2968
2969         return 0;
2970 }
2971
2972 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
2973         .func           = bpf_xdp_adjust_head,
2974         .gpl_only       = false,
2975         .ret_type       = RET_INTEGER,
2976         .arg1_type      = ARG_PTR_TO_CTX,
2977         .arg2_type      = ARG_ANYTHING,
2978 };
2979
2980 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
2981 {
2982         void *data_end = xdp->data_end + offset;
2983
2984         /* only shrinking is allowed for now. */
2985         if (unlikely(offset >= 0))
2986                 return -EINVAL;
2987
2988         if (unlikely(data_end < xdp->data + ETH_HLEN))
2989                 return -EINVAL;
2990
2991         xdp->data_end = data_end;
2992
2993         return 0;
2994 }
2995
2996 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
2997         .func           = bpf_xdp_adjust_tail,
2998         .gpl_only       = false,
2999         .ret_type       = RET_INTEGER,
3000         .arg1_type      = ARG_PTR_TO_CTX,
3001         .arg2_type      = ARG_ANYTHING,
3002 };
3003
3004 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3005 {
3006         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3007         void *meta = xdp->data_meta + offset;
3008         unsigned long metalen = xdp->data - meta;
3009
3010         if (xdp_data_meta_unsupported(xdp))
3011                 return -ENOTSUPP;
3012         if (unlikely(meta < xdp_frame_end ||
3013                      meta > xdp->data))
3014                 return -EINVAL;
3015         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3016                      (metalen > 32)))
3017                 return -EACCES;
3018
3019         xdp->data_meta = meta;
3020
3021         return 0;
3022 }
3023
3024 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3025         .func           = bpf_xdp_adjust_meta,
3026         .gpl_only       = false,
3027         .ret_type       = RET_INTEGER,
3028         .arg1_type      = ARG_PTR_TO_CTX,
3029         .arg2_type      = ARG_ANYTHING,
3030 };
3031
3032 static int __bpf_tx_xdp(struct net_device *dev,
3033                         struct bpf_map *map,
3034                         struct xdp_buff *xdp,
3035                         u32 index)
3036 {
3037         struct xdp_frame *xdpf;
3038         int err, sent;
3039
3040         if (!dev->netdev_ops->ndo_xdp_xmit) {
3041                 return -EOPNOTSUPP;
3042         }
3043
3044         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3045         if (unlikely(err))
3046                 return err;
3047
3048         xdpf = convert_to_xdp_frame(xdp);
3049         if (unlikely(!xdpf))
3050                 return -EOVERFLOW;
3051
3052         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3053         if (sent <= 0)
3054                 return sent;
3055         return 0;
3056 }
3057
3058 static noinline int
3059 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3060                      struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3061 {
3062         struct net_device *fwd;
3063         u32 index = ri->ifindex;
3064         int err;
3065
3066         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3067         ri->ifindex = 0;
3068         if (unlikely(!fwd)) {
3069                 err = -EINVAL;
3070                 goto err;
3071         }
3072
3073         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3074         if (unlikely(err))
3075                 goto err;
3076
3077         _trace_xdp_redirect(dev, xdp_prog, index);
3078         return 0;
3079 err:
3080         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3081         return err;
3082 }
3083
3084 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3085                             struct bpf_map *map,
3086                             struct xdp_buff *xdp,
3087                             u32 index)
3088 {
3089         int err;
3090
3091         switch (map->map_type) {
3092         case BPF_MAP_TYPE_DEVMAP: {
3093                 struct bpf_dtab_netdev *dst = fwd;
3094
3095                 err = dev_map_enqueue(dst, xdp, dev_rx);
3096                 if (unlikely(err))
3097                         return err;
3098                 __dev_map_insert_ctx(map, index);
3099                 break;
3100         }
3101         case BPF_MAP_TYPE_CPUMAP: {
3102                 struct bpf_cpu_map_entry *rcpu = fwd;
3103
3104                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3105                 if (unlikely(err))
3106                         return err;
3107                 __cpu_map_insert_ctx(map, index);
3108                 break;
3109         }
3110         case BPF_MAP_TYPE_XSKMAP: {
3111                 struct xdp_sock *xs = fwd;
3112
3113                 err = __xsk_map_redirect(map, xdp, xs);
3114                 return err;
3115         }
3116         default:
3117                 break;
3118         }
3119         return 0;
3120 }
3121
3122 void xdp_do_flush_map(void)
3123 {
3124         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3125         struct bpf_map *map = ri->map_to_flush;
3126
3127         ri->map_to_flush = NULL;
3128         if (map) {
3129                 switch (map->map_type) {
3130                 case BPF_MAP_TYPE_DEVMAP:
3131                         __dev_map_flush(map);
3132                         break;
3133                 case BPF_MAP_TYPE_CPUMAP:
3134                         __cpu_map_flush(map);
3135                         break;
3136                 case BPF_MAP_TYPE_XSKMAP:
3137                         __xsk_map_flush(map);
3138                         break;
3139                 default:
3140                         break;
3141                 }
3142         }
3143 }
3144 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3145
3146 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3147 {
3148         switch (map->map_type) {
3149         case BPF_MAP_TYPE_DEVMAP:
3150                 return __dev_map_lookup_elem(map, index);
3151         case BPF_MAP_TYPE_CPUMAP:
3152                 return __cpu_map_lookup_elem(map, index);
3153         case BPF_MAP_TYPE_XSKMAP:
3154                 return __xsk_map_lookup_elem(map, index);
3155         default:
3156                 return NULL;
3157         }
3158 }
3159
3160 void bpf_clear_redirect_map(struct bpf_map *map)
3161 {
3162         struct bpf_redirect_info *ri;
3163         int cpu;
3164
3165         for_each_possible_cpu(cpu) {
3166                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3167                 /* Avoid polluting remote cacheline due to writes if
3168                  * not needed. Once we pass this test, we need the
3169                  * cmpxchg() to make sure it hasn't been changed in
3170                  * the meantime by remote CPU.
3171                  */
3172                 if (unlikely(READ_ONCE(ri->map) == map))
3173                         cmpxchg(&ri->map, map, NULL);
3174         }
3175 }
3176
3177 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3178                                struct bpf_prog *xdp_prog, struct bpf_map *map,
3179                                struct bpf_redirect_info *ri)
3180 {
3181         u32 index = ri->ifindex;
3182         void *fwd = NULL;
3183         int err;
3184
3185         ri->ifindex = 0;
3186         WRITE_ONCE(ri->map, NULL);
3187
3188         fwd = __xdp_map_lookup_elem(map, index);
3189         if (unlikely(!fwd)) {
3190                 err = -EINVAL;
3191                 goto err;
3192         }
3193         if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3194                 xdp_do_flush_map();
3195
3196         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3197         if (unlikely(err))
3198                 goto err;
3199
3200         ri->map_to_flush = map;
3201         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3202         return 0;
3203 err:
3204         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3205         return err;
3206 }
3207
3208 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3209                     struct bpf_prog *xdp_prog)
3210 {
3211         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3212         struct bpf_map *map = READ_ONCE(ri->map);
3213
3214         if (likely(map))
3215                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3216
3217         return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3218 }
3219 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3220
3221 static int xdp_do_generic_redirect_map(struct net_device *dev,
3222                                        struct sk_buff *skb,
3223                                        struct xdp_buff *xdp,
3224                                        struct bpf_prog *xdp_prog,
3225                                        struct bpf_map *map)
3226 {
3227         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3228         u32 index = ri->ifindex;
3229         void *fwd = NULL;
3230         int err = 0;
3231
3232         ri->ifindex = 0;
3233         WRITE_ONCE(ri->map, NULL);
3234
3235         fwd = __xdp_map_lookup_elem(map, index);
3236         if (unlikely(!fwd)) {
3237                 err = -EINVAL;
3238                 goto err;
3239         }
3240
3241         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3242                 struct bpf_dtab_netdev *dst = fwd;
3243
3244                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3245                 if (unlikely(err))
3246                         goto err;
3247         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3248                 struct xdp_sock *xs = fwd;
3249
3250                 err = xsk_generic_rcv(xs, xdp);
3251                 if (err)
3252                         goto err;
3253                 consume_skb(skb);
3254         } else {
3255                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3256                 err = -EBADRQC;
3257                 goto err;
3258         }
3259
3260         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3261         return 0;
3262 err:
3263         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3264         return err;
3265 }
3266
3267 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3268                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3269 {
3270         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3271         struct bpf_map *map = READ_ONCE(ri->map);
3272         u32 index = ri->ifindex;
3273         struct net_device *fwd;
3274         int err = 0;
3275
3276         if (map)
3277                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3278                                                    map);
3279         ri->ifindex = 0;
3280         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3281         if (unlikely(!fwd)) {
3282                 err = -EINVAL;
3283                 goto err;
3284         }
3285
3286         err = xdp_ok_fwd_dev(fwd, skb->len);
3287         if (unlikely(err))
3288                 goto err;
3289
3290         skb->dev = fwd;
3291         _trace_xdp_redirect(dev, xdp_prog, index);
3292         generic_xdp_tx(skb, xdp_prog);
3293         return 0;
3294 err:
3295         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3296         return err;
3297 }
3298 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3299
3300 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3301 {
3302         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3303
3304         if (unlikely(flags))
3305                 return XDP_ABORTED;
3306
3307         ri->ifindex = ifindex;
3308         ri->flags = flags;
3309         WRITE_ONCE(ri->map, NULL);
3310
3311         return XDP_REDIRECT;
3312 }
3313
3314 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3315         .func           = bpf_xdp_redirect,
3316         .gpl_only       = false,
3317         .ret_type       = RET_INTEGER,
3318         .arg1_type      = ARG_ANYTHING,
3319         .arg2_type      = ARG_ANYTHING,
3320 };
3321
3322 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3323            u64, flags)
3324 {
3325         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3326
3327         if (unlikely(flags))
3328                 return XDP_ABORTED;
3329
3330         ri->ifindex = ifindex;
3331         ri->flags = flags;
3332         WRITE_ONCE(ri->map, map);
3333
3334         return XDP_REDIRECT;
3335 }
3336
3337 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3338         .func           = bpf_xdp_redirect_map,
3339         .gpl_only       = false,
3340         .ret_type       = RET_INTEGER,
3341         .arg1_type      = ARG_CONST_MAP_PTR,
3342         .arg2_type      = ARG_ANYTHING,
3343         .arg3_type      = ARG_ANYTHING,
3344 };
3345
3346 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3347                                   unsigned long off, unsigned long len)
3348 {
3349         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3350
3351         if (unlikely(!ptr))
3352                 return len;
3353         if (ptr != dst_buff)
3354                 memcpy(dst_buff, ptr, len);
3355
3356         return 0;
3357 }
3358
3359 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3360            u64, flags, void *, meta, u64, meta_size)
3361 {
3362         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3363
3364         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3365                 return -EINVAL;
3366         if (unlikely(skb_size > skb->len))
3367                 return -EFAULT;
3368
3369         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3370                                 bpf_skb_copy);
3371 }
3372
3373 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3374         .func           = bpf_skb_event_output,
3375         .gpl_only       = true,
3376         .ret_type       = RET_INTEGER,
3377         .arg1_type      = ARG_PTR_TO_CTX,
3378         .arg2_type      = ARG_CONST_MAP_PTR,
3379         .arg3_type      = ARG_ANYTHING,
3380         .arg4_type      = ARG_PTR_TO_MEM,
3381         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3382 };
3383
3384 static unsigned short bpf_tunnel_key_af(u64 flags)
3385 {
3386         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3387 }
3388
3389 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3390            u32, size, u64, flags)
3391 {
3392         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3393         u8 compat[sizeof(struct bpf_tunnel_key)];
3394         void *to_orig = to;
3395         int err;
3396
3397         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3398                 err = -EINVAL;
3399                 goto err_clear;
3400         }
3401         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3402                 err = -EPROTO;
3403                 goto err_clear;
3404         }
3405         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3406                 err = -EINVAL;
3407                 switch (size) {
3408                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3409                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3410                         goto set_compat;
3411                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3412                         /* Fixup deprecated structure layouts here, so we have
3413                          * a common path later on.
3414                          */
3415                         if (ip_tunnel_info_af(info) != AF_INET)
3416                                 goto err_clear;
3417 set_compat:
3418                         to = (struct bpf_tunnel_key *)compat;
3419                         break;
3420                 default:
3421                         goto err_clear;
3422                 }
3423         }
3424
3425         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3426         to->tunnel_tos = info->key.tos;
3427         to->tunnel_ttl = info->key.ttl;
3428         to->tunnel_ext = 0;
3429
3430         if (flags & BPF_F_TUNINFO_IPV6) {
3431                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3432                        sizeof(to->remote_ipv6));
3433                 to->tunnel_label = be32_to_cpu(info->key.label);
3434         } else {
3435                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3436                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3437                 to->tunnel_label = 0;
3438         }
3439
3440         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3441                 memcpy(to_orig, to, size);
3442
3443         return 0;
3444 err_clear:
3445         memset(to_orig, 0, size);
3446         return err;
3447 }
3448
3449 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3450         .func           = bpf_skb_get_tunnel_key,
3451         .gpl_only       = false,
3452         .ret_type       = RET_INTEGER,
3453         .arg1_type      = ARG_PTR_TO_CTX,
3454         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3455         .arg3_type      = ARG_CONST_SIZE,
3456         .arg4_type      = ARG_ANYTHING,
3457 };
3458
3459 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3460 {
3461         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3462         int err;
3463
3464         if (unlikely(!info ||
3465                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3466                 err = -ENOENT;
3467                 goto err_clear;
3468         }
3469         if (unlikely(size < info->options_len)) {
3470                 err = -ENOMEM;
3471                 goto err_clear;
3472         }
3473
3474         ip_tunnel_info_opts_get(to, info);
3475         if (size > info->options_len)
3476                 memset(to + info->options_len, 0, size - info->options_len);
3477
3478         return info->options_len;
3479 err_clear:
3480         memset(to, 0, size);
3481         return err;
3482 }
3483
3484 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3485         .func           = bpf_skb_get_tunnel_opt,
3486         .gpl_only       = false,
3487         .ret_type       = RET_INTEGER,
3488         .arg1_type      = ARG_PTR_TO_CTX,
3489         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3490         .arg3_type      = ARG_CONST_SIZE,
3491 };
3492
3493 static struct metadata_dst __percpu *md_dst;
3494
3495 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3496            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3497 {
3498         struct metadata_dst *md = this_cpu_ptr(md_dst);
3499         u8 compat[sizeof(struct bpf_tunnel_key)];
3500         struct ip_tunnel_info *info;
3501
3502         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3503                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3504                 return -EINVAL;
3505         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3506                 switch (size) {
3507                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3508                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3509                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3510                         /* Fixup deprecated structure layouts here, so we have
3511                          * a common path later on.
