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_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start,
2301            u32, len, u64, flags)
2302 {
2303         struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge;
2304         u32 new, i = 0, l, space, copy = 0, offset = 0;
2305         u8 *raw, *to, *from;
2306         struct page *page;
2307
2308         if (unlikely(flags))
2309                 return -EINVAL;
2310
2311         /* First find the starting scatterlist element */
2312         i = msg->sg.start;
2313         do {
2314                 l = sk_msg_elem(msg, i)->length;
2315
2316                 if (start < offset + l)
2317                         break;
2318                 offset += l;
2319                 sk_msg_iter_var_next(i);
2320         } while (i != msg->sg.end);
2321
2322         if (start >= offset + l)
2323                 return -EINVAL;
2324
2325         space = MAX_MSG_FRAGS - sk_msg_elem_used(msg);
2326
2327         /* If no space available will fallback to copy, we need at
2328          * least one scatterlist elem available to push data into
2329          * when start aligns to the beginning of an element or two
2330          * when it falls inside an element. We handle the start equals
2331          * offset case because its the common case for inserting a
2332          * header.
2333          */
2334         if (!space || (space == 1 && start != offset))
2335                 copy = msg->sg.data[i].length;
2336
2337         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2338                            get_order(copy + len));
2339         if (unlikely(!page))
2340                 return -ENOMEM;
2341
2342         if (copy) {
2343                 int front, back;
2344
2345                 raw = page_address(page);
2346
2347                 psge = sk_msg_elem(msg, i);
2348                 front = start - offset;
2349                 back = psge->length - front;
2350                 from = sg_virt(psge);
2351
2352                 if (front)
2353                         memcpy(raw, from, front);
2354
2355                 if (back) {
2356                         from += front;
2357                         to = raw + front + len;
2358
2359                         memcpy(to, from, back);
2360                 }
2361
2362                 put_page(sg_page(psge));
2363         } else if (start - offset) {
2364                 psge = sk_msg_elem(msg, i);
2365                 rsge = sk_msg_elem_cpy(msg, i);
2366
2367                 psge->length = start - offset;
2368                 rsge.length -= psge->length;
2369                 rsge.offset += start;
2370
2371                 sk_msg_iter_var_next(i);
2372                 sg_unmark_end(psge);
2373                 sk_msg_iter_next(msg, end);
2374         }
2375
2376         /* Slot(s) to place newly allocated data */
2377         new = i;
2378
2379         /* Shift one or two slots as needed */
2380         if (!copy) {
2381                 sge = sk_msg_elem_cpy(msg, i);
2382
2383                 sk_msg_iter_var_next(i);
2384                 sg_unmark_end(&sge);
2385                 sk_msg_iter_next(msg, end);
2386
2387                 nsge = sk_msg_elem_cpy(msg, i);
2388                 if (rsge.length) {
2389                         sk_msg_iter_var_next(i);
2390                         nnsge = sk_msg_elem_cpy(msg, i);
2391                 }
2392
2393                 while (i != msg->sg.end) {
2394                         msg->sg.data[i] = sge;
2395                         sge = nsge;
2396                         sk_msg_iter_var_next(i);
2397                         if (rsge.length) {
2398                                 nsge = nnsge;
2399                                 nnsge = sk_msg_elem_cpy(msg, i);
2400                         } else {
2401                                 nsge = sk_msg_elem_cpy(msg, i);
2402                         }
2403                 }
2404         }
2405
2406         /* Place newly allocated data buffer */
2407         sk_mem_charge(msg->sk, len);
2408         msg->sg.size += len;
2409         msg->sg.copy[new] = false;
2410         sg_set_page(&msg->sg.data[new], page, len + copy, 0);
2411         if (rsge.length) {
2412                 get_page(sg_page(&rsge));
2413                 sk_msg_iter_var_next(new);
2414                 msg->sg.data[new] = rsge;
2415         }
2416
2417         sk_msg_compute_data_pointers(msg);
2418         return 0;
2419 }
2420
2421 static const struct bpf_func_proto bpf_msg_push_data_proto = {