3512                          */
3513                         memcpy(compat, from, size);
3514                         memset(compat + size, 0, sizeof(compat) - size);
3515                         from = (const struct bpf_tunnel_key *) compat;
3516                         break;
3517                 default:
3518                         return -EINVAL;
3519                 }
3520         }
3521         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3522                      from->tunnel_ext))
3523                 return -EINVAL;
3524
3525         skb_dst_drop(skb);
3526         dst_hold((struct dst_entry *) md);
3527         skb_dst_set(skb, (struct dst_entry *) md);
3528
3529         info = &md->u.tun_info;
3530         memset(info, 0, sizeof(*info));
3531         info->mode = IP_TUNNEL_INFO_TX;
3532
3533         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3534         if (flags & BPF_F_DONT_FRAGMENT)
3535                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3536         if (flags & BPF_F_ZERO_CSUM_TX)
3537                 info->key.tun_flags &= ~TUNNEL_CSUM;
3538         if (flags & BPF_F_SEQ_NUMBER)
3539                 info->key.tun_flags |= TUNNEL_SEQ;
3540
3541         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3542         info->key.tos = from->tunnel_tos;
3543         info->key.ttl = from->tunnel_ttl;
3544
3545         if (flags & BPF_F_TUNINFO_IPV6) {
3546                 info->mode |= IP_TUNNEL_INFO_IPV6;
3547                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3548                        sizeof(from->remote_ipv6));
3549                 info->key.label = cpu_to_be32(from->tunnel_label) &
3550                                   IPV6_FLOWLABEL_MASK;
3551         } else {
3552                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3553         }
3554
3555         return 0;
3556 }
3557
3558 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3559         .func           = bpf_skb_set_tunnel_key,
3560         .gpl_only       = false,
3561         .ret_type       = RET_INTEGER,
3562         .arg1_type      = ARG_PTR_TO_CTX,
3563         .arg2_type      = ARG_PTR_TO_MEM,
3564         .arg3_type      = ARG_CONST_SIZE,
3565         .arg4_type      = ARG_ANYTHING,
3566 };
3567
3568 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3569            const u8 *, from, u32, size)
3570 {
3571         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3572         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3573
3574         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3575                 return -EINVAL;
3576         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3577                 return -ENOMEM;
3578
3579         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3580
3581         return 0;
3582 }
3583
3584 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3585         .func           = bpf_skb_set_tunnel_opt,
3586         .gpl_only       = false,
3587         .ret_type       = RET_INTEGER,
3588         .arg1_type      = ARG_PTR_TO_CTX,
3589         .arg2_type      = ARG_PTR_TO_MEM,
3590         .arg3_type      = ARG_CONST_SIZE,
3591 };
3592
3593 static const struct bpf_func_proto *
3594 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3595 {
3596         if (!md_dst) {
3597                 struct metadata_dst __percpu *tmp;
3598
3599                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3600                                                 METADATA_IP_TUNNEL,
3601                                                 GFP_KERNEL);
3602                 if (!tmp)
3603                         return NULL;
3604                 if (cmpxchg(&md_dst, NULL, tmp))
3605                         metadata_dst_free_percpu(tmp);
3606         }
3607
3608         switch (which) {
3609         case BPF_FUNC_skb_set_tunnel_key:
3610                 return &bpf_skb_set_tunnel_key_proto;
3611         case BPF_FUNC_skb_set_tunnel_opt:
3612                 return &bpf_skb_set_tunnel_opt_proto;
3613         default:
3614                 return NULL;
3615         }
3616 }
3617
3618 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3619            u32, idx)
3620 {
3621         struct bpf_array *array = container_of(map, struct bpf_array, map);
3622         struct cgroup *cgrp;
3623         struct sock *sk;
3624
3625         sk = skb_to_full_sk(skb);
3626         if (!sk || !sk_fullsock(sk))
3627                 return -ENOENT;
3628         if (unlikely(idx >= array->map.max_entries))
3629                 return -E2BIG;
3630
3631         cgrp = READ_ONCE(array->ptrs[idx]);
3632         if (unlikely(!cgrp))
3633                 return -EAGAIN;
3634
3635         return sk_under_cgroup_hierarchy(sk, cgrp);
3636 }
3637
3638 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3639         .func           = bpf_skb_under_cgroup,
3640         .gpl_only       = false,
3641         .ret_type       = RET_INTEGER,
3642         .arg1_type      = ARG_PTR_TO_CTX,
3643         .arg2_type      = ARG_CONST_MAP_PTR,
3644         .arg3_type      = ARG_ANYTHING,
3645 };
3646
3647 #ifdef CONFIG_SOCK_CGROUP_DATA
3648 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3649 {
3650         struct sock *sk = skb_to_full_sk(skb);
3651         struct cgroup *cgrp;
3652
3653         if (!sk || !sk_fullsock(sk))
3654                 return 0;
3655
3656         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3657         return cgrp->kn->id.id;
3658 }
3659
3660 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3661         .func           = bpf_skb_cgroup_id,
3662         .gpl_only       = false,
3663         .ret_type       = RET_INTEGER,
3664         .arg1_type      = ARG_PTR_TO_CTX,
3665 };
3666
3667 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3668            ancestor_level)
3669 {
3670         struct sock *sk = skb_to_full_sk(skb);
3671         struct cgroup *ancestor;
3672         struct cgroup *cgrp;
3673
3674         if (!sk || !sk_fullsock(sk))
3675                 return 0;
3676
3677         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3678         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3679         if (!ancestor)
3680                 return 0;
3681
3682         return ancestor->kn->id.id;
3683 }
3684
3685 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3686         .func           = bpf_skb_ancestor_cgroup_id,
3687         .gpl_only       = false,
3688         .ret_type       = RET_INTEGER,
3689         .arg1_type      = ARG_PTR_TO_CTX,
3690         .arg2_type      = ARG_ANYTHING,
3691 };
3692 #endif
3693
3694 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3695                                   unsigned long off, unsigned long len)
3696 {
3697         memcpy(dst_buff, src_buff + off, len);
3698         return 0;
3699 }
3700
3701 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3702            u64, flags, void *, meta, u64, meta_size)
3703 {
3704         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3705
3706         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3707                 return -EINVAL;
3708         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3709                 return -EFAULT;
3710
3711         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3712                                 xdp_size, bpf_xdp_copy);
3713 }
3714
3715 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3716         .func           = bpf_xdp_event_output,
3717         .gpl_only       = true,
3718         .ret_type       = RET_INTEGER,
3719         .arg1_type      = ARG_PTR_TO_CTX,
3720         .arg2_type      = ARG_CONST_MAP_PTR,
3721         .arg3_type      = ARG_ANYTHING,
3722         .arg4_type      = ARG_PTR_TO_MEM,
3723         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3724 };
3725
3726 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3727 {
3728         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3729 }
3730
3731 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3732         .func           = bpf_get_socket_cookie,
3733         .gpl_only       = false,
3734         .ret_type       = RET_INTEGER,
3735         .arg1_type      = ARG_PTR_TO_CTX,
3736 };
3737
3738 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
3739 {
3740         return sock_gen_cookie(ctx->sk);
3741 }
3742
3743 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
3744         .func           = bpf_get_socket_cookie_sock_addr,
3745         .gpl_only       = false,
3746         .ret_type       = RET_INTEGER,
3747         .arg1_type      = ARG_PTR_TO_CTX,
3748 };
3749
3750 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
3751 {
3752         return sock_gen_cookie(ctx->sk);
3753 }
3754
3755 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
3756         .func           = bpf_get_socket_cookie_sock_ops,
3757         .gpl_only       = false,
3758         .ret_type       = RET_INTEGER,
3759         .arg1_type      = ARG_PTR_TO_CTX,
3760 };
3761
3762 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3763 {
3764         struct sock *sk = sk_to_full_sk(skb->sk);
3765         kuid_t kuid;
3766
3767         if (!sk || !sk_fullsock(sk))
3768                 return overflowuid;
3769         kuid = sock_net_uid(sock_net(sk), sk);
3770         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3771 }
3772
3773 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3774         .func           = bpf_get_socket_uid,
3775         .gpl_only       = false,
3776         .ret_type       = RET_INTEGER,
3777         .arg1_type      = ARG_PTR_TO_CTX,
3778 };
3779
3780 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3781            int, level, int, optname, char *, optval, int, optlen)
3782 {
3783         struct sock *sk = bpf_sock->sk;
3784         int ret = 0;
3785         int val;
3786
3787         if (!sk_fullsock(sk))
3788                 return -EINVAL;
3789
3790         if (level == SOL_SOCKET) {
3791                 if (optlen != sizeof(int))
3792                         return -EINVAL;
3793                 val = *((int *)optval);
3794
3795                 /* Only some socketops are supported */
3796                 switch (optname) {
3797                 case SO_RCVBUF:
3798                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3799                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3800                         break;
3801                 case SO_SNDBUF:
3802                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3803                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3804                         break;
3805                 case SO_MAX_PACING_RATE:
3806                         sk->sk_max_pacing_rate = val;
3807                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3808                                                  sk->sk_max_pacing_rate);
3809                         break;
3810                 case SO_PRIORITY:
3811                         sk->sk_priority = val;
3812                         break;
3813                 case SO_RCVLOWAT:
3814                         if (val < 0)
3815                                 val = INT_MAX;
3816                         sk->sk_rcvlowat = val ? : 1;
3817                         break;
3818                 case SO_MARK:
3819                         sk->sk_mark = val;
3820                         break;
3821                 default:
3822                         ret = -EINVAL;
3823                 }
3824 #ifdef CONFIG_INET
3825         } else if (level == SOL_IP) {
3826                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3827                         return -EINVAL;
3828
3829                 val = *((int *)optval);
3830                 /* Only some options are supported */
3831                 switch (optname) {
3832                 case IP_TOS:
3833                         if (val < -1 || val > 0xff) {
3834                                 ret = -EINVAL;
3835                         } else {
3836                                 struct inet_sock *inet = inet_sk(sk);
3837
3838                                 if (val == -1)
3839                                         val = 0;
3840                                 inet->tos = val;
3841                         }
3842                         break;
3843                 default:
3844                         ret = -EINVAL;
3845                 }
3846 #if IS_ENABLED(CONFIG_IPV6)
3847         } else if (level == SOL_IPV6) {
3848                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3849                         return -EINVAL;
3850
3851                 val = *((int *)optval);
3852                 /* Only some options are supported */
3853                 switch (optname) {
3854                 case IPV6_TCLASS:
3855                         if (val < -1 || val > 0xff) {
3856                                 ret = -EINVAL;
3857                         } else {
3858                                 struct ipv6_pinfo *np = inet6_sk(sk);
3859
3860                                 if (val == -1)
3861                                         val = 0;
3862                                 np->tclass = val;
3863                         }
3864                         break;
3865                 default:
3866                         ret = -EINVAL;
3867                 }
3868 #endif
3869         } else if (level == SOL_TCP &&
3870                    sk->sk_prot->setsockopt == tcp_setsockopt) {
3871                 if (optname == TCP_CONGESTION) {
3872                         char name[TCP_CA_NAME_MAX];
3873                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3874
3875                         strncpy(name, optval, min_t(long, optlen,
3876                                                     TCP_CA_NAME_MAX-1));
3877                         name[TCP_CA_NAME_MAX-1] = 0;
3878                         ret = tcp_set_congestion_control(sk, name, false,
3879                                                          reinit);
3880                 } else {
3881                         struct tcp_sock *tp = tcp_sk(sk);
3882
3883                         if (optlen != sizeof(int))
3884                                 return -EINVAL;
3885
3886                         val = *((int *)optval);
3887                         /* Only some options are supported */
3888                         switch (optname) {
3889                         case TCP_BPF_IW:
3890                                 if (val <= 0 || tp->data_segs_out > 0)
3891                                         ret = -EINVAL;
3892                                 else
3893                                         tp->snd_cwnd = val;
3894                                 break;
3895                         case TCP_BPF_SNDCWND_CLAMP:
3896                                 if (val <= 0) {
3897                                         ret = -EINVAL;
3898                                 } else {
3899                                         tp->snd_cwnd_clamp = val;
3900                                         tp->snd_ssthresh = val;
3901                                 }
3902                                 break;
3903                         case TCP_SAVE_SYN:
3904                                 if (val < 0 || val > 1)
3905                                         ret = -EINVAL;
3906                                 else
3907                                         tp->save_syn = val;
3908                                 break;
3909                         default:
3910                                 ret = -EINVAL;
3911                         }
3912                 }
3913 #endif
3914         } else {
3915                 ret = -EINVAL;
3916         }
3917         return ret;
3918 }
3919
3920 static const struct bpf_func_proto bpf_setsockopt_proto = {
3921         .func           = bpf_setsockopt,
3922         .gpl_only       = false,
3923         .ret_type       = RET_INTEGER,
3924         .arg1_type      = ARG_PTR_TO_CTX,
3925         .arg2_type      = ARG_ANYTHING,
3926         .arg3_type      = ARG_ANYTHING,
3927         .arg4_type      = ARG_PTR_TO_MEM,
3928         .arg5_type      = ARG_CONST_SIZE,
3929 };
3930
3931 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3932            int, level, int, optname, char *, optval, int, optlen)
3933 {
3934         struct sock *sk = bpf_sock->sk;
3935
3936         if (!sk_fullsock(sk))
3937                 goto err_clear;
3938 #ifdef CONFIG_INET
3939         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
3940                 struct inet_connection_sock *icsk;
3941                 struct tcp_sock *tp;
3942
3943                 switch (optname) {
3944                 case TCP_CONGESTION:
3945                         icsk = inet_csk(sk);
3946
3947                         if (!icsk->icsk_ca_ops || optlen <= 1)
3948                                 goto err_clear;
3949                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
3950                         optval[optlen - 1] = 0;
3951                         break;
3952                 case TCP_SAVED_SYN:
3953                         tp = tcp_sk(sk);
3954
3955                         if (optlen <= 0 || !tp->saved_syn ||
3956                             optlen > tp->saved_syn[0])
3957                                 goto err_clear;
3958                         memcpy(optval, tp->saved_syn + 1, optlen);
3959                         break;
3960                 default:
3961                         goto err_clear;
3962                 }
3963         } else if (level == SOL_IP) {
3964                 struct inet_sock *inet = inet_sk(sk);
3965
3966                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3967                         goto err_clear;
3968
3969                 /* Only some options are supported */
3970                 switch (optname) {
3971                 case IP_TOS:
3972                         *((int *)optval) = (int)inet->tos;
3973                         break;
3974                 default:
3975                         goto err_clear;
3976                 }
3977 #if IS_ENABLED(CONFIG_IPV6)
3978         } else if (level == SOL_IPV6) {
3979                 struct ipv6_pinfo *np = inet6_sk(sk);
3980
3981                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3982                         goto err_clear;
3983
3984                 /* Only some options are supported */
3985                 switch (optname) {
3986                 case IPV6_TCLASS:
3987                         *((int *)optval) = (int)np->tclass;
3988                         break;
3989                 default:
3990                         goto err_clear;
3991                 }
3992 #endif
3993         } else {
3994                 goto err_clear;
3995         }
3996         return 0;
3997 #endif
3998 err_clear:
3999         memset(optval, 0, optlen);
4000         return -EINVAL;
4001 }
4002
4003 static const struct bpf_func_proto bpf_getsockopt_proto = {
4004         .func           = bpf_getsockopt,
4005         .gpl_only       = false,
4006         .ret_type       = RET_INTEGER,
4007         .arg1_type      = ARG_PTR_TO_CTX,
4008         .arg2_type      = ARG_ANYTHING,
4009         .arg3_type      = ARG_ANYTHING,
4010         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4011         .arg5_type      = ARG_CONST_SIZE,
4012 };
4013
4014 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4015            int, argval)
4016 {
4017         struct sock *sk = bpf_sock->sk;
4018         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4019
4020         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4021                 return -EINVAL;
4022
4023         if (val)
4024                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4025
4026         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4027 }
4028
4029 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4030         .func           = bpf_sock_ops_cb_flags_set,
4031         .gpl_only       = false,
4032         .ret_type       = RET_INTEGER,
4033         .arg1_type      = ARG_PTR_TO_CTX,
4034         .arg2_type      = ARG_ANYTHING,
4035 };
4036
4037 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4038 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4039
4040 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4041            int, addr_len)
4042 {
4043 #ifdef CONFIG_INET
4044         struct sock *sk = ctx->sk;
4045         int err;
4046
4047         /* Binding to port can be expensive so it's prohibited in the helper.
4048          * Only binding to IP is supported.
4049          */
4050         err = -EINVAL;
4051         if (addr->sa_family == AF_INET) {
4052                 if (addr_len < sizeof(struct sockaddr_in))
4053                         return err;
4054                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4055                         return err;
4056                 return __inet_bind(sk, addr, addr_len, true, false);
4057 #if IS_ENABLED(CONFIG_IPV6)
4058         } else if (addr->sa_family == AF_INET6) {
4059                 if (addr_len < SIN6_LEN_RFC2133)
4060                         return err;
4061                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4062                         return err;
4063                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4064                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4065                  */
4066                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4067 #endif /* CONFIG_IPV6 */
4068         }
4069 #endif /* CONFIG_INET */
4070
4071         return -EAFNOSUPPORT;
4072 }
4073
4074 static const struct bpf_func_proto bpf_bind_proto = {
4075         .func           = bpf_bind,
4076         .gpl_only       = false,
4077         .ret_type       = RET_INTEGER,
4078         .arg1_type      = ARG_PTR_TO_CTX,
4079         .arg2_type      = ARG_PTR_TO_MEM,
4080         .arg3_type      = ARG_CONST_SIZE,
4081 };
4082
4083 #ifdef CONFIG_XFRM
4084 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4085            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4086 {
4087         const struct sec_path *sp = skb_sec_path(skb);
4088         const struct xfrm_state *x;
4089
4090         if (!sp || unlikely(index >= sp->len || flags))
4091                 goto err_clear;
4092
4093         x = sp->xvec[index];
4094
4095         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4096                 goto err_clear;
4097
4098         to->reqid = x->props.reqid;
4099         to->spi = x->id.spi;
4100         to->family = x->props.family;
4101         to->ext = 0;
4102
4103         if (to->family == AF_INET6) {
4104                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4105                        sizeof(to->remote_ipv6));
4106         } else {
4107                 to->remote_ipv4 = x->props.saddr.a4;
4108                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4109         }
4110
4111         return 0;
4112 err_clear:
4113         memset(to, 0, size);
4114         return -EINVAL;
4115 }
4116
4117 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4118         .func           = bpf_skb_get_xfrm_state,
4119         .gpl_only       = false,
4120         .ret_type       = RET_INTEGER,
4121         .arg1_type      = ARG_PTR_TO_CTX,
4122         .arg2_type      = ARG_ANYTHING,
4123         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4124         .arg4_type      = ARG_CONST_SIZE,
4125         .arg5_type      = ARG_ANYTHING,
4126 };
4127 #endif
4128
4129 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4130 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4131                                   const struct neighbour *neigh,
4132                                   const struct net_device *dev)
4133 {
4134         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4135         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4136         params->h_vlan_TCI = 0;
4137         params->h_vlan_proto = 0;
4138         params->ifindex = dev->ifindex;
4139
4140         return 0;
4141 }
4142 #endif
4143
4144 #if IS_ENABLED(CONFIG_INET)
4145 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4146                                u32 flags, bool check_mtu)
4147 {
4148         struct in_device *in_dev;
4149         struct neighbour *neigh;
4150         struct net_device *dev;
4151         struct fib_result res;
4152         struct fib_nh *nh;
4153         struct flowi4 fl4;
4154         int err;
4155         u32 mtu;
4156
4157         dev = dev_get_by_index_rcu(net, params->ifindex);
4158         if (unlikely(!dev))
4159                 return -ENODEV;
4160
4161         /* verify forwarding is enabled on this interface */
4162         in_dev = __in_dev_get_rcu(dev);
4163         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4164                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4165
4166         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4167                 fl4.flowi4_iif = 1;
4168                 fl4.flowi4_oif = params->ifindex;
4169         } else {
4170                 fl4.flowi4_iif = params->ifindex;
4171                 fl4.flowi4_oif = 0;
4172         }
4173         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4174         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4175         fl4.flowi4_flags = 0;
4176
4177         fl4.flowi4_proto = params->l4_protocol;
4178         fl4.daddr = params->ipv4_dst;
4179         fl4.saddr = params->ipv4_src;
4180         fl4.fl4_sport = params->sport;
4181         fl4.fl4_dport = params->dport;
4182
4183         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4184                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4185                 struct fib_table *tb;
4186
4187                 tb = fib_get_table(net, tbid);
4188                 if (unlikely(!tb))
4189                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4190
4191                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4192         } else {
4193                 fl4.flowi4_mark = 0;
4194                 fl4.flowi4_secid = 0;
4195                 fl4.flowi4_tun_key.tun_id = 0;
4196                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4197
4198                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4199         }
4200
4201         if (err) {
4202                 /* map fib lookup errors to RTN_ type */
4203                 if (err == -EINVAL)
4204                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4205                 if (err == -EHOSTUNREACH)
4206                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4207                 if (err == -EACCES)
4208                         return BPF_FIB_LKUP_RET_PROHIBIT;
4209
4210                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4211         }
4212
4213         if (res.type != RTN_UNICAST)
4214                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4215
4216         if (res.fi->fib_nhs > 1)
4217                 fib_select_path(net, &res, &fl4, NULL);
4218
4219         if (check_mtu) {
4220                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4221                 if (params->tot_len > mtu)
4222                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4223         }
4224
4225         nh = &res.fi->fib_nh[res.nh_sel];
4226
4227         /* do not handle lwt encaps right now */
4228         if (nh->nh_lwtstate)
4229                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4230
4231         dev = nh->nh_dev;
4232         if (nh->nh_gw)
4233                 params->ipv4_dst = nh->nh_gw;
4234
4235         params->rt_metric = res.fi->fib_priority;
4236
4237         /* xdp and cls_bpf programs are run in RCU-bh so
4238          * rcu_read_lock_bh is not needed here
4239          */
4240         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4241         if (!neigh)
4242                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4243
4244         return bpf_fib_set_fwd_params(params, neigh, dev);
4245 }
4246 #endif
4247
4248 #if IS_ENABLED(CONFIG_IPV6)
4249 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4250                                u32 flags, bool check_mtu)
4251 {
4252         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4253         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4254         struct neighbour *neigh;
4255         struct net_device *dev;
4256         struct inet6_dev *idev;
4257         struct fib6_info *f6i;
4258         struct flowi6 fl6;
4259         int strict = 0;
4260         int oif;
4261         u32 mtu;
4262
4263         /* link local addresses are never forwarded */
4264         if (rt6_need_strict(dst) || rt6_need_strict(src))
4265                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4266
4267         dev = dev_get_by_index_rcu(net, params->ifindex);
4268         if (unlikely(!dev))
4269                 return -ENODEV;
4270
4271         idev = __in6_dev_get_safely(dev);
4272         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4273                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4274
4275         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4276                 fl6.flowi6_iif = 1;
4277                 oif = fl6.flowi6_oif = params->ifindex;
4278         } else {
4279                 oif = fl6.flowi6_iif = params->ifindex;
4280                 fl6.flowi6_oif = 0;
4281                 strict = RT6_LOOKUP_F_HAS_SADDR;
4282         }
4283         fl6.flowlabel = params->flowinfo;
4284         fl6.flowi6_scope = 0;
4285         fl6.flowi6_flags = 0;
4286         fl6.mp_hash = 0;
4287
4288         fl6.flowi6_proto = params->l4_protocol;
4289         fl6.daddr = *dst;
4290         fl6.saddr = *src;
4291         fl6.fl6_sport = params->sport;
4292         fl6.fl6_dport = params->dport;
4293
4294         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4295                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4296                 struct fib6_table *tb;
4297
4298                 tb = ipv6_stub->fib6_get_table(net, tbid);
4299                 if (unlikely(!tb))
4300                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4301
4302                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4303         } else {
4304                 fl6.flowi6_mark = 0;
4305                 fl6.flowi6_secid = 0;
4306                 fl6.flowi6_tun_key.tun_id = 0;
4307                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4308
4309                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4310         }
4311
4312         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4313                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4314
4315         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4316                 switch (f6i->fib6_type) {
4317                 case RTN_BLACKHOLE:
4318                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4319                 case RTN_UNREACHABLE:
4320                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4321                 case RTN_PROHIBIT:
4322                         return BPF_FIB_LKUP_RET_PROHIBIT;
4323                 default:
4324                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4325                 }
4326         }
4327
4328         if (f6i->fib6_type != RTN_UNICAST)
4329                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4330
4331         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4332                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4333                                                        fl6.flowi6_oif, NULL,
4334                                                        strict);
4335
4336         if (check_mtu) {
4337                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4338                 if (params->tot_len > mtu)
4339                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4340         }
4341
4342         if (f6i->fib6_nh.nh_lwtstate)
4343                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4344
4345         if (f6i->fib6_flags & RTF_GATEWAY)
4346                 *dst = f6i->fib6_nh.nh_gw;
4347
4348         dev = f6i->fib6_nh.nh_dev;
4349         params->rt_metric = f6i->fib6_metric;
4350
4351         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4352          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4353          * because we need to get nd_tbl via the stub
4354          */
4355         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4356                                       ndisc_hashfn, dst, dev);
4357         if (!neigh)
4358                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4359
4360         return bpf_fib_set_fwd_params(params, neigh, dev);
4361 }
4362 #endif
4363
4364 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4365            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4366 {
4367         if (plen < sizeof(*params))
4368                 return -EINVAL;
4369
4370         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4371                 return -EINVAL;
4372
4373         switch (params->family) {
4374 #if IS_ENABLED(CONFIG_INET)
4375         case AF_INET:
4376                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4377                                            flags, true);
4378 #endif
4379 #if IS_ENABLED(CONFIG_IPV6)
4380         case AF_INET6:
4381                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4382                                            flags, true);
4383 #endif
4384         }
4385         return -EAFNOSUPPORT;
4386 }
4387
4388 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4389         .func           = bpf_xdp_fib_lookup,
4390         .gpl_only       = true,
4391         .ret_type       = RET_INTEGER,
4392         .arg1_type      = ARG_PTR_TO_CTX,
4393         .arg2_type      = ARG_PTR_TO_MEM,
4394         .arg3_type      = ARG_CONST_SIZE,
4395         .arg4_type      = ARG_ANYTHING,
4396 };
4397
4398 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4399            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4400 {
4401         struct net *net = dev_net(skb->dev);
4402         int rc = -EAFNOSUPPORT;
4403
4404         if (plen < sizeof(*params))
4405                 return -EINVAL;
4406
4407         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4408                 return -EINVAL;
4409
4410         switch (params->family) {
4411 #if IS_ENABLED(CONFIG_INET)
4412         case AF_INET:
4413                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4414                 break;
4415 #endif
4416 #if IS_ENABLED(CONFIG_IPV6)
4417         case AF_INET6:
4418                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4419                 break;
4420 #endif
4421         }
4422
4423         if (!rc) {
4424                 struct net_device *dev;
4425
4426                 dev = dev_get_by_index_rcu(net, params->ifindex);
4427                 if (!is_skb_forwardable(dev, skb))
4428                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4429         }
4430
4431         return rc;
4432 }
4433
4434 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4435         .func           = bpf_skb_fib_lookup,
4436         .gpl_only       = true,
4437         .ret_type       = RET_INTEGER,
4438         .arg1_type      = ARG_PTR_TO_CTX,
4439         .arg2_type      = ARG_PTR_TO_MEM,
4440         .arg3_type      = ARG_CONST_SIZE,
4441         .arg4_type      = ARG_ANYTHING,
4442 };
4443
4444 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4445 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4446 {
4447         int err;
4448         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4449
4450         if (!seg6_validate_srh(srh, len))
4451                 return -EINVAL;
4452
4453         switch (type) {
4454         case BPF_LWT_ENCAP_SEG6_INLINE:
4455                 if (skb->protocol != htons(ETH_P_IPV6))
4456                         return -EBADMSG;
4457
4458                 err = seg6_do_srh_inline(skb, srh);
4459                 break;
4460         case BPF_LWT_ENCAP_SEG6:
4461                 skb_reset_inner_headers(skb);
4462                 skb->encapsulation = 1;
4463                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4464                 break;
4465         default:
4466                 return -EINVAL;
4467         }
4468
4469         bpf_compute_data_pointers(skb);
4470         if (err)
4471                 return err;
4472
4473         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4474         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4475
4476         return seg6_lookup_nexthop(skb, NULL, 0);
4477 }
4478 #endif /* CONFIG_IPV6_SEG6_BPF */
4479
4480 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4481            u32, len)
4482 {
4483         switch (type) {
4484 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4485         case BPF_LWT_ENCAP_SEG6:
4486         case BPF_LWT_ENCAP_SEG6_INLINE:
4487                 return bpf_push_seg6_encap(skb, type, hdr, len);
4488 #endif
4489         default:
4490                 return -EINVAL;
4491         }
4492 }
4493
4494 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4495         .func           = bpf_lwt_push_encap,
4496         .gpl_only       = false,
4497         .ret_type       = RET_INTEGER,
4498         .arg1_type      = ARG_PTR_TO_CTX,
4499         .arg2_type      = ARG_ANYTHING,
4500         .arg3_type      = ARG_PTR_TO_MEM,
4501         .arg4_type      = ARG_CONST_SIZE
4502 };
4503
4504 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4505 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4506            const void *, from, u32, len)
4507 {
4508         struct seg6_bpf_srh_state *srh_state =
4509                 this_cpu_ptr(&seg6_bpf_srh_states);
4510         struct ipv6_sr_hdr *srh = srh_state->srh;
4511         void *srh_tlvs, *srh_end, *ptr;
4512         int srhoff = 0;
4513
4514         if (srh == NULL)
4515                 return -EINVAL;
4516
4517         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4518         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4519
4520         ptr = skb->data + offset;
4521         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4522                 srh_state->valid = false;
4523         else if (ptr < (void *)&srh->flags ||
4524                  ptr + len > (void *)&srh->segments)
4525                 return -EFAULT;
4526
4527         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4528                 return -EFAULT;
4529         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4530                 return -EINVAL;
4531         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4532
4533         memcpy(skb->data + offset, from, len);
4534         return 0;
4535 }
4536
4537 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4538         .func           = bpf_lwt_seg6_store_bytes,
4539         .gpl_only       = false,
4540         .ret_type       = RET_INTEGER,
4541         .arg1_type      = ARG_PTR_TO_CTX,
4542         .arg2_type      = ARG_ANYTHING,
4543         .arg3_type      = ARG_PTR_TO_MEM,
4544         .arg4_type      = ARG_CONST_SIZE
4545 };
4546
4547 static void bpf_update_srh_state(struct sk_buff *skb)
4548 {
4549         struct seg6_bpf_srh_state *srh_state =
4550                 this_cpu_ptr(&seg6_bpf_srh_states);
4551         int srhoff = 0;
4552
4553         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4554                 srh_state->srh = NULL;
4555         } else {
4556                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4557                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4558                 srh_state->valid = true;
4559         }
4560 }
4561
4562 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4563            u32, action, void *, param, u32, param_len)
4564 {
4565         struct seg6_bpf_srh_state *srh_state =
4566                 this_cpu_ptr(&seg6_bpf_srh_states);
4567         int hdroff = 0;
4568         int err;
4569
4570         switch (action) {
4571         case SEG6_LOCAL_ACTION_END_X:
4572                 if (!seg6_bpf_has_valid_srh(skb))
4573                         return -EBADMSG;
4574                 if (param_len != sizeof(struct in6_addr))
4575                         return -EINVAL;
4576                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4577         case SEG6_LOCAL_ACTION_END_T:
4578                 if (!seg6_bpf_has_valid_srh(skb))
4579                         return -EBADMSG;
4580                 if (param_len != sizeof(int))
4581                         return -EINVAL;
4582                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4583         case SEG6_LOCAL_ACTION_END_DT6:
4584                 if (!seg6_bpf_has_valid_srh(skb))
4585                         return -EBADMSG;
4586                 if (param_len != sizeof(int))
4587                         return -EINVAL;
4588
4589                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4590                         return -EBADMSG;
4591                 if (!pskb_pull(skb, hdroff))
4592                         return -EBADMSG;
4593
4594                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4595                 skb_reset_network_header(skb);
4596                 skb_reset_transport_header(skb);
4597                 skb->encapsulation = 0;
4598
4599                 bpf_compute_data_pointers(skb);
4600                 bpf_update_srh_state(skb);
4601                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4602         case SEG6_LOCAL_ACTION_END_B6:
4603                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4604                         return -EBADMSG;
4605                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4606                                           param, param_len);
4607                 if (!err)
4608                         bpf_update_srh_state(skb);
4609
4610                 return err;
4611         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4612                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4613                         return -EBADMSG;
4614                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4615                                           param, param_len);
4616                 if (!err)
4617                         bpf_update_srh_state(skb);
4618
4619                 return err;
4620         default:
4621                 return -EINVAL;
4622         }
4623 }
4624
4625 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4626         .func           = bpf_lwt_seg6_action,
4627         .gpl_only       = false,
4628         .ret_type       = RET_INTEGER,
4629         .arg1_type      = ARG_PTR_TO_CTX,
4630         .arg2_type      = ARG_ANYTHING,
4631         .arg3_type      = ARG_PTR_TO_MEM,
4632         .arg4_type      = ARG_CONST_SIZE
4633 };
4634
4635 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4636            s32, len)
4637 {
4638         struct seg6_bpf_srh_state *srh_state =
4639                 this_cpu_ptr(&seg6_bpf_srh_states);
4640         struct ipv6_sr_hdr *srh = srh_state->srh;
4641         void *srh_end, *srh_tlvs, *ptr;
4642         struct ipv6hdr *hdr;
4643         int srhoff = 0;
4644         int ret;
4645
4646         if (unlikely(srh == NULL))
4647                 return -EINVAL;
4648
4649         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4650                         ((srh->first_segment + 1) << 4));
4651         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4652                         srh_state->hdrlen);
4653         ptr = skb->data + offset;
4654
4655         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4656                 return -EFAULT;
4657         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4658                 return -EFAULT;
4659
4660         if (len > 0) {
4661                 ret = skb_cow_head(skb, len);
4662                 if (unlikely(ret < 0))
4663                         return ret;
4664
4665                 ret = bpf_skb_net_hdr_push(skb, offset, len);
4666         } else {
4667                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4668         }
4669
4670         bpf_compute_data_pointers(skb);
4671         if (unlikely(ret < 0))
4672                 return ret;
4673
4674         hdr = (struct ipv6hdr *)skb->data;
4675         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4676
4677         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4678                 return -EINVAL;
4679         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4680         srh_state->hdrlen += len;
4681         srh_state->valid = false;
4682         return 0;
4683 }
4684
4685 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4686         .func           = bpf_lwt_seg6_adjust_srh,
4687         .gpl_only       = false,
4688         .ret_type       = RET_INTEGER,
4689         .arg1_type      = ARG_PTR_TO_CTX,
4690         .arg2_type      = ARG_ANYTHING,
4691         .arg3_type      = ARG_ANYTHING,
4692 };
4693 #endif /* CONFIG_IPV6_SEG6_BPF */
4694
4695 #ifdef CONFIG_INET
4696 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
4697                               struct sk_buff *skb, u8 family, u8 proto)
4698 {
4699         bool refcounted = false;
4700         struct sock *sk = NULL;
4701         int dif = 0;
4702
4703         if (skb->dev)
4704                 dif = skb->dev->ifindex;
4705
4706         if (family == AF_INET) {
4707                 __be32 src4 = tuple->ipv4.saddr;
4708                 __be32 dst4 = tuple->ipv4.daddr;
4709                 int sdif = inet_sdif(skb);
4710
4711                 if (proto == IPPROTO_TCP)
4712                         sk = __inet_lookup(net, &tcp_hashinfo, skb, 0,
4713                                            src4, tuple->ipv4.sport,
4714                                            dst4, tuple->ipv4.dport,
4715                                            dif, sdif, &refcounted);
4716                 else
4717                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
4718                                                dst4, tuple->ipv4.dport,
4719                                                dif, sdif, &udp_table, skb);
4720 #if IS_REACHABLE(CONFIG_IPV6)
4721         } else {
4722                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
4723                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
4724                 int sdif = inet6_sdif(skb);
4725
4726                 if (proto == IPPROTO_TCP)
4727                         sk = __inet6_lookup(net, &tcp_hashinfo, skb, 0,
4728                                             src6, tuple->ipv6.sport,
4729                                             dst6, tuple->ipv6.dport,
4730                                             dif, sdif, &refcounted);
4731                 else
4732                         sk = __udp6_lib_lookup(net, src6, tuple->ipv6.sport,
4733                                                dst6, tuple->ipv6.dport,
4734                                                dif, sdif, &udp_table, skb);
4735 #endif
4736         }
4737
4738         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
4739                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
4740                 sk = NULL;
4741         }
4742         return sk;
4743 }
4744
4745 /* bpf_sk_lookup performs the core lookup for different types of sockets,
4746  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
4747  * Returns the socket as an 'unsigned long' to simplify the casting in the
4748  * callers to satisfy BPF_CALL declarations.