2422         .func           = bpf_msg_push_data,
2423         .gpl_only       = false,
2424         .ret_type       = RET_INTEGER,
2425         .arg1_type      = ARG_PTR_TO_CTX,
2426         .arg2_type      = ARG_ANYTHING,
2427         .arg3_type      = ARG_ANYTHING,
2428         .arg4_type      = ARG_ANYTHING,
2429 };
2430
2431 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2432 {
2433         return task_get_classid(skb);
2434 }
2435
2436 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2437         .func           = bpf_get_cgroup_classid,
2438         .gpl_only       = false,
2439         .ret_type       = RET_INTEGER,
2440         .arg1_type      = ARG_PTR_TO_CTX,
2441 };
2442
2443 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2444 {
2445         return dst_tclassid(skb);
2446 }
2447
2448 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2449         .func           = bpf_get_route_realm,
2450         .gpl_only       = false,
2451         .ret_type       = RET_INTEGER,
2452         .arg1_type      = ARG_PTR_TO_CTX,
2453 };
2454
2455 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2456 {
2457         /* If skb_clear_hash() was called due to mangling, we can
2458          * trigger SW recalculation here. Later access to hash
2459          * can then use the inline skb->hash via context directly
2460          * instead of calling this helper again.
2461          */
2462         return skb_get_hash(skb);
2463 }
2464
2465 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2466         .func           = bpf_get_hash_recalc,
2467         .gpl_only       = false,
2468         .ret_type       = RET_INTEGER,
2469         .arg1_type      = ARG_PTR_TO_CTX,
2470 };
2471
2472 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2473 {
2474         /* After all direct packet write, this can be used once for
2475          * triggering a lazy recalc on next skb_get_hash() invocation.
2476          */
2477         skb_clear_hash(skb);
2478         return 0;
2479 }
2480
2481 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2482         .func           = bpf_set_hash_invalid,
2483         .gpl_only       = false,
2484         .ret_type       = RET_INTEGER,
2485         .arg1_type      = ARG_PTR_TO_CTX,
2486 };
2487
2488 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2489 {
2490         /* Set user specified hash as L4(+), so that it gets returned
2491          * on skb_get_hash() call unless BPF prog later on triggers a
2492          * skb_clear_hash().
2493          */
2494         __skb_set_sw_hash(skb, hash, true);
2495         return 0;
2496 }
2497
2498 static const struct bpf_func_proto bpf_set_hash_proto = {
2499         .func           = bpf_set_hash,
2500         .gpl_only       = false,
2501         .ret_type       = RET_INTEGER,
2502         .arg1_type      = ARG_PTR_TO_CTX,
2503         .arg2_type      = ARG_ANYTHING,
2504 };
2505
2506 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2507            u16, vlan_tci)
2508 {
2509         int ret;
2510
2511         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2512                      vlan_proto != htons(ETH_P_8021AD)))
2513                 vlan_proto = htons(ETH_P_8021Q);
2514
2515         bpf_push_mac_rcsum(skb);
2516         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2517         bpf_pull_mac_rcsum(skb);
2518
2519         bpf_compute_data_pointers(skb);
2520         return ret;
2521 }
2522
2523 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2524         .func           = bpf_skb_vlan_push,
2525         .gpl_only       = false,
2526         .ret_type       = RET_INTEGER,
2527         .arg1_type      = ARG_PTR_TO_CTX,
2528         .arg2_type      = ARG_ANYTHING,
2529         .arg3_type      = ARG_ANYTHING,
2530 };
2531
2532 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2533 {
2534         int ret;
2535
2536         bpf_push_mac_rcsum(skb);
2537         ret = skb_vlan_pop(skb);
2538         bpf_pull_mac_rcsum(skb);
2539
2540         bpf_compute_data_pointers(skb);
2541         return ret;
2542 }
2543
2544 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2545         .func           = bpf_skb_vlan_pop,
2546         .gpl_only       = false,
2547         .ret_type       = RET_INTEGER,
2548         .arg1_type      = ARG_PTR_TO_CTX,
2549 };
2550
2551 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2552 {
2553         /* Caller already did skb_cow() with len as headroom,
2554          * so no need to do it here.