4749  */
4750 static unsigned long
4751 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
4752               u8 proto, u64 netns_id, u64 flags)
4753 {
4754         struct net *caller_net;
4755         struct sock *sk = NULL;
4756         u8 family = AF_UNSPEC;
4757         struct net *net;
4758
4759         family = len == sizeof(tuple->ipv4) ? AF_INET : AF_INET6;
4760         if (unlikely(family == AF_UNSPEC || netns_id > U32_MAX || flags))
4761                 goto out;
4762
4763         if (skb->dev)
4764                 caller_net = dev_net(skb->dev);
4765         else
4766                 caller_net = sock_net(skb->sk);
4767         if (netns_id) {
4768                 net = get_net_ns_by_id(caller_net, netns_id);
4769                 if (unlikely(!net))
4770                         goto out;
4771                 sk = sk_lookup(net, tuple, skb, family, proto);
4772                 put_net(net);
4773         } else {
4774                 net = caller_net;
4775                 sk = sk_lookup(net, tuple, skb, family, proto);
4776         }
4777
4778         if (sk)
4779                 sk = sk_to_full_sk(sk);
4780 out:
4781         return (unsigned long) sk;
4782 }
4783
4784 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
4785            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
4786 {
4787         return bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, netns_id, flags);
4788 }
4789
4790 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
4791         .func           = bpf_sk_lookup_tcp,
4792         .gpl_only       = false,
4793         .pkt_access     = true,
4794         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
4795         .arg1_type      = ARG_PTR_TO_CTX,
4796         .arg2_type      = ARG_PTR_TO_MEM,
4797         .arg3_type      = ARG_CONST_SIZE,
4798         .arg4_type      = ARG_ANYTHING,
4799         .arg5_type      = ARG_ANYTHING,
4800 };
4801
4802 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
4803            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
4804 {
4805         return bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, netns_id, flags);
4806 }
4807
4808 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
4809         .func           = bpf_sk_lookup_udp,
4810         .gpl_only       = false,
4811         .pkt_access     = true,
4812         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
4813         .arg1_type      = ARG_PTR_TO_CTX,
4814         .arg2_type      = ARG_PTR_TO_MEM,
4815         .arg3_type      = ARG_CONST_SIZE,
4816         .arg4_type      = ARG_ANYTHING,
4817         .arg5_type      = ARG_ANYTHING,
4818 };
4819
4820 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
4821 {
4822         if (!sock_flag(sk, SOCK_RCU_FREE))
4823                 sock_gen_put(sk);
4824         return 0;
4825 }
4826
4827 static const struct bpf_func_proto bpf_sk_release_proto = {
4828         .func           = bpf_sk_release,
4829         .gpl_only       = false,
4830         .ret_type       = RET_INTEGER,
4831         .arg1_type      = ARG_PTR_TO_SOCKET,
4832 };
4833 #endif /* CONFIG_INET */
4834
4835 bool bpf_helper_changes_pkt_data(void *func)
4836 {
4837         if (func == bpf_skb_vlan_push ||
4838             func == bpf_skb_vlan_pop ||
4839             func == bpf_skb_store_bytes ||
4840             func == bpf_skb_change_proto ||
4841             func == bpf_skb_change_head ||
4842             func == sk_skb_change_head ||
4843             func == bpf_skb_change_tail ||
4844             func == sk_skb_change_tail ||
4845             func == bpf_skb_adjust_room ||
4846             func == bpf_skb_pull_data ||
4847             func == sk_skb_pull_data ||
4848             func == bpf_clone_redirect ||
4849             func == bpf_l3_csum_replace ||
4850             func == bpf_l4_csum_replace ||
4851             func == bpf_xdp_adjust_head ||
4852             func == bpf_xdp_adjust_meta ||
4853             func == bpf_msg_pull_data ||
4854             func == bpf_xdp_adjust_tail ||
4855 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4856             func == bpf_lwt_seg6_store_bytes ||
4857             func == bpf_lwt_seg6_adjust_srh ||
4858             func == bpf_lwt_seg6_action ||
4859 #endif
4860             func == bpf_lwt_push_encap)
4861                 return true;
4862
4863         return false;
4864 }
4865
4866 static const struct bpf_func_proto *
4867 bpf_base_func_proto(enum bpf_func_id func_id)
4868 {
4869         switch (func_id) {
4870         case BPF_FUNC_map_lookup_elem:
4871                 return &bpf_map_lookup_elem_proto;
4872         case BPF_FUNC_map_update_elem:
4873                 return &bpf_map_update_elem_proto;
4874         case BPF_FUNC_map_delete_elem:
4875                 return &bpf_map_delete_elem_proto;
4876         case BPF_FUNC_get_prandom_u32:
4877                 return &bpf_get_prandom_u32_proto;
4878         case BPF_FUNC_get_smp_processor_id:
4879                 return &bpf_get_raw_smp_processor_id_proto;
4880         case BPF_FUNC_get_numa_node_id:
4881                 return &bpf_get_numa_node_id_proto;
4882         case BPF_FUNC_tail_call:
4883                 return &bpf_tail_call_proto;
4884         case BPF_FUNC_ktime_get_ns:
4885                 return &bpf_ktime_get_ns_proto;
4886         case BPF_FUNC_trace_printk:
4887                 if (capable(CAP_SYS_ADMIN))
4888                         return bpf_get_trace_printk_proto();
4889                 /* else: fall through */
4890         default:
4891                 return NULL;
4892         }
4893 }
4894
4895 static const struct bpf_func_proto *
4896 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4897 {
4898         switch (func_id) {
4899         /* inet and inet6 sockets are created in a process
4900          * context so there is always a valid uid/gid
4901          */
4902         case BPF_FUNC_get_current_uid_gid:
4903                 return &bpf_get_current_uid_gid_proto;
4904         case BPF_FUNC_get_local_storage:
4905                 return &bpf_get_local_storage_proto;
4906         default:
4907                 return bpf_base_func_proto(func_id);
4908         }
4909 }
4910
4911 static const struct bpf_func_proto *
4912 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4913 {
4914         switch (func_id) {
4915         /* inet and inet6 sockets are created in a process
4916          * context so there is always a valid uid/gid
4917          */
4918         case BPF_FUNC_get_current_uid_gid:
4919                 return &bpf_get_current_uid_gid_proto;
4920         case BPF_FUNC_bind:
4921                 switch (prog->expected_attach_type) {
4922                 case BPF_CGROUP_INET4_CONNECT:
4923                 case BPF_CGROUP_INET6_CONNECT:
4924                         return &bpf_bind_proto;
4925                 default:
4926                         return NULL;
4927                 }
4928         case BPF_FUNC_get_socket_cookie:
4929                 return &bpf_get_socket_cookie_sock_addr_proto;
4930         case BPF_FUNC_get_local_storage:
4931                 return &bpf_get_local_storage_proto;
4932         default:
4933                 return bpf_base_func_proto(func_id);
4934         }
4935 }
4936
4937 static const struct bpf_func_proto *
4938 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4939 {
4940         switch (func_id) {
4941         case BPF_FUNC_skb_load_bytes:
4942                 return &bpf_skb_load_bytes_proto;
4943         case BPF_FUNC_skb_load_bytes_relative:
4944                 return &bpf_skb_load_bytes_relative_proto;
4945         case BPF_FUNC_get_socket_cookie:
4946                 return &bpf_get_socket_cookie_proto;
4947         case BPF_FUNC_get_socket_uid:
4948                 return &bpf_get_socket_uid_proto;
4949         default:
4950                 return bpf_base_func_proto(func_id);
4951         }
4952 }
4953
4954 static const struct bpf_func_proto *
4955 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4956 {
4957         switch (func_id) {
4958         case BPF_FUNC_get_local_storage:
4959                 return &bpf_get_local_storage_proto;
4960         default:
4961                 return sk_filter_func_proto(func_id, prog);
4962         }
4963 }
4964
4965 static const struct bpf_func_proto *
4966 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4967 {
4968         switch (func_id) {
4969         case BPF_FUNC_skb_store_bytes:
4970                 return &bpf_skb_store_bytes_proto;
4971         case BPF_FUNC_skb_load_bytes:
4972                 return &bpf_skb_load_bytes_proto;
4973         case BPF_FUNC_skb_load_bytes_relative:
4974                 return &bpf_skb_load_bytes_relative_proto;
4975         case BPF_FUNC_skb_pull_data:
4976                 return &bpf_skb_pull_data_proto;
4977         case BPF_FUNC_csum_diff:
4978                 return &bpf_csum_diff_proto;
4979         case BPF_FUNC_csum_update:
4980                 return &bpf_csum_update_proto;
4981         case BPF_FUNC_l3_csum_replace:
4982                 return &bpf_l3_csum_replace_proto;
4983         case BPF_FUNC_l4_csum_replace:
4984                 return &bpf_l4_csum_replace_proto;
4985         case BPF_FUNC_clone_redirect:
4986                 return &bpf_clone_redirect_proto;
4987         case BPF_FUNC_get_cgroup_classid:
4988                 return &bpf_get_cgroup_classid_proto;
4989         case BPF_FUNC_skb_vlan_push:
4990                 return &bpf_skb_vlan_push_proto;
4991         case BPF_FUNC_skb_vlan_pop:
4992                 return &bpf_skb_vlan_pop_proto;
4993         case BPF_FUNC_skb_change_proto:
4994                 return &bpf_skb_change_proto_proto;
4995         case BPF_FUNC_skb_change_type:
4996                 return &bpf_skb_change_type_proto;
4997         case BPF_FUNC_skb_adjust_room:
4998                 return &bpf_skb_adjust_room_proto;
4999         case BPF_FUNC_skb_change_tail:
5000                 return &bpf_skb_change_tail_proto;
5001         case BPF_FUNC_skb_get_tunnel_key:
5002                 return &bpf_skb_get_tunnel_key_proto;
5003         case BPF_FUNC_skb_set_tunnel_key:
5004                 return bpf_get_skb_set_tunnel_proto(func_id);
5005         case BPF_FUNC_skb_get_tunnel_opt:
5006                 return &bpf_skb_get_tunnel_opt_proto;
5007         case BPF_FUNC_skb_set_tunnel_opt:
5008                 return bpf_get_skb_set_tunnel_proto(func_id);
5009         case BPF_FUNC_redirect:
5010                 return &bpf_redirect_proto;
5011         case BPF_FUNC_get_route_realm:
5012                 return &bpf_get_route_realm_proto;
5013         case BPF_FUNC_get_hash_recalc:
5014                 return &bpf_get_hash_recalc_proto;
5015         case BPF_FUNC_set_hash_invalid:
5016                 return &bpf_set_hash_invalid_proto;
5017         case BPF_FUNC_set_hash:
5018                 return &bpf_set_hash_proto;
5019         case BPF_FUNC_perf_event_output:
5020                 return &bpf_skb_event_output_proto;
5021         case BPF_FUNC_get_smp_processor_id:
5022                 return &bpf_get_smp_processor_id_proto;
5023         case BPF_FUNC_skb_under_cgroup:
5024                 return &bpf_skb_under_cgroup_proto;
5025         case BPF_FUNC_get_socket_cookie:
5026                 return &bpf_get_socket_cookie_proto;
5027         case BPF_FUNC_get_socket_uid:
5028                 return &bpf_get_socket_uid_proto;
5029         case BPF_FUNC_fib_lookup:
5030                 return &bpf_skb_fib_lookup_proto;
5031 #ifdef CONFIG_XFRM
5032         case BPF_FUNC_skb_get_xfrm_state:
5033                 return &bpf_skb_get_xfrm_state_proto;
5034 #endif
5035 #ifdef CONFIG_SOCK_CGROUP_DATA
5036         case BPF_FUNC_skb_cgroup_id:
5037                 return &bpf_skb_cgroup_id_proto;
5038         case BPF_FUNC_skb_ancestor_cgroup_id:
5039                 return &bpf_skb_ancestor_cgroup_id_proto;
5040 #endif
5041 #ifdef CONFIG_INET
5042         case BPF_FUNC_sk_lookup_tcp:
5043                 return &bpf_sk_lookup_tcp_proto;
5044         case BPF_FUNC_sk_lookup_udp:
5045                 return &bpf_sk_lookup_udp_proto;
5046         case BPF_FUNC_sk_release:
5047                 return &bpf_sk_release_proto;
5048 #endif
5049         default:
5050                 return bpf_base_func_proto(func_id);
5051         }
5052 }
5053
5054 static const struct bpf_func_proto *
5055 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5056 {
5057         switch (func_id) {
5058         case BPF_FUNC_perf_event_output:
5059                 return &bpf_xdp_event_output_proto;
5060         case BPF_FUNC_get_smp_processor_id:
5061                 return &bpf_get_smp_processor_id_proto;
5062         case BPF_FUNC_csum_diff:
5063                 return &bpf_csum_diff_proto;
5064         case BPF_FUNC_xdp_adjust_head:
5065                 return &bpf_xdp_adjust_head_proto;
5066         case BPF_FUNC_xdp_adjust_meta:
5067                 return &bpf_xdp_adjust_meta_proto;
5068         case BPF_FUNC_redirect:
5069                 return &bpf_xdp_redirect_proto;
5070         case BPF_FUNC_redirect_map:
5071                 return &bpf_xdp_redirect_map_proto;
5072         case BPF_FUNC_xdp_adjust_tail:
5073                 return &bpf_xdp_adjust_tail_proto;
5074         case BPF_FUNC_fib_lookup:
5075                 return &bpf_xdp_fib_lookup_proto;
5076         default:
5077                 return bpf_base_func_proto(func_id);
5078         }
5079 }
5080
5081 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5082 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5083
5084 static const struct bpf_func_proto *
5085 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5086 {
5087         switch (func_id) {
5088         case BPF_FUNC_setsockopt:
5089                 return &bpf_setsockopt_proto;
5090         case BPF_FUNC_getsockopt:
5091                 return &bpf_getsockopt_proto;
5092         case BPF_FUNC_sock_ops_cb_flags_set:
5093                 return &bpf_sock_ops_cb_flags_set_proto;
5094         case BPF_FUNC_sock_map_update:
5095                 return &bpf_sock_map_update_proto;
5096         case BPF_FUNC_sock_hash_update:
5097                 return &bpf_sock_hash_update_proto;
5098         case BPF_FUNC_get_socket_cookie:
5099                 return &bpf_get_socket_cookie_sock_ops_proto;
5100         case BPF_FUNC_get_local_storage:
5101                 return &bpf_get_local_storage_proto;
5102         default:
5103                 return bpf_base_func_proto(func_id);
5104         }
5105 }
5106
5107 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5108 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5109
5110 static const struct bpf_func_proto *
5111 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5112 {
5113         switch (func_id) {
5114         case BPF_FUNC_msg_redirect_map:
5115                 return &bpf_msg_redirect_map_proto;
5116         case BPF_FUNC_msg_redirect_hash:
5117                 return &bpf_msg_redirect_hash_proto;
5118         case BPF_FUNC_msg_apply_bytes:
5119                 return &bpf_msg_apply_bytes_proto;
5120         case BPF_FUNC_msg_cork_bytes:
5121                 return &bpf_msg_cork_bytes_proto;
5122         case BPF_FUNC_msg_pull_data:
5123                 return &bpf_msg_pull_data_proto;
5124         case BPF_FUNC_get_local_storage:
5125                 return &bpf_get_local_storage_proto;
5126         default:
5127                 return bpf_base_func_proto(func_id);
5128         }
5129 }
5130
5131 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5132 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5133
5134 static const struct bpf_func_proto *
5135 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5136 {
5137         switch (func_id) {
5138         case BPF_FUNC_skb_store_bytes:
5139                 return &bpf_skb_store_bytes_proto;
5140         case BPF_FUNC_skb_load_bytes:
5141                 return &bpf_skb_load_bytes_proto;
5142         case BPF_FUNC_skb_pull_data:
5143                 return &sk_skb_pull_data_proto;
5144         case BPF_FUNC_skb_change_tail:
5145                 return &sk_skb_change_tail_proto;
5146         case BPF_FUNC_skb_change_head:
5147                 return &sk_skb_change_head_proto;
5148         case BPF_FUNC_get_socket_cookie:
5149                 return &bpf_get_socket_cookie_proto;
5150         case BPF_FUNC_get_socket_uid:
5151                 return &bpf_get_socket_uid_proto;
5152         case BPF_FUNC_sk_redirect_map:
5153                 return &bpf_sk_redirect_map_proto;
5154         case BPF_FUNC_sk_redirect_hash:
5155                 return &bpf_sk_redirect_hash_proto;
5156         case BPF_FUNC_get_local_storage:
5157                 return &bpf_get_local_storage_proto;
5158 #ifdef CONFIG_INET
5159         case BPF_FUNC_sk_lookup_tcp:
5160                 return &bpf_sk_lookup_tcp_proto;
5161         case BPF_FUNC_sk_lookup_udp:
5162                 return &bpf_sk_lookup_udp_proto;
5163         case BPF_FUNC_sk_release:
5164                 return &bpf_sk_release_proto;
5165 #endif
5166         default:
5167                 return bpf_base_func_proto(func_id);
5168         }
5169 }
5170
5171 static const struct bpf_func_proto *
5172 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5173 {
5174         switch (func_id) {
5175         case BPF_FUNC_skb_load_bytes:
5176                 return &bpf_skb_load_bytes_proto;
5177         default:
5178                 return bpf_base_func_proto(func_id);
5179         }
5180 }
5181
5182 static const struct bpf_func_proto *
5183 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5184 {
5185         switch (func_id) {
5186         case BPF_FUNC_skb_load_bytes:
5187                 return &bpf_skb_load_bytes_proto;
5188         case BPF_FUNC_skb_pull_data:
5189                 return &bpf_skb_pull_data_proto;
5190         case BPF_FUNC_csum_diff:
5191                 return &bpf_csum_diff_proto;
5192         case BPF_FUNC_get_cgroup_classid:
5193                 return &bpf_get_cgroup_classid_proto;
5194         case BPF_FUNC_get_route_realm:
5195                 return &bpf_get_route_realm_proto;
5196         case BPF_FUNC_get_hash_recalc:
5197                 return &bpf_get_hash_recalc_proto;
5198         case BPF_FUNC_perf_event_output:
5199                 return &bpf_skb_event_output_proto;
5200         case BPF_FUNC_get_smp_processor_id:
5201                 return &bpf_get_smp_processor_id_proto;
5202         case BPF_FUNC_skb_under_cgroup:
5203                 return &bpf_skb_under_cgroup_proto;
5204         default:
5205                 return bpf_base_func_proto(func_id);
5206         }
5207 }
5208
5209 static const struct bpf_func_proto *
5210 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5211 {
5212         switch (func_id) {
5213         case BPF_FUNC_lwt_push_encap:
5214                 return &bpf_lwt_push_encap_proto;
5215         default:
5216                 return lwt_out_func_proto(func_id, prog);
5217         }
5218 }
5219
5220 static const struct bpf_func_proto *
5221 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5222 {
5223         switch (func_id) {
5224         case BPF_FUNC_skb_get_tunnel_key:
5225                 return &bpf_skb_get_tunnel_key_proto;
5226         case BPF_FUNC_skb_set_tunnel_key:
5227                 return bpf_get_skb_set_tunnel_proto(func_id);
5228         case BPF_FUNC_skb_get_tunnel_opt:
5229                 return &bpf_skb_get_tunnel_opt_proto;
5230         case BPF_FUNC_skb_set_tunnel_opt:
5231                 return bpf_get_skb_set_tunnel_proto(func_id);
5232         case BPF_FUNC_redirect:
5233                 return &bpf_redirect_proto;
5234         case BPF_FUNC_clone_redirect:
5235                 return &bpf_clone_redirect_proto;
5236         case BPF_FUNC_skb_change_tail:
5237                 return &bpf_skb_change_tail_proto;
5238         case BPF_FUNC_skb_change_head:
5239                 return &bpf_skb_change_head_proto;
5240         case BPF_FUNC_skb_store_bytes:
5241                 return &bpf_skb_store_bytes_proto;
5242         case BPF_FUNC_csum_update:
5243                 return &bpf_csum_update_proto;
5244         case BPF_FUNC_l3_csum_replace:
5245                 return &bpf_l3_csum_replace_proto;
5246         case BPF_FUNC_l4_csum_replace:
5247                 return &bpf_l4_csum_replace_proto;
5248         case BPF_FUNC_set_hash_invalid:
5249                 return &bpf_set_hash_invalid_proto;
5250         default:
5251                 return lwt_out_func_proto(func_id, prog);
5252         }
5253 }
5254
5255 static const struct bpf_func_proto *
5256 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5257 {
5258         switch (func_id) {
5259 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5260         case BPF_FUNC_lwt_seg6_store_bytes:
5261                 return &bpf_lwt_seg6_store_bytes_proto;
5262         case BPF_FUNC_lwt_seg6_action:
5263                 return &bpf_lwt_seg6_action_proto;
5264         case BPF_FUNC_lwt_seg6_adjust_srh:
5265                 return &bpf_lwt_seg6_adjust_srh_proto;
5266 #endif
5267         default:
5268                 return lwt_out_func_proto(func_id, prog);
5269         }
5270 }
5271
5272 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5273                                     const struct bpf_prog *prog,
5274                                     struct bpf_insn_access_aux *info)
5275 {
5276         const int size_default = sizeof(__u32);
5277
5278         if (off < 0 || off >= sizeof(struct __sk_buff))
5279                 return false;
5280
5281         /* The verifier guarantees that size > 0. */
5282         if (off % size != 0)
5283                 return false;
5284
5285         switch (off) {
5286         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5287                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5288                         return false;
5289                 break;
5290         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5291         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5292         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5293         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5294         case bpf_ctx_range(struct __sk_buff, data):
5295         case bpf_ctx_range(struct __sk_buff, data_meta):
5296         case bpf_ctx_range(struct __sk_buff, data_end):
5297                 if (size != size_default)
5298                         return false;
5299                 break;
5300         case bpf_ctx_range(struct __sk_buff, flow_keys):
5301                 if (size != sizeof(struct bpf_flow_keys *))
5302                         return false;
5303                 break;
5304         default:
5305                 /* Only narrow read access allowed for now. */
5306                 if (type == BPF_WRITE) {
5307                         if (size != size_default)
5308                                 return false;
5309                 } else {
5310                         bpf_ctx_record_field_size(info, size_default);
5311                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5312                                 return false;
5313                 }
5314         }
5315
5316         return true;
5317 }
5318
5319 static bool sk_filter_is_valid_access(int off, int size,
5320                                       enum bpf_access_type type,
5321                                       const struct bpf_prog *prog,
5322                                       struct bpf_insn_access_aux *info)
5323 {
5324         switch (off) {
5325         case bpf_ctx_range(struct __sk_buff, tc_classid):
5326         case bpf_ctx_range(struct __sk_buff, data):
5327         case bpf_ctx_range(struct __sk_buff, data_meta):
5328         case bpf_ctx_range(struct __sk_buff, data_end):
5329         case bpf_ctx_range(struct __sk_buff, flow_keys):
5330         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5331                 return false;
5332         }
5333
5334         if (type == BPF_WRITE) {
5335                 switch (off) {
5336                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5337                         break;
5338                 default:
5339                         return false;
5340                 }
5341         }
5342
5343         return bpf_skb_is_valid_access(off, size, type, prog, info);
5344 }
5345
5346 static bool lwt_is_valid_access(int off, int size,
5347                                 enum bpf_access_type type,
5348                                 const struct bpf_prog *prog,
5349                                 struct bpf_insn_access_aux *info)
5350 {
5351         switch (off) {
5352         case bpf_ctx_range(struct __sk_buff, tc_classid):
5353         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5354         case bpf_ctx_range(struct __sk_buff, data_meta):
5355         case bpf_ctx_range(struct __sk_buff, flow_keys):
5356                 return false;
5357         }
5358
5359         if (type == BPF_WRITE) {
5360                 switch (off) {
5361                 case bpf_ctx_range(struct __sk_buff, mark):
5362                 case bpf_ctx_range(struct __sk_buff, priority):
5363                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5364                         break;
5365                 default:
5366                         return false;
5367                 }
5368         }
5369
5370         switch (off) {
5371         case bpf_ctx_range(struct __sk_buff, data):
5372                 info->reg_type = PTR_TO_PACKET;
5373                 break;
5374         case bpf_ctx_range(struct __sk_buff, data_end):
5375                 info->reg_type = PTR_TO_PACKET_END;
5376                 break;
5377         }
5378
5379         return bpf_skb_is_valid_access(off, size, type, prog, info);
5380 }
5381
5382 /* Attach type specific accesses */
5383 static bool __sock_filter_check_attach_type(int off,
5384                                             enum bpf_access_type access_type,
5385                                             enum bpf_attach_type attach_type)
5386 {
5387         switch (off) {
5388         case offsetof(struct bpf_sock, bound_dev_if):
5389         case offsetof(struct bpf_sock, mark):
5390         case offsetof(struct bpf_sock, priority):
5391                 switch (attach_type) {
5392                 case BPF_CGROUP_INET_SOCK_CREATE:
5393                         goto full_access;
5394                 default:
5395                         return false;
5396                 }
5397         case bpf_ctx_range(struct bpf_sock, src_ip4):
5398                 switch (attach_type) {
5399                 case BPF_CGROUP_INET4_POST_BIND:
5400                         goto read_only;
5401                 default:
5402                         return false;
5403                 }
5404         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5405                 switch (attach_type) {
5406                 case BPF_CGROUP_INET6_POST_BIND:
5407                         goto read_only;
5408                 default:
5409                         return false;
5410                 }
5411         case bpf_ctx_range(struct bpf_sock, src_port):
5412                 switch (attach_type) {
5413                 case BPF_CGROUP_INET4_POST_BIND:
5414                 case BPF_CGROUP_INET6_POST_BIND:
5415                         goto read_only;
5416                 default:
5417                         return false;
5418                 }
5419         }
5420 read_only:
5421         return access_type == BPF_READ;
5422 full_access:
5423         return true;
5424 }
5425
5426 static bool __sock_filter_check_size(int off, int size,
5427                                      struct bpf_insn_access_aux *info)
5428 {
5429         const int size_default = sizeof(__u32);
5430
5431         switch (off) {
5432         case bpf_ctx_range(struct bpf_sock, src_ip4):
5433         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5434                 bpf_ctx_record_field_size(info, size_default);
5435                 return bpf_ctx_narrow_access_ok(off, size, size_default);
5436         }
5437
5438         return size == size_default;
5439 }
5440
5441 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5442                               struct bpf_insn_access_aux *info)
5443 {
5444         if (off < 0 || off >= sizeof(struct bpf_sock))
5445                 return false;
5446         if (off % size != 0)
5447                 return false;
5448         if (!__sock_filter_check_size(off, size, info))
5449                 return false;
5450         return true;
5451 }
5452
5453 static bool sock_filter_is_valid_access(int off, int size,
5454                                         enum bpf_access_type type,
5455                                         const struct bpf_prog *prog,
5456                                         struct bpf_insn_access_aux *info)
5457 {
5458         if (!bpf_sock_is_valid_access(off, size, type, info))
5459                 return false;
5460         return __sock_filter_check_attach_type(off, type,
5461                                                prog->expected_attach_type);
5462 }
5463
5464 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5465                                 const struct bpf_prog *prog, int drop_verdict)
5466 {
5467         struct bpf_insn *insn = insn_buf;
5468
5469         if (!direct_write)
5470                 return 0;
5471
5472         /* if (!skb->cloned)
5473          *       goto start;
5474          *
5475          * (Fast-path, otherwise approximation that we might be
5476          *  a clone, do the rest in helper.)
5477          */
5478         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5479         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5480         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5481
5482         /* ret = bpf_skb_pull_data(skb, 0); */
5483         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5484         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5485         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5486                                BPF_FUNC_skb_pull_data);
5487         /* if (!ret)
5488          *      goto restore;
5489          * return TC_ACT_SHOT;
5490          */
5491         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5492         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5493         *insn++ = BPF_EXIT_INSN();
5494
5495         /* restore: */
5496         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5497         /* start: */
5498         *insn++ = prog->insnsi[0];
5499
5500         return insn - insn_buf;
5501 }
5502
5503 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5504                           struct bpf_insn *insn_buf)
5505 {
5506         bool indirect = BPF_MODE(orig->code) == BPF_IND;
5507         struct bpf_insn *insn = insn_buf;
5508
5509         /* We're guaranteed here that CTX is in R6. */
5510         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5511         if (!indirect) {
5512                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5513         } else {
5514                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5515                 if (orig->imm)
5516                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5517         }
5518
5519         switch (BPF_SIZE(orig->code)) {
5520         case BPF_B:
5521                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5522                 break;
5523         case BPF_H:
5524                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5525                 break;
5526         case BPF_W:
5527                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5528                 break;
5529         }
5530
5531         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5532         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5533         *insn++ = BPF_EXIT_INSN();
5534
5535         return insn - insn_buf;
5536 }
5537
5538 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5539                                const struct bpf_prog *prog)
5540 {
5541         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5542 }
5543
5544 static bool tc_cls_act_is_valid_access(int off, int size,
5545                                        enum bpf_access_type type,
5546                                        const struct bpf_prog *prog,
5547                                        struct bpf_insn_access_aux *info)
5548 {
5549         if (type == BPF_WRITE) {
5550                 switch (off) {
5551                 case bpf_ctx_range(struct __sk_buff, mark):
5552                 case bpf_ctx_range(struct __sk_buff, tc_index):
5553                 case bpf_ctx_range(struct __sk_buff, priority):
5554                 case bpf_ctx_range(struct __sk_buff, tc_classid):
5555                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5556                         break;
5557                 default:
5558                         return false;
5559                 }
5560         }
5561
5562         switch (off) {
5563         case bpf_ctx_range(struct __sk_buff, data):
5564                 info->reg_type = PTR_TO_PACKET;
5565                 break;
5566         case bpf_ctx_range(struct __sk_buff, data_meta):
5567                 info->reg_type = PTR_TO_PACKET_META;
5568                 break;
5569         case bpf_ctx_range(struct __sk_buff, data_end):
5570                 info->reg_type = PTR_TO_PACKET_END;
5571                 break;
5572         case bpf_ctx_range(struct __sk_buff, flow_keys):
5573         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5574                 return false;
5575         }
5576
5577         return bpf_skb_is_valid_access(off, size, type, prog, info);
5578 }
5579
5580 static bool __is_valid_xdp_access(int off, int size)
5581 {
5582         if (off < 0 || off >= sizeof(struct xdp_md))
5583                 return false;
5584         if (off % size != 0)
5585                 return false;
5586         if (size != sizeof(__u32))
5587                 return false;
5588
5589         return true;
5590 }
5591
5592 static bool xdp_is_valid_access(int off, int size,
5593                                 enum bpf_access_type type,
5594                                 const struct bpf_prog *prog,
5595                                 struct bpf_insn_access_aux *info)
5596 {
5597         if (type == BPF_WRITE) {
5598                 if (bpf_prog_is_dev_bound(prog->aux)) {
5599                         switch (off) {
5600                         case offsetof(struct xdp_md, rx_queue_index):
5601                                 return __is_valid_xdp_access(off, size);
5602                         }
5603                 }
5604                 return false;
5605         }
5606
5607         switch (off) {
5608         case offsetof(struct xdp_md, data):
5609                 info->reg_type = PTR_TO_PACKET;
5610                 break;
5611         case offsetof(struct xdp_md, data_meta):
5612                 info->reg_type = PTR_TO_PACKET_META;
5613                 break;
5614         case offsetof(struct xdp_md, data_end):
5615                 info->reg_type = PTR_TO_PACKET_END;
5616                 break;
5617         }
5618
5619         return __is_valid_xdp_access(off, size);
5620 }
5621
5622 void bpf_warn_invalid_xdp_action(u32 act)
5623 {
5624         const u32 act_max = XDP_REDIRECT;
5625
5626         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5627                   act > act_max ? "Illegal" : "Driver unsupported",
5628                   act);
5629 }
5630 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5631
5632 static bool sock_addr_is_valid_access(int off, int size,
5633                                       enum bpf_access_type type,
5634                                       const struct bpf_prog *prog,
5635                                       struct bpf_insn_access_aux *info)
5636 {
5637         const int size_default = sizeof(__u32);
5638
5639         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5640                 return false;
5641         if (off % size != 0)
5642                 return false;
5643
5644         /* Disallow access to IPv6 fields from IPv4 contex and vise
5645          * versa.