2555          */
2556         skb_push(skb, len);
2557         memmove(skb->data, skb->data + len, off);
2558         memset(skb->data + off, 0, len);
2559
2560         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2561          * needed here as it does not change the skb->csum
2562          * result for checksum complete when summing over
2563          * zeroed blocks.
2564          */
2565         return 0;
2566 }
2567
2568 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2569 {
2570         /* skb_ensure_writable() is not needed here, as we're
2571          * already working on an uncloned skb.
2572          */
2573         if (unlikely(!pskb_may_pull(skb, off + len)))
2574                 return -ENOMEM;
2575
2576         skb_postpull_rcsum(skb, skb->data + off, len);
2577         memmove(skb->data + len, skb->data, off);
2578         __skb_pull(skb, len);
2579
2580         return 0;
2581 }
2582
2583 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2584 {
2585         bool trans_same = skb->transport_header == skb->network_header;
2586         int ret;
2587
2588         /* There's no need for __skb_push()/__skb_pull() pair to
2589          * get to the start of the mac header as we're guaranteed
2590          * to always start from here under eBPF.
2591          */
2592         ret = bpf_skb_generic_push(skb, off, len);
2593         if (likely(!ret)) {
2594                 skb->mac_header -= len;
2595                 skb->network_header -= len;
2596                 if (trans_same)
2597                         skb->transport_header = skb->network_header;
2598         }
2599
2600         return ret;
2601 }
2602
2603 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2604 {
2605         bool trans_same = skb->transport_header == skb->network_header;
2606         int ret;
2607
2608         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2609         ret = bpf_skb_generic_pop(skb, off, len);
2610         if (likely(!ret)) {
2611                 skb->mac_header += len;
2612                 skb->network_header += len;
2613                 if (trans_same)
2614                         skb->transport_header = skb->network_header;
2615         }
2616
2617         return ret;
2618 }
2619
2620 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2621 {
2622         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2623         u32 off = skb_mac_header_len(skb);
2624         int ret;
2625
2626         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2627         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2628                 return -ENOTSUPP;
2629
2630         ret = skb_cow(skb, len_diff);
2631         if (unlikely(ret < 0))
2632                 return ret;
2633
2634         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2635         if (unlikely(ret < 0))
2636                 return ret;
2637
2638         if (skb_is_gso(skb)) {
2639                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2640
2641                 /* SKB_GSO_TCPV4 needs to be changed into
2642                  * SKB_GSO_TCPV6.
2643                  */
2644                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2645                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2646                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2647                 }
2648
2649                 /* Due to IPv6 header, MSS needs to be downgraded. */
2650                 skb_decrease_gso_size(shinfo, len_diff);
2651                 /* Header must be checked, and gso_segs recomputed. */
2652                 shinfo->gso_type |= SKB_GSO_DODGY;
2653                 shinfo->gso_segs = 0;
2654         }
2655
2656         skb->protocol = htons(ETH_P_IPV6);
2657         skb_clear_hash(skb);
2658
2659         return 0;
2660 }
2661
2662 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2663 {
2664         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2665         u32 off = skb_mac_header_len(skb);
2666         int ret;
2667
2668         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2669         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2670                 return -ENOTSUPP;
2671
2672         ret = skb_unclone(skb, GFP_ATOMIC);
2673         if (unlikely(ret < 0))
2674                 return ret;
2675
2676         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2677         if (unlikely(ret < 0))
2678                 return ret;
2679
2680         if (skb_is_gso(skb)) {
2681                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2682
2683                 /* SKB_GSO_TCPV6 needs to be changed into
2684                  * SKB_GSO_TCPV4.