5646          */
5647         switch (off) {
5648         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5649                 switch (prog->expected_attach_type) {
5650                 case BPF_CGROUP_INET4_BIND:
5651                 case BPF_CGROUP_INET4_CONNECT:
5652                 case BPF_CGROUP_UDP4_SENDMSG:
5653                         break;
5654                 default:
5655                         return false;
5656                 }
5657                 break;
5658         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5659                 switch (prog->expected_attach_type) {
5660                 case BPF_CGROUP_INET6_BIND:
5661                 case BPF_CGROUP_INET6_CONNECT:
5662                 case BPF_CGROUP_UDP6_SENDMSG:
5663                         break;
5664                 default:
5665                         return false;
5666                 }
5667                 break;
5668         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5669                 switch (prog->expected_attach_type) {
5670                 case BPF_CGROUP_UDP4_SENDMSG:
5671                         break;
5672                 default:
5673                         return false;
5674                 }
5675                 break;
5676         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5677                                 msg_src_ip6[3]):
5678                 switch (prog->expected_attach_type) {
5679                 case BPF_CGROUP_UDP6_SENDMSG:
5680                         break;
5681                 default:
5682                         return false;
5683                 }
5684                 break;
5685         }
5686
5687         switch (off) {
5688         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5689         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5690         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5691         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5692                                 msg_src_ip6[3]):
5693                 /* Only narrow read access allowed for now. */
5694                 if (type == BPF_READ) {
5695                         bpf_ctx_record_field_size(info, size_default);
5696                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5697                                 return false;
5698                 } else {
5699                         if (size != size_default)
5700                                 return false;
5701                 }
5702                 break;
5703         case bpf_ctx_range(struct bpf_sock_addr, user_port):
5704                 if (size != size_default)
5705                         return false;
5706                 break;
5707         default:
5708                 if (type == BPF_READ) {
5709                         if (size != size_default)
5710                                 return false;
5711                 } else {
5712                         return false;
5713                 }
5714         }
5715
5716         return true;
5717 }
5718
5719 static bool sock_ops_is_valid_access(int off, int size,
5720                                      enum bpf_access_type type,
5721                                      const struct bpf_prog *prog,
5722                                      struct bpf_insn_access_aux *info)
5723 {
5724         const int size_default = sizeof(__u32);
5725
5726         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5727                 return false;
5728
5729         /* The verifier guarantees that size > 0. */
5730         if (off % size != 0)
5731                 return false;
5732
5733         if (type == BPF_WRITE) {
5734                 switch (off) {
5735                 case offsetof(struct bpf_sock_ops, reply):
5736                 case offsetof(struct bpf_sock_ops, sk_txhash):
5737                         if (size != size_default)
5738                                 return false;
5739                         break;
5740                 default:
5741                         return false;
5742                 }
5743         } else {
5744                 switch (off) {
5745                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5746                                         bytes_acked):
5747                         if (size != sizeof(__u64))
5748                                 return false;
5749                         break;
5750                 default:
5751                         if (size != size_default)
5752                                 return false;
5753                         break;
5754                 }
5755         }
5756
5757         return true;
5758 }
5759
5760 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5761                            const struct bpf_prog *prog)
5762 {
5763         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5764 }
5765
5766 static bool sk_skb_is_valid_access(int off, int size,
5767                                    enum bpf_access_type type,
5768                                    const struct bpf_prog *prog,
5769                                    struct bpf_insn_access_aux *info)
5770 {
5771         switch (off) {
5772         case bpf_ctx_range(struct __sk_buff, tc_classid):
5773         case bpf_ctx_range(struct __sk_buff, data_meta):
5774         case bpf_ctx_range(struct __sk_buff, flow_keys):
5775                 return false;
5776         }
5777
5778         if (type == BPF_WRITE) {
5779                 switch (off) {
5780                 case bpf_ctx_range(struct __sk_buff, tc_index):
5781                 case bpf_ctx_range(struct __sk_buff, priority):
5782                         break;
5783                 default:
5784                         return false;
5785                 }
5786         }
5787
5788         switch (off) {
5789         case bpf_ctx_range(struct __sk_buff, mark):
5790                 return false;
5791         case bpf_ctx_range(struct __sk_buff, data):
5792                 info->reg_type = PTR_TO_PACKET;
5793                 break;
5794         case bpf_ctx_range(struct __sk_buff, data_end):
5795                 info->reg_type = PTR_TO_PACKET_END;
5796                 break;
5797         }
5798
5799         return bpf_skb_is_valid_access(off, size, type, prog, info);
5800 }
5801
5802 static bool sk_msg_is_valid_access(int off, int size,
5803                                    enum bpf_access_type type,
5804                                    const struct bpf_prog *prog,
5805                                    struct bpf_insn_access_aux *info)
5806 {
5807         if (type == BPF_WRITE)
5808                 return false;
5809
5810         switch (off) {
5811         case offsetof(struct sk_msg_md, data):
5812                 info->reg_type = PTR_TO_PACKET;
5813                 if (size != sizeof(__u64))
5814                         return false;
5815                 break;
5816         case offsetof(struct sk_msg_md, data_end):
5817                 info->reg_type = PTR_TO_PACKET_END;
5818                 if (size != sizeof(__u64))
5819                         return false;
5820                 break;
5821         default:
5822                 if (size != sizeof(__u32))
5823                         return false;
5824         }
5825
5826         if (off < 0 || off >= sizeof(struct sk_msg_md))
5827                 return false;
5828         if (off % size != 0)
5829                 return false;
5830
5831         return true;
5832 }
5833
5834 static bool flow_dissector_is_valid_access(int off, int size,
5835                                            enum bpf_access_type type,
5836                                            const struct bpf_prog *prog,
5837                                            struct bpf_insn_access_aux *info)
5838 {
5839         if (type == BPF_WRITE) {
5840                 switch (off) {
5841                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5842                         break;
5843                 default:
5844                         return false;
5845                 }
5846         }
5847
5848         switch (off) {
5849         case bpf_ctx_range(struct __sk_buff, data):
5850                 info->reg_type = PTR_TO_PACKET;
5851                 break;
5852         case bpf_ctx_range(struct __sk_buff, data_end):
5853                 info->reg_type = PTR_TO_PACKET_END;
5854                 break;
5855         case bpf_ctx_range(struct __sk_buff, flow_keys):
5856                 info->reg_type = PTR_TO_FLOW_KEYS;
5857                 break;
5858         case bpf_ctx_range(struct __sk_buff, tc_classid):
5859         case bpf_ctx_range(struct __sk_buff, data_meta):
5860         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5861                 return false;
5862         }
5863
5864         return bpf_skb_is_valid_access(off, size, type, prog, info);
5865 }
5866
5867 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5868                                   const struct bpf_insn *si,
5869                                   struct bpf_insn *insn_buf,
5870                                   struct bpf_prog *prog, u32 *target_size)
5871 {
5872         struct bpf_insn *insn = insn_buf;
5873         int off;
5874
5875         switch (si->off) {
5876         case offsetof(struct __sk_buff, len):
5877                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5878                                       bpf_target_off(struct sk_buff, len, 4,
5879                                                      target_size));
5880                 break;
5881
5882         case offsetof(struct __sk_buff, protocol):
5883                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5884                                       bpf_target_off(struct sk_buff, protocol, 2,
5885                                                      target_size));
5886                 break;
5887
5888         case offsetof(struct __sk_buff, vlan_proto):
5889                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5890                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
5891                                                      target_size));
5892                 break;
5893
5894         case offsetof(struct __sk_buff, priority):
5895                 if (type == BPF_WRITE)
5896                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5897                                               bpf_target_off(struct sk_buff, priority, 4,
5898                                                              target_size));
5899                 else
5900                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5901                                               bpf_target_off(struct sk_buff, priority, 4,
5902                                                              target_size));
5903                 break;
5904
5905         case offsetof(struct __sk_buff, ingress_ifindex):
5906                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5907                                       bpf_target_off(struct sk_buff, skb_iif, 4,
5908                                                      target_size));
5909                 break;
5910
5911         case offsetof(struct __sk_buff, ifindex):
5912                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5913                                       si->dst_reg, si->src_reg,
5914                                       offsetof(struct sk_buff, dev));
5915                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5916                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5917                                       bpf_target_off(struct net_device, ifindex, 4,
5918                                                      target_size));
5919                 break;
5920
5921         case offsetof(struct __sk_buff, hash):
5922                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5923                                       bpf_target_off(struct sk_buff, hash, 4,
5924                                                      target_size));
5925                 break;
5926
5927         case offsetof(struct __sk_buff, mark):
5928                 if (type == BPF_WRITE)
5929                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5930                                               bpf_target_off(struct sk_buff, mark, 4,
5931                                                              target_size));
5932                 else
5933                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5934                                               bpf_target_off(struct sk_buff, mark, 4,
5935                                                              target_size));
5936                 break;
5937
5938         case offsetof(struct __sk_buff, pkt_type):
5939                 *target_size = 1;
5940                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5941                                       PKT_TYPE_OFFSET());
5942                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5943 #ifdef __BIG_ENDIAN_BITFIELD
5944                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5945 #endif
5946                 break;
5947
5948         case offsetof(struct __sk_buff, queue_mapping):
5949                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5950                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
5951                                                      target_size));
5952                 break;
5953
5954         case offsetof(struct __sk_buff, vlan_present):
5955         case offsetof(struct __sk_buff, vlan_tci):
5956                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5957
5958                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5959                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
5960                                                      target_size));
5961                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5962                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5963                                                 ~VLAN_TAG_PRESENT);
5964                 } else {
5965                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5966                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5967                 }
5968                 break;
5969
5970         case offsetof(struct __sk_buff, cb[0]) ...
5971              offsetofend(struct __sk_buff, cb[4]) - 1:
5972                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5973                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5974                               offsetof(struct qdisc_skb_cb, data)) %
5975                              sizeof(__u64));
5976
5977                 prog->cb_access = 1;
5978                 off  = si->off;
5979                 off -= offsetof(struct __sk_buff, cb[0]);
5980                 off += offsetof(struct sk_buff, cb);
5981                 off += offsetof(struct qdisc_skb_cb, data);
5982                 if (type == BPF_WRITE)
5983                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5984                                               si->src_reg, off);
5985                 else
5986                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5987                                               si->src_reg, off);
5988                 break;
5989
5990         case offsetof(struct __sk_buff, tc_classid):
5991                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5992
5993                 off  = si->off;
5994                 off -= offsetof(struct __sk_buff, tc_classid);
5995                 off += offsetof(struct sk_buff, cb);
5996                 off += offsetof(struct qdisc_skb_cb, tc_classid);
5997                 *target_size = 2;
5998                 if (type == BPF_WRITE)
5999                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6000                                               si->src_reg, off);
6001                 else
6002                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6003                                               si->src_reg, off);
6004                 break;
6005
6006         case offsetof(struct __sk_buff, data):
6007                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6008                                       si->dst_reg, si->src_reg,
6009                                       offsetof(struct sk_buff, data));
6010                 break;
6011
6012         case offsetof(struct __sk_buff, data_meta):
6013                 off  = si->off;
6014                 off -= offsetof(struct __sk_buff, data_meta);
6015                 off += offsetof(struct sk_buff, cb);
6016                 off += offsetof(struct bpf_skb_data_end, data_meta);
6017                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6018                                       si->src_reg, off);
6019                 break;
6020
6021         case offsetof(struct __sk_buff, data_end):
6022                 off  = si->off;
6023                 off -= offsetof(struct __sk_buff, data_end);
6024                 off += offsetof(struct sk_buff, cb);
6025                 off += offsetof(struct bpf_skb_data_end, data_end);
6026                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6027                                       si->src_reg, off);
6028                 break;
6029
6030         case offsetof(struct __sk_buff, tc_index):
6031 #ifdef CONFIG_NET_SCHED
6032                 if (type == BPF_WRITE)
6033                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6034                                               bpf_target_off(struct sk_buff, tc_index, 2,
6035                                                              target_size));
6036                 else
6037                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6038                                               bpf_target_off(struct sk_buff, tc_index, 2,
6039                                                              target_size));
6040 #else
6041                 *target_size = 2;
6042                 if (type == BPF_WRITE)
6043                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6044                 else
6045                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6046 #endif
6047                 break;
6048
6049         case offsetof(struct __sk_buff, napi_id):
6050 #if defined(CONFIG_NET_RX_BUSY_POLL)
6051                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6052                                       bpf_target_off(struct sk_buff, napi_id, 4,
6053                                                      target_size));
6054                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6055                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6056 #else
6057                 *target_size = 4;
6058                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6059 #endif
6060                 break;
6061         case offsetof(struct __sk_buff, family):
6062                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6063
6064                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6065                                       si->dst_reg, si->src_reg,
6066                                       offsetof(struct sk_buff, sk));
6067                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6068                                       bpf_target_off(struct sock_common,
6069                                                      skc_family,
6070                                                      2, target_size));
6071                 break;
6072         case offsetof(struct __sk_buff, remote_ip4):
6073                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6074
6075                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6076                                       si->dst_reg, si->src_reg,
6077                                       offsetof(struct sk_buff, sk));
6078                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6079                                       bpf_target_off(struct sock_common,
6080                                                      skc_daddr,
6081                                                      4, target_size));
6082                 break;
6083         case offsetof(struct __sk_buff, local_ip4):
6084                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6085                                           skc_rcv_saddr) != 4);
6086
6087                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6088                                       si->dst_reg, si->src_reg,
6089                                       offsetof(struct sk_buff, sk));
6090                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6091                                       bpf_target_off(struct sock_common,
6092                                                      skc_rcv_saddr,
6093                                                      4, target_size));
6094                 break;
6095         case offsetof(struct __sk_buff, remote_ip6[0]) ...
6096              offsetof(struct __sk_buff, remote_ip6[3]):
6097 #if IS_ENABLED(CONFIG_IPV6)
6098                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6099                                           skc_v6_daddr.s6_addr32[0]) != 4);
6100
6101                 off = si->off;
6102                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6103
6104                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6105                                       si->dst_reg, si->src_reg,
6106                                       offsetof(struct sk_buff, sk));
6107                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6108                                       offsetof(struct sock_common,
6109                                                skc_v6_daddr.s6_addr32[0]) +
6110                                       off);
6111 #else
6112                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6113 #endif
6114                 break;
6115         case offsetof(struct __sk_buff, local_ip6[0]) ...
6116              offsetof(struct __sk_buff, local_ip6[3]):
6117 #if IS_ENABLED(CONFIG_IPV6)
6118                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6119                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6120
6121                 off = si->off;
6122                 off -= offsetof(struct __sk_buff, local_ip6[0]);
6123
6124                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6125                                       si->dst_reg, si->src_reg,
6126                                       offsetof(struct sk_buff, sk));
6127                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6128                                       offsetof(struct sock_common,
6129                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6130                                       off);
6131 #else
6132                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6133 #endif
6134                 break;
6135
6136         case offsetof(struct __sk_buff, remote_port):
6137                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6138
6139                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6140                                       si->dst_reg, si->src_reg,
6141                                       offsetof(struct sk_buff, sk));
6142                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6143                                       bpf_target_off(struct sock_common,
6144                                                      skc_dport,
6145                                                      2, target_size));
6146 #ifndef __BIG_ENDIAN_BITFIELD
6147                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6148 #endif
6149                 break;
6150
6151         case offsetof(struct __sk_buff, local_port):
6152                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6153
6154                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6155                                       si->dst_reg, si->src_reg,
6156                                       offsetof(struct sk_buff, sk));
6157                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6158                                       bpf_target_off(struct sock_common,
6159                                                      skc_num, 2, target_size));
6160                 break;
6161
6162         case offsetof(struct __sk_buff, flow_keys):
6163                 off  = si->off;
6164                 off -= offsetof(struct __sk_buff, flow_keys);
6165                 off += offsetof(struct sk_buff, cb);
6166                 off += offsetof(struct qdisc_skb_cb, flow_keys);
6167                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6168                                       si->src_reg, off);
6169                 break;
6170         }
6171
6172         return insn - insn_buf;
6173 }
6174
6175 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
6176                                 const struct bpf_insn *si,
6177                                 struct bpf_insn *insn_buf,
6178                                 struct bpf_prog *prog, u32 *target_size)
6179 {
6180         struct bpf_insn *insn = insn_buf;
6181         int off;
6182
6183         switch (si->off) {
6184         case offsetof(struct bpf_sock, bound_dev_if):
6185                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6186
6187                 if (type == BPF_WRITE)
6188                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6189                                         offsetof(struct sock, sk_bound_dev_if));
6190                 else
6191                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6192                                       offsetof(struct sock, sk_bound_dev_if));
6193                 break;
6194
6195         case offsetof(struct bpf_sock, mark):
6196                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6197
6198                 if (type == BPF_WRITE)
6199                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6200                                         offsetof(struct sock, sk_mark));
6201                 else
6202                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6203                                       offsetof(struct sock, sk_mark));
6204                 break;
6205
6206         case offsetof(struct bpf_sock, priority):
6207                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6208
6209                 if (type == BPF_WRITE)
6210                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6211                                         offsetof(struct sock, sk_priority));
6212                 else
6213                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6214                                       offsetof(struct sock, sk_priority));
6215                 break;
6216
6217         case offsetof(struct bpf_sock, family):
6218                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6219
6220                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6221                                       offsetof(struct sock, sk_family));
6222                 break;
6223
6224         case offsetof(struct bpf_sock, type):
6225                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6226                                       offsetof(struct sock, __sk_flags_offset));
6227                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6228                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6229                 break;
6230
6231         case offsetof(struct bpf_sock, protocol):
6232                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6233                                       offsetof(struct sock, __sk_flags_offset));
6234                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6235                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6236                 break;
6237
6238         case offsetof(struct bpf_sock, src_ip4):
6239                 *insn++ = BPF_LDX_MEM(
6240                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6241                         bpf_target_off(struct sock_common, skc_rcv_saddr,
6242                                        FIELD_SIZEOF(struct sock_common,
6243                                                     skc_rcv_saddr),
6244                                        target_size));
6245                 break;
6246
6247         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6248 #if IS_ENABLED(CONFIG_IPV6)
6249                 off = si->off;
6250                 off -= offsetof(struct bpf_sock, src_ip6[0]);
6251                 *insn++ = BPF_LDX_MEM(
6252                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6253                         bpf_target_off(
6254                                 struct sock_common,
6255                                 skc_v6_rcv_saddr.s6_addr32[0],
6256                                 FIELD_SIZEOF(struct sock_common,
6257                                              skc_v6_rcv_saddr.s6_addr32[0]),
6258                                 target_size) + off);
6259 #else
6260                 (void)off;
6261                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6262 #endif
6263                 break;
6264
6265         case offsetof(struct bpf_sock, src_port):
6266                 *insn++ = BPF_LDX_MEM(
6267                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6268                         si->dst_reg, si->src_reg,
6269                         bpf_target_off(struct sock_common, skc_num,
6270                                        FIELD_SIZEOF(struct sock_common,
6271                                                     skc_num),
6272                                        target_size));
6273                 break;
6274         }
6275
6276         return insn - insn_buf;
6277 }
6278
6279 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6280                                          const struct bpf_insn *si,
6281                                          struct bpf_insn *insn_buf,
6282                                          struct bpf_prog *prog, u32 *target_size)
6283 {
6284         struct bpf_insn *insn = insn_buf;
6285
6286         switch (si->off) {
6287         case offsetof(struct __sk_buff, ifindex):
6288                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6289                                       si->dst_reg, si->src_reg,
6290                                       offsetof(struct sk_buff, dev));
6291                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6292                                       bpf_target_off(struct net_device, ifindex, 4,
6293                                                      target_size));
6294                 break;
6295         default:
6296                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6297                                               target_size);
6298         }
6299
6300         return insn - insn_buf;
6301 }
6302
6303 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6304                                   const struct bpf_insn *si,
6305                                   struct bpf_insn *insn_buf,
6306                                   struct bpf_prog *prog, u32 *target_size)
6307 {
6308         struct bpf_insn *insn = insn_buf;
6309
6310         switch (si->off) {
6311         case offsetof(struct xdp_md, data):
6312                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6313                                       si->dst_reg, si->src_reg,
6314                                       offsetof(struct xdp_buff, data));
6315                 break;
6316         case offsetof(struct xdp_md, data_meta):
6317                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6318                                       si->dst_reg, si->src_reg,
6319                                       offsetof(struct xdp_buff, data_meta));
6320                 break;
6321         case offsetof(struct xdp_md, data_end):
6322                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6323                                       si->dst_reg, si->src_reg,
6324                                       offsetof(struct xdp_buff, data_end));
6325                 break;
6326         case offsetof(struct xdp_md, ingress_ifindex):
6327                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6328                                       si->dst_reg, si->src_reg,
6329                                       offsetof(struct xdp_buff, rxq));
6330                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6331                                       si->dst_reg, si->dst_reg,
6332                                       offsetof(struct xdp_rxq_info, dev));
6333                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6334                                       offsetof(struct net_device, ifindex));
6335                 break;
6336         case offsetof(struct xdp_md, rx_queue_index):
6337                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6338                                       si->dst_reg, si->src_reg,
6339                                       offsetof(struct xdp_buff, rxq));
6340                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6341                                       offsetof(struct xdp_rxq_info,
6342                                                queue_index));
6343                 break;
6344         }
6345
6346         return insn - insn_buf;
6347 }
6348
6349 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6350  * context Structure, F is Field in context structure that contains a pointer
6351  * to Nested Structure of type NS that has the field NF.
6352  *
6353  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6354  * sure that SIZE is not greater than actual size of S.F.NF.
6355  *
6356  * If offset OFF is provided, the load happens from that offset relative to
6357  * offset of NF.
6358  */
6359 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
6360         do {                                                                   \
6361                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
6362                                       si->src_reg, offsetof(S, F));            \
6363                 *insn++ = BPF_LDX_MEM(                                         \
6364                         SIZE, si->dst_reg, si->dst_reg,                        \
6365                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6366                                        target_size)                            \
6367                                 + OFF);                                        \
6368         } while (0)
6369
6370 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
6371         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
6372                                              BPF_FIELD_SIZEOF(NS, NF), 0)
6373
6374 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6375  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6376  *
6377  * It doesn't support SIZE argument though since narrow stores are not
6378  * supported for now.
6379  *
6380  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6381  * "register" since two registers available in convert_ctx_access are not
6382  * enough: we can't override neither SRC, since it contains value to store, nor
6383  * DST since it contains pointer to context that may be used by later
6384  * instructions. But we need a temporary place to save pointer to nested
6385  * structure whose field we want to store to.
6386  */
6387 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
6388         do {                                                                   \
6389                 int tmp_reg = BPF_REG_9;                                       \
6390                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6391                         --tmp_reg;                                             \
6392                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6393                         --tmp_reg;                                             \
6394                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
6395                                       offsetof(S, TF));                        \
6396                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
6397                                       si->dst_reg, offsetof(S, F));            \
6398                 *insn++ = BPF_STX_MEM(                                         \
6399                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
6400                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6401                                        target_size)                            \
6402                                 + OFF);                                        \
6403                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
6404                                       offsetof(S, TF));                        \
6405         } while (0)
6406
6407 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6408                                                       TF)                      \
6409         do {                                                                   \
6410                 if (type == BPF_WRITE) {                                       \
6411                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
6412                                                          TF);                  \
6413                 } else {                                                       \
6414                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
6415                                 S, NS, F, NF, SIZE, OFF);  \
6416                 }                                                              \
6417         } while (0)
6418
6419 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
6420         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
6421                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6422
6423 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6424                                         const struct bpf_insn *si,
6425                                         struct bpf_insn *insn_buf,
6426                                         struct bpf_prog *prog, u32 *target_size)
6427 {
6428         struct bpf_insn *insn = insn_buf;
6429         int off;
6430
6431         switch (si->off) {
6432         case offsetof(struct bpf_sock_addr, user_family):
6433                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6434                                             struct sockaddr, uaddr, sa_family);
6435                 break;
6436
6437         case offsetof(struct bpf_sock_addr, user_ip4):
6438                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6439                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6440                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6441                 break;
6442
6443         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6444                 off = si->off;
6445                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6446                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6447                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6448                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6449                         tmp_reg);
6450                 break;
6451
6452         case offsetof(struct bpf_sock_addr, user_port):
6453                 /* To get port we need to know sa_family first and then treat
6454                  * sockaddr as either sockaddr_in or sockaddr_in6.
6455                  * Though we can simplify since port field has same offset and
6456                  * size in both structures.
6457                  * Here we check this invariant and use just one of the
6458                  * structures if it's true.
6459                  */
6460                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6461                              offsetof(struct sockaddr_in6, sin6_port));
6462                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6463                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6464                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6465                                                      struct sockaddr_in6, uaddr,
6466                                                      sin6_port, tmp_reg);
6467                 break;
6468
6469         case offsetof(struct bpf_sock_addr, family):
6470                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6471                                             struct sock, sk, sk_family);
6472                 break;
6473
6474         case offsetof(struct bpf_sock_addr, type):
6475                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6476                         struct bpf_sock_addr_kern, struct sock, sk,
6477                         __sk_flags_offset, BPF_W, 0);
6478                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6479                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6480                 break;
6481
6482         case offsetof(struct bpf_sock_addr, protocol):
6483                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6484                         struct bpf_sock_addr_kern, struct sock, sk,
6485                         __sk_flags_offset, BPF_W, 0);
6486                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6487                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6488                                         SK_FL_PROTO_SHIFT);
6489                 break;
6490
6491         case offsetof(struct bpf_sock_addr, msg_src_ip4):
6492                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6493                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6494                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6495                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6496                 break;
6497
6498         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6499                                 msg_src_ip6[3]):
6500                 off = si->off;
6501                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6502                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6503                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6504                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6505                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6506                 break;
6507         }
6508
6509         return insn - insn_buf;
6510 }
6511
6512 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6513                                        const struct bpf_insn *si,
6514                                        struct bpf_insn *insn_buf,
6515                                        struct bpf_prog *prog,
6516                                        u32 *target_size)
6517 {
6518         struct bpf_insn *insn = insn_buf;
6519         int off;
6520
6521         switch (si->off) {
6522         case offsetof(struct bpf_sock_ops, op) ...
6523              offsetof(struct bpf_sock_ops, replylong[3]):
6524                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6525                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6526                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6527                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6528                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6529                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6530                 off = si->off;
6531                 off -= offsetof(struct bpf_sock_ops, op);
6532                 off += offsetof(struct bpf_sock_ops_kern, op);
6533                 if (type == BPF_WRITE)
6534                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6535                                               off);
6536                 else
6537                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6538                                               off);
6539                 break;
6540
6541         case offsetof(struct bpf_sock_ops, family):
6542                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6543
6544                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6545                                               struct bpf_sock_ops_kern, sk),
6546                                       si->dst_reg, si->src_reg,
6547                                       offsetof(struct bpf_sock_ops_kern, sk));
6548                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6549                                       offsetof(struct sock_common, skc_family));
6550                 break;
6551
6552         case offsetof(struct bpf_sock_ops, remote_ip4):
6553                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6554
6555                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6556                                                 struct bpf_sock_ops_kern, sk),
6557                                       si->dst_reg, si->src_reg,
6558                                       offsetof(struct bpf_sock_ops_kern, sk));
6559                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6560                                       offsetof(struct sock_common, skc_daddr));
6561                 break;
6562
6563         case offsetof(struct bpf_sock_ops, local_ip4):
6564                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6565                                           skc_rcv_saddr) != 4);
6566
6567                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6568                                               struct bpf_sock_ops_kern, sk),
6569                                       si->dst_reg, si->src_reg,
6570                                       offsetof(struct bpf_sock_ops_kern, sk));
6571                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6572                                       offsetof(struct sock_common,
6573                                                skc_rcv_saddr));
6574                 break;
6575
6576         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6577              offsetof(struct bpf_sock_ops, remote_ip6[3]):
6578 #if IS_ENABLED(CONFIG_IPV6)
6579                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6580                                           skc_v6_daddr.s6_addr32[0]) != 4);
6581
6582                 off = si->off;
6583                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6585                                                 struct bpf_sock_ops_kern, sk),
6586                                       si->dst_reg, si->src_reg,
6587                                       offsetof(struct bpf_sock_ops_kern, sk));
6588                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6589                                       offsetof(struct sock_common,
6590                                                skc_v6_daddr.s6_addr32[0]) +
6591                                       off);
6592 #else
6593                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6594 #endif
6595                 break;
6596
6597         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6598              offsetof(struct bpf_sock_ops, local_ip6[3]):
6599 #if IS_ENABLED(CONFIG_IPV6)
6600                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6601                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6602
6603                 off = si->off;
6604                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6605                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6606                                                 struct bpf_sock_ops_kern, sk),
6607                                       si->dst_reg, si->src_reg,
6608                                       offsetof(struct bpf_sock_ops_kern, sk));
6609                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6610                                       offsetof(struct sock_common,
6611                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6612                                       off);
6613 #else
6614                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6615 #endif
6616                 break;
6617
6618         case offsetof(struct bpf_sock_ops, remote_port):
6619                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6620
6621                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6622                                                 struct bpf_sock_ops_kern, sk),
6623                                       si->dst_reg, si->src_reg,
6624                                       offsetof(struct bpf_sock_ops_kern, sk));
6625                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6626                                       offsetof(struct sock_common, skc_dport));
6627 #ifndef __BIG_ENDIAN_BITFIELD
6628                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6629 #endif
6630                 break;
6631
6632         case offsetof(struct bpf_sock_ops, local_port):
6633                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6634
6635                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6636                                                 struct bpf_sock_ops_kern, sk),
6637                                       si->dst_reg, si->src_reg,
6638                                       offsetof(struct bpf_sock_ops_kern, sk));
6639                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6640                                       offsetof(struct sock_common, skc_num));
6641                 break;
6642
6643         case offsetof(struct bpf_sock_ops, is_fullsock):
6644                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6645                                                 struct bpf_sock_ops_kern,
6646                                                 is_fullsock),
6647                                       si->dst_reg, si->src_reg,
6648                                       offsetof(struct bpf_sock_ops_kern,
6649                                                is_fullsock));
6650                 break;
6651
6652         case offsetof(struct bpf_sock_ops, state):
6653                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6654
6655                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6656                                                 struct bpf_sock_ops_kern, sk),
6657                                       si->dst_reg, si->src_reg,
6658                                       offsetof(struct bpf_sock_ops_kern, sk));
6659                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6660                                       offsetof(struct sock_common, skc_state));
6661                 break;
6662
6663         case offsetof(struct bpf_sock_ops, rtt_min):
6664                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6665                              sizeof(struct minmax));
6666                 BUILD_BUG_ON(sizeof(struct minmax) <
6667                              sizeof(struct minmax_sample));
6668
6669                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6670                                                 struct bpf_sock_ops_kern, sk),
6671                                       si->dst_reg, si->src_reg,
6672                                       offsetof(struct bpf_sock_ops_kern, sk));
6673                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6674                                       offsetof(struct tcp_sock, rtt_min) +
6675                                       FIELD_SIZEOF(struct minmax_sample, t));
6676                 break;
6677
6678 /* Helper macro for adding read access to tcp_sock or sock fields. */
6679 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6680         do {                                                                  \
6681                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6682                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6683                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6684                                                 struct bpf_sock_ops_kern,     \
6685                                                 is_fullsock),                 \
6686                                       si->dst_reg, si->src_reg,               \
6687                                       offsetof(struct bpf_sock_ops_kern,      \
6688                                                is_fullsock));                 \
6689                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
6690                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6691                                                 struct bpf_sock_ops_kern, sk),\
6692                                       si->dst_reg, si->src_reg,               \
6693                                       offsetof(struct bpf_sock_ops_kern, sk));\
6694                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
6695                                                        OBJ_FIELD),            \
6696                                       si->dst_reg, si->dst_reg,               \
6697                                       offsetof(OBJ, OBJ_FIELD));              \
6698         } while (0)
6699
6700 /* Helper macro for adding write access to tcp_sock or sock fields.
6701  * The macro is called with two registers, dst_reg which contains a pointer
6702  * to ctx (context) and src_reg which contains the value that should be
6703  * stored. However, we need an additional register since we cannot overwrite
6704  * dst_reg because it may be used later in the program.
6705  * Instead we "borrow" one of the other register. We first save its value
6706  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6707  * it at the end of the macro.
6708  */
6709 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6710         do {                                                                  \
6711                 int reg = BPF_REG_9;                                          \
6712                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6713                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6714                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6715                         reg--;                                                \
6716                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6717                         reg--;                                                \
6718                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
6719                                       offsetof(struct bpf_sock_ops_kern,      \
6720                                                temp));                        \
6721                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6722                                                 struct bpf_sock_ops_kern,     \
6723                                                 is_fullsock),                 \
6724                                       reg, si->dst_reg,                       \
6725                                       offsetof(struct bpf_sock_ops_kern,      \
6726                                                is_fullsock));                 \
6727                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
6728                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6729                                                 struct bpf_sock_ops_kern, sk),\
6730                                       reg, si->dst_reg,                       \
6731                                       offsetof(struct bpf_sock_ops_kern, sk));\
6732                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
6733                                       reg, si->src_reg,                       \
6734                                       offsetof(OBJ, OBJ_FIELD));              \
6735                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
6736                                       offsetof(struct bpf_sock_ops_kern,      \
6737                                                temp));                        \
6738         } while (0)
6739
6740 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
6741         do {                                                                  \
6742                 if (TYPE == BPF_WRITE)                                        \
6743                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6744                 else                                                          \
6745                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6746         } while (0)
6747
6748         case offsetof(struct bpf_sock_ops, snd_cwnd):
6749                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6750                 break;
6751
6752         case offsetof(struct bpf_sock_ops, srtt_us):
6753                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6754                 break;
6755
6756         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6757                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6758                                    struct tcp_sock);
6759                 break;
6760
6761         case offsetof(struct bpf_sock_ops, snd_ssthresh):
6762                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6763                 break;
6764
6765         case offsetof(struct bpf_sock_ops, rcv_nxt):
6766                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6767                 break;
6768
6769         case offsetof(struct bpf_sock_ops, snd_nxt):
6770                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6771                 break;
6772
6773         case offsetof(struct bpf_sock_ops, snd_una):
6774                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6775                 break;
6776
6777         case offsetof(struct bpf_sock_ops, mss_cache):
6778                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6779                 break;
6780
6781         case offsetof(struct bpf_sock_ops, ecn_flags):
6782                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6783                 break;
6784
6785         case offsetof(struct bpf_sock_ops, rate_delivered):
6786                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6787                                    struct tcp_sock);
6788                 break;
6789
6790         case offsetof(struct bpf_sock_ops, rate_interval_us):
6791                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6792                                    struct tcp_sock);
6793                 break;
6794
6795         case offsetof(struct bpf_sock_ops, packets_out):
6796                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6797                 break;
6798
6799         case offsetof(struct bpf_sock_ops, retrans_out):
6800                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6801                 break;
6802
6803         case offsetof(struct bpf_sock_ops, total_retrans):
6804                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6805                                    struct tcp_sock);
6806                 break;
6807
6808         case offsetof(struct bpf_sock_ops, segs_in):
6809                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6810                 break;
6811
6812         case offsetof(struct bpf_sock_ops, data_segs_in):
6813                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6814                 break;
6815
6816         case offsetof(struct bpf_sock_ops, segs_out):
6817                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6818                 break;
6819
6820         case offsetof(struct bpf_sock_ops, data_segs_out):
6821                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6822                                    struct tcp_sock);
6823                 break;
6824
6825         case offsetof(struct bpf_sock_ops, lost_out):
6826                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6827                 break;
6828
6829         case offsetof(struct bpf_sock_ops, sacked_out):
6830                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6831                 break;
6832
6833         case offsetof(struct bpf_sock_ops, sk_txhash):
6834                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6835                                           struct sock, type);
6836                 break;
6837
6838         case offsetof(struct bpf_sock_ops, bytes_received):
6839                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6840                                    struct tcp_sock);
6841                 break;
6842
6843         case offsetof(struct bpf_sock_ops, bytes_acked):
6844                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6845                 break;
6846
6847         }
6848         return insn - insn_buf;
6849 }
6850
6851 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6852                                      const struct bpf_insn *si,
6853                                      struct bpf_insn *insn_buf,
6854                                      struct bpf_prog *prog, u32 *target_size)
6855 {
6856         struct bpf_insn *insn = insn_buf;
6857         int off;
6858
6859         switch (si->off) {
6860         case offsetof(struct __sk_buff, data_end):
6861                 off  = si->off;
6862                 off -= offsetof(struct __sk_buff, data_end);
6863                 off += offsetof(struct sk_buff, cb);
6864                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6865                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6866                                       si->src_reg, off);
6867                 break;
6868         default:
6869                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6870                                               target_size);
6871         }
6872
6873         return insn - insn_buf;
6874 }
6875
6876 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6877                                      const struct bpf_insn *si,
6878                                      struct bpf_insn *insn_buf,
6879                                      struct bpf_prog *prog, u32 *target_size)
6880 {
6881         struct bpf_insn *insn = insn_buf;
6882 #if IS_ENABLED(CONFIG_IPV6)
6883         int off;
6884 #endif
6885
6886         switch (si->off) {
6887         case offsetof(struct sk_msg_md, data):
6888                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
6889                                       si->dst_reg, si->src_reg,
6890                                       offsetof(struct sk_msg, data));
6891                 break;
6892         case offsetof(struct sk_msg_md, data_end):
6893                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
6894                                       si->dst_reg, si->src_reg,
6895                                       offsetof(struct sk_msg, data_end));
6896                 break;
6897         case offsetof(struct sk_msg_md, family):
6898                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6899
6900                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6901                                               struct sk_msg, sk),
6902                                       si->dst_reg, si->src_reg,
6903                                       offsetof(struct sk_msg, sk));
6904                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6905                                       offsetof(struct sock_common, skc_family));
6906                 break;
6907
6908         case offsetof(struct sk_msg_md, remote_ip4):
6909                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6910
6911                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6912                                                 struct sk_msg, sk),
6913                                       si->dst_reg, si->src_reg,
6914                                       offsetof(struct sk_msg, sk));
6915                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6916                                       offsetof(struct sock_common, skc_daddr));
6917                 break;
6918
6919         case offsetof(struct sk_msg_md, local_ip4):
6920                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6921                                           skc_rcv_saddr) != 4);
6922
6923                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6924                                               struct sk_msg, sk),
6925                                       si->dst_reg, si->src_reg,
6926                                       offsetof(struct sk_msg, sk));
6927                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6928                                       offsetof(struct sock_common,
6929                                                skc_rcv_saddr));
6930                 break;
6931
6932         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6933              offsetof(struct sk_msg_md, remote_ip6[3]):
6934 #if IS_ENABLED(CONFIG_IPV6)
6935                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6936                                           skc_v6_daddr.s6_addr32[0]) != 4);
6937
6938                 off = si->off;
6939                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6940                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6941                                                 struct sk_msg, sk),
6942                                       si->dst_reg, si->src_reg,
6943                                       offsetof(struct sk_msg, sk));
6944                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6945                                       offsetof(struct sock_common,
6946                                                skc_v6_daddr.s6_addr32[0]) +
6947                                       off);
6948 #else
6949                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6950 #endif
6951                 break;
6952
6953         case offsetof(struct sk_msg_md, local_ip6[0]) ...