2685                  */
2686                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2687                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2688                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2689                 }
2690
2691                 /* Due to IPv4 header, MSS can be upgraded. */
2692                 skb_increase_gso_size(shinfo, len_diff);
2693                 /* Header must be checked, and gso_segs recomputed. */
2694                 shinfo->gso_type |= SKB_GSO_DODGY;
2695                 shinfo->gso_segs = 0;
2696         }
2697
2698         skb->protocol = htons(ETH_P_IP);
2699         skb_clear_hash(skb);
2700
2701         return 0;
2702 }
2703
2704 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2705 {
2706         __be16 from_proto = skb->protocol;
2707
2708         if (from_proto == htons(ETH_P_IP) &&
2709               to_proto == htons(ETH_P_IPV6))
2710                 return bpf_skb_proto_4_to_6(skb);
2711
2712         if (from_proto == htons(ETH_P_IPV6) &&
2713               to_proto == htons(ETH_P_IP))
2714                 return bpf_skb_proto_6_to_4(skb);
2715
2716         return -ENOTSUPP;
2717 }
2718
2719 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2720            u64, flags)
2721 {
2722         int ret;
2723
2724         if (unlikely(flags))
2725                 return -EINVAL;
2726
2727         /* General idea is that this helper does the basic groundwork
2728          * needed for changing the protocol, and eBPF program fills the
2729          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2730          * and other helpers, rather than passing a raw buffer here.
2731          *
2732          * The rationale is to keep this minimal and without a need to
2733          * deal with raw packet data. F.e. even if we would pass buffers
2734          * here, the program still needs to call the bpf_lX_csum_replace()
2735          * helpers anyway. Plus, this way we keep also separation of
2736          * concerns, since f.e. bpf_skb_store_bytes() should only take
2737          * care of stores.
2738          *
2739          * Currently, additional options and extension header space are
2740          * not supported, but flags register is reserved so we can adapt
2741          * that. For offloads, we mark packet as dodgy, so that headers
2742          * need to be verified first.
2743          */
2744         ret = bpf_skb_proto_xlat(skb, proto);
2745         bpf_compute_data_pointers(skb);
2746         return ret;
2747 }
2748
2749 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2750         .func           = bpf_skb_change_proto,
2751         .gpl_only       = false,
2752         .ret_type       = RET_INTEGER,
2753         .arg1_type      = ARG_PTR_TO_CTX,
2754         .arg2_type      = ARG_ANYTHING,
2755         .arg3_type      = ARG_ANYTHING,
2756 };
2757
2758 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2759 {
2760         /* We only allow a restricted subset to be changed for now. */
2761         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2762                      !skb_pkt_type_ok(pkt_type)))
2763                 return -EINVAL;
2764
2765         skb->pkt_type = pkt_type;
2766         return 0;
2767 }
2768
2769 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2770         .func           = bpf_skb_change_type,
2771         .gpl_only       = false,
2772         .ret_type       = RET_INTEGER,
2773         .arg1_type      = ARG_PTR_TO_CTX,
2774         .arg2_type      = ARG_ANYTHING,
2775 };
2776
2777 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2778 {
2779         switch (skb->protocol) {
2780         case htons(ETH_P_IP):
2781                 return sizeof(struct iphdr);
2782         case htons(ETH_P_IPV6):
2783                 return sizeof(struct ipv6hdr);
2784         default:
2785                 return ~0U;
2786         }
2787 }
2788
2789 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2790 {
2791         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2792         int ret;
2793
2794         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2795         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2796                 return -ENOTSUPP;
2797
2798         ret = skb_cow(skb, len_diff);
2799         if (unlikely(ret < 0))
2800                 return ret;
2801
2802         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2803         if (unlikely(ret < 0))
2804                 return ret;
2805
2806         if (skb_is_gso(skb)) {
2807                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2808
2809                 /* Due to header grow, MSS needs to be downgraded. */
2810                 skb_decrease_gso_size(shinfo, len_diff);
2811                 /* Header must be checked, and gso_segs recomputed. */
2812                 shinfo->gso_type |= SKB_GSO_DODGY;
2813                 shinfo->gso_segs = 0;
2814         }
2815
2816         return 0;
2817 }
2818
2819 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2820 {
2821         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2822         int ret;
2823
2824         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2825         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2826                 return -ENOTSUPP;
2827
2828         ret = skb_unclone(skb, GFP_ATOMIC);
2829         if (unlikely(ret < 0))
2830                 return ret;
2831
2832         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2833         if (unlikely(ret < 0))
2834                 return ret;
2835
2836         if (skb_is_gso(skb)) {
2837                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2838
2839                 /* Due to header shrink, MSS can be upgraded. */
2840                 skb_increase_gso_size(shinfo, len_diff);
2841                 /* Header must be checked, and gso_segs recomputed. */
2842                 shinfo->gso_type |= SKB_GSO_DODGY;
2843                 shinfo->gso_segs = 0;
2844         }
2845
2846         return 0;
2847 }
2848
2849 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2850 {
2851         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2852                           SKB_MAX_ALLOC;
2853 }
2854
2855 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2856 {
2857         bool trans_same = skb->transport_header == skb->network_header;
2858         u32 len_cur, len_diff_abs = abs(len_diff);
2859         u32 len_min = bpf_skb_net_base_len(skb);
2860         u32 len_max = __bpf_skb_max_len(skb);
2861         __be16 proto = skb->protocol;
2862         bool shrink = len_diff < 0;
2863         int ret;
2864
2865         if (unlikely(len_diff_abs > 0xfffU))
2866                 return -EFAULT;
2867         if (unlikely(proto != htons(ETH_P_IP) &&
2868                      proto != htons(ETH_P_IPV6)))
2869                 return -ENOTSUPP;
2870
2871         len_cur = skb->len - skb_network_offset(skb);
2872         if (skb_transport_header_was_set(skb) && !trans_same)
2873                 len_cur = skb_network_header_len(skb);
2874         if ((shrink && (len_diff_abs >= len_cur ||
2875                         len_cur - len_diff_abs < len_min)) ||
2876             (!shrink && (skb->len + len_diff_abs > len_max &&
2877                          !skb_is_gso(skb))))
2878                 return -ENOTSUPP;
2879
2880         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2881                        bpf_skb_net_grow(skb, len_diff_abs);
2882
2883         bpf_compute_data_pointers(skb);
2884         return ret;
2885 }
2886
2887 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2888            u32, mode, u64, flags)
2889 {
2890         if (unlikely(flags))
2891                 return -EINVAL;
2892         if (likely(mode == BPF_ADJ_ROOM_NET))
2893                 return bpf_skb_adjust_net(skb, len_diff);
2894
2895         return -ENOTSUPP;
2896 }
2897
2898 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2899         .func           = bpf_skb_adjust_room,
2900         .gpl_only       = false,
2901         .ret_type       = RET_INTEGER,
2902         .arg1_type      = ARG_PTR_TO_CTX,
2903         .arg2_type      = ARG_ANYTHING,
2904         .arg3_type      = ARG_ANYTHING,
2905         .arg4_type      = ARG_ANYTHING,
2906 };
2907
2908 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2909 {
2910         u32 min_len = skb_network_offset(skb);
2911
2912         if (skb_transport_header_was_set(skb))
2913                 min_len = skb_transport_offset(skb);
2914         if (skb->ip_summed == CHECKSUM_PARTIAL)
2915                 min_len = skb_checksum_start_offset(skb) +
2916                           skb->csum_offset + sizeof(__sum16);
2917         return min_len;
2918 }
2919
2920 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2921 {
2922         unsigned int old_len = skb->len;
2923         int ret;
2924
2925         ret = __skb_grow_rcsum(skb, new_len);
2926         if (!ret)
2927                 memset(skb->data + old