6954              offsetof(struct sk_msg_md, local_ip6[3]):
6955 #if IS_ENABLED(CONFIG_IPV6)
6956                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6957                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6958
6959                 off = si->off;
6960                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6961                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6962                                                 struct sk_msg, sk),
6963                                       si->dst_reg, si->src_reg,
6964                                       offsetof(struct sk_msg, sk));
6965                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6966                                       offsetof(struct sock_common,
6967                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6968                                       off);
6969 #else
6970                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6971 #endif
6972                 break;
6973
6974         case offsetof(struct sk_msg_md, remote_port):
6975                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6976
6977                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6978                                                 struct sk_msg, sk),
6979                                       si->dst_reg, si->src_reg,
6980                                       offsetof(struct sk_msg, sk));
6981                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6982                                       offsetof(struct sock_common, skc_dport));
6983 #ifndef __BIG_ENDIAN_BITFIELD
6984                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6985 #endif
6986                 break;
6987
6988         case offsetof(struct sk_msg_md, local_port):
6989                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6990
6991                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6992                                                 struct sk_msg, sk),
6993                                       si->dst_reg, si->src_reg,
6994                                       offsetof(struct sk_msg, sk));
6995                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6996                                       offsetof(struct sock_common, skc_num));
6997                 break;
6998         }
6999
7000         return insn - insn_buf;
7001 }
7002
7003 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7004         .get_func_proto         = sk_filter_func_proto,
7005         .is_valid_access        = sk_filter_is_valid_access,
7006         .convert_ctx_access     = bpf_convert_ctx_access,
7007         .gen_ld_abs             = bpf_gen_ld_abs,
7008 };
7009
7010 const struct bpf_prog_ops sk_filter_prog_ops = {
7011         .test_run               = bpf_prog_test_run_skb,
7012 };
7013
7014 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7015         .get_func_proto         = tc_cls_act_func_proto,
7016         .is_valid_access        = tc_cls_act_is_valid_access,
7017         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
7018         .gen_prologue           = tc_cls_act_prologue,
7019         .gen_ld_abs             = bpf_gen_ld_abs,
7020 };
7021
7022 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7023         .test_run               = bpf_prog_test_run_skb,
7024 };
7025
7026 const struct bpf_verifier_ops xdp_verifier_ops = {
7027         .get_func_proto         = xdp_func_proto,
7028         .is_valid_access        = xdp_is_valid_access,
7029         .convert_ctx_access     = xdp_convert_ctx_access,
7030 };
7031
7032 const struct bpf_prog_ops xdp_prog_ops = {
7033         .test_run               = bpf_prog_test_run_xdp,
7034 };
7035
7036 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7037         .get_func_proto         = cg_skb_func_proto,
7038         .is_valid_access        = sk_filter_is_valid_access,
7039         .convert_ctx_access     = bpf_convert_ctx_access,
7040 };
7041
7042 const struct bpf_prog_ops cg_skb_prog_ops = {
7043         .test_run               = bpf_prog_test_run_skb,
7044 };
7045
7046 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7047         .get_func_proto         = lwt_in_func_proto,
7048         .is_valid_access        = lwt_is_valid_access,
7049         .convert_ctx_access     = bpf_convert_ctx_access,
7050 };
7051
7052 const struct bpf_prog_ops lwt_in_prog_ops = {
7053         .test_run               = bpf_prog_test_run_skb,
7054 };
7055
7056 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7057         .get_func_proto         = lwt_out_func_proto,
7058         .is_valid_access        = lwt_is_valid_access,
7059         .convert_ctx_access     = bpf_convert_ctx_access,
7060 };
7061
7062 const struct bpf_prog_ops lwt_out_prog_ops = {
7063         .test_run               = bpf_prog_test_run_skb,
7064 };
7065
7066 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7067         .get_func_proto         = lwt_xmit_func_proto,
7068         .is_valid_access        = lwt_is_valid_access,
7069         .convert_ctx_access     = bpf_convert_ctx_access,
7070         .gen_prologue           = tc_cls_act_prologue,
7071 };
7072
7073 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7074         .test_run               = bpf_prog_test_run_skb,
7075 };
7076
7077 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7078         .get_func_proto         = lwt_seg6local_func_proto,
7079         .is_valid_access        = lwt_is_valid_access,
7080         .convert_ctx_access     = bpf_convert_ctx_access,
7081 };
7082
7083 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7084         .test_run               = bpf_prog_test_run_skb,
7085 };
7086
7087 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7088         .get_func_proto         = sock_filter_func_proto,
7089         .is_valid_access        = sock_filter_is_valid_access,
7090         .convert_ctx_access     = bpf_sock_convert_ctx_access,
7091 };
7092
7093 const struct bpf_prog_ops cg_sock_prog_ops = {
7094 };
7095
7096 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7097         .get_func_proto         = sock_addr_func_proto,
7098         .is_valid_access        = sock_addr_is_valid_access,
7099         .convert_ctx_access     = sock_addr_convert_ctx_access,
7100 };
7101
7102 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7103 };
7104
7105 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7106         .get_func_proto         = sock_ops_func_proto,
7107         .is_valid_access        = sock_ops_is_valid_access,
7108         .convert_ctx_access     = sock_ops_convert_ctx_access,
7109 };
7110
7111 const struct bpf_prog_ops sock_ops_prog_ops = {
7112 };
7113
7114 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7115         .get_func_proto         = sk_skb_func_proto,
7116         .is_valid_access        = sk_skb_is_valid_access,
7117         .convert_ctx_access     = sk_skb_convert_ctx_access,
7118         .gen_prologue           = sk_skb_prologue,
7119 };
7120
7121 const struct bpf_prog_ops sk_skb_prog_ops = {
7122 };
7123
7124 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7125         .get_func_proto         = sk_msg_func_proto,
7126         .is_valid_access        = sk_msg_is_valid_access,
7127         .convert_ctx_access     = sk_msg_convert_ctx_access,
7128 };
7129
7130 const struct bpf_prog_ops sk_msg_prog_ops = {
7131 };
7132
7133 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7134         .get_func_proto         = flow_dissector_func_proto,
7135         .is_valid_access        = flow_dissector_is_valid_access,
7136         .convert_ctx_access     = bpf_convert_ctx_access,
7137 };
7138
7139 const struct bpf_prog_ops flow_dissector_prog_ops = {
7140 };
7141
7142 int sk_detach_filter(struct sock *sk)
7143 {
7144         int ret = -ENOENT;
7145         struct sk_filter *filter;
7146
7147         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7148                 return -EPERM;
7149
7150         filter = rcu_dereference_protected(sk->sk_filter,
7151                                            lockdep_sock_is_held(sk));
7152         if (filter) {
7153                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7154                 sk_filter_uncharge(sk, filter);
7155                 ret = 0;
7156         }
7157
7158         return ret;
7159 }
7160 EXPORT_SYMBOL_GPL(sk_detach_filter);
7161
7162 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7163                   unsigned int len)
7164 {
7165         struct sock_fprog_kern *fprog;
7166         struct sk_filter *filter;
7167         int ret = 0;
7168
7169         lock_sock(sk);
7170         filter = rcu_dereference_protected(sk->sk_filter,
7171                                            lockdep_sock_is_held(sk));
7172         if (!filter)
7173                 goto out;
7174
7175         /* We're copying the filter that has been originally attached,
7176          * so no conversion/decode needed anymore. eBPF programs that
7177          * have no original program cannot be dumped through this.
7178          */
7179         ret = -EACCES;
7180         fprog = filter->prog->orig_prog;
7181         if (!fprog)
7182                 goto out;
7183
7184         ret = fprog->len;
7185         if (!len)
7186                 /* User space only enquires number of filter blocks. */
7187                 goto out;
7188
7189         ret = -EINVAL;
7190         if (len < fprog->len)
7191                 goto out;
7192
7193         ret = -EFAULT;
7194         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7195                 goto out;
7196
7197         /* Instead of bytes, the API requests to return the number
7198          * of filter blocks.
7199          */
7200         ret = fprog->len;
7201 out:
7202         release_sock(sk);
7203         return ret;
7204 }
7205
7206 #ifdef CONFIG_INET
7207 struct sk_reuseport_kern {
7208         struct sk_buff *skb;
7209         struct sock *sk;
7210         struct sock *selected_sk;
7211         void *data_end;
7212         u32 hash;
7213         u32 reuseport_id;
7214         bool bind_inany;
7215 };
7216
7217 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7218                                     struct sock_reuseport *reuse,
7219                                     struct sock *sk, struct sk_buff *skb,
7220                                     u32 hash)
7221 {
7222         reuse_kern->skb = skb;
7223         reuse_kern->sk = sk;
7224         reuse_kern->selected_sk = NULL;
7225         reuse_kern->data_end = skb->data + skb_headlen(skb);
7226         reuse_kern->hash = hash;
7227         reuse_kern->reuseport_id = reuse->reuseport_id;
7228         reuse_kern->bind_inany = reuse->bind_inany;
7229 }
7230
7231 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7232                                   struct bpf_prog *prog, struct sk_buff *skb,
7233                                   u32 hash)
7234 {
7235         struct sk_reuseport_kern reuse_kern;
7236         enum sk_action action;
7237
7238         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7239         action = BPF_PROG_RUN(prog, &reuse_kern);
7240
7241         if (action == SK_PASS)
7242                 return reuse_kern.selected_sk;
7243         else
7244                 return ERR_PTR(-ECONNREFUSED);
7245 }
7246
7247 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7248            struct bpf_map *, map, void *, key, u32, flags)
7249 {
7250         struct sock_reuseport *reuse;
7251         struct sock *selected_sk;
7252
7253         selected_sk = map->ops->map_lookup_elem(map, key);
7254         if (!selected_sk)
7255                 return -ENOENT;
7256
7257         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7258         if (!reuse)
7259                 /* selected_sk is unhashed (e.g. by close()) after the
7260                  * above map_lookup_elem().  Treat selected_sk has already
7261                  * been removed from the map.
7262                  */
7263                 return -ENOENT;
7264
7265         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7266                 struct sock *sk;
7267
7268                 if (unlikely(!reuse_kern->reuseport_id))
7269                         /* There is a small race between adding the
7270                          * sk to the map and setting the
7271                          * reuse_kern->reuseport_id.
7272                          * Treat it as the sk has not been added to
7273                          * the bpf map yet.
7274                          */
7275                         return -ENOENT;
7276
7277                 sk = reuse_kern->sk;
7278                 if (sk->sk_protocol != selected_sk->sk_protocol)
7279                         return -EPROTOTYPE;
7280                 else if (sk->sk_family != selected_sk->sk_family)
7281                         return -EAFNOSUPPORT;
7282
7283                 /* Catch all. Likely bound to a different sockaddr. */
7284                 return -EBADFD;
7285         }
7286
7287         reuse_kern->selected_sk = selected_sk;
7288
7289         return 0;
7290 }
7291
7292 static const struct bpf_func_proto sk_select_reuseport_proto = {
7293         .func           = sk_select_reuseport,
7294         .gpl_only       = false,
7295         .ret_type       = RET_INTEGER,
7296         .arg1_type      = ARG_PTR_TO_CTX,
7297         .arg2_type      = ARG_CONST_MAP_PTR,
7298         .arg3_type      = ARG_PTR_TO_MAP_KEY,
7299         .arg4_type      = ARG_ANYTHING,
7300 };
7301
7302 BPF_CALL_4(sk_reuseport_load_bytes,
7303            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7304            void *, to, u32, len)
7305 {
7306         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7307 }
7308
7309 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7310         .func           = sk_reuseport_load_bytes,
7311         .gpl_only       = false,
7312         .ret_type       = RET_INTEGER,
7313         .arg1_type      = ARG_PTR_TO_CTX,
7314         .arg2_type      = ARG_ANYTHING,
7315         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7316         .arg4_type      = ARG_CONST_SIZE,
7317 };
7318
7319 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7320            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7321            void *, to, u32, len, u32, start_header)
7322 {
7323         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7324                                                len, start_header);
7325 }
7326
7327 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7328         .func           = sk_reuseport_load_bytes_relative,
7329         .gpl_only       = false,
7330         .ret_type       = RET_INTEGER,
7331         .arg1_type      = ARG_PTR_TO_CTX,
7332         .arg2_type      = ARG_ANYTHING,
7333         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7334         .arg4_type      = ARG_CONST_SIZE,
7335         .arg5_type      = ARG_ANYTHING,
7336 };
7337
7338 static const struct bpf_func_proto *
7339 sk_reuseport_func_proto(enum bpf_func_id func_id,
7340                         const struct bpf_prog *prog)
7341 {
7342         switch (func_id) {
7343         case BPF_FUNC_sk_select_reuseport:
7344                 return &sk_select_reuseport_proto;
7345         case BPF_FUNC_skb_load_bytes:
7346                 return &sk_reuseport_load_bytes_proto;
7347         case BPF_FUNC_skb_load_bytes_relative:
7348                 return &sk_reuseport_load_bytes_relative_proto;
7349         default:
7350                 return bpf_base_func_proto(func_id);
7351         }
7352 }
7353
7354 static bool
7355 sk_reuseport_is_valid_access(int off, int size,
7356                              enum bpf_access_type type,
7357                              const struct bpf_prog *prog,
7358                              struct bpf_insn_access_aux *info)
7359 {
7360         const u32 size_default = sizeof(__u32);
7361
7362         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7363             off % size || type != BPF_READ)
7364                 return false;
7365
7366         switch (off) {
7367         case offsetof(struct sk_reuseport_md, data):
7368                 info->reg_type = PTR_TO_PACKET;
7369                 return size == sizeof(__u64);
7370
7371         case offsetof(struct sk_reuseport_md, data_end):
7372                 info->reg_type = PTR_TO_PACKET_END;
7373                 return size == sizeof(__u64);
7374
7375         case offsetof(struct sk_reuseport_md, hash):
7376                 return size == size_default;
7377
7378         /* Fields that allow narrowing */
7379         case offsetof(struct sk_reuseport_md, eth_protocol):
7380                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7381                         return false;
7382                 /* fall through */
7383         case offsetof(struct sk_reuseport_md, ip_protocol):
7384         case offsetof(struct sk_reuseport_md, bind_inany):
7385         case offsetof(struct sk_reuseport_md, len):
7386                 bpf_ctx_record_field_size(info, size_default);
7387                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7388
7389         default:
7390                 return false;
7391         }
7392 }
7393
7394 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
7395         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7396                               si->dst_reg, si->src_reg,                 \
7397                               bpf_target_off(struct sk_reuseport_kern, F, \
7398                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7399                                              target_size));             \
7400         })
7401
7402 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
7403         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
7404                                     struct sk_buff,                     \
7405                                     skb,                                \
7406                                     SKB_FIELD)
7407
7408 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7409         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
7410                                              struct sock,               \
7411                                              sk,                        \
7412                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
7413
7414 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7415                                            const struct bpf_insn *si,
7416                                            struct bpf_insn *insn_buf,
7417                                            struct bpf_prog *prog,
7418                                            u32 *target_size)
7419 {
7420         struct bpf_insn *insn = insn_buf;
7421
7422         switch (si->off) {
7423         case offsetof(struct sk_reuseport_md, data):
7424                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7425                 break;
7426
7427         case offsetof(struct sk_reuseport_md, len):
7428                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7429                 break;
7430
7431         case offsetof(struct sk_reuseport_md, eth_protocol):
7432                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7433                 break;
7434
7435         case offsetof(struct sk_reuseport_md, ip_protocol):
7436                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7437                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7438                                                     BPF_W, 0);
7439                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7440                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7441                                         SK_FL_PROTO_SHIFT);
7442                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
7443                  * aware.  No further narrowing or masking is needed.
7444                  */
7445                 *target_size = 1;
7446                 break;
7447
7448         case offsetof(struct sk_reuseport_md, data_end):
7449                 SK_REUSEPORT_LOAD_FIELD(data_end);
7450                 break;
7451
7452         case offsetof(struct sk_reuseport_md, hash):
7453                 SK_REUSEPORT_LOAD_FIELD(hash);
7454                 break;
7455
7456         case offsetof(struct sk_reuseport_md, bind_inany):
7457                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
7458                 break;
7459         }
7460
7461         return insn - insn_buf;
7462 }
7463
7464 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
7465         .get_func_proto         = sk_reuseport_func_proto,
7466         .is_valid_access        = sk_reuseport_is_valid_access,
7467         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
7468 };
7469
7470 const struct bpf_prog_ops sk_reuseport_prog_ops = {
7471 };
7472 #endif /* CONFIG_INET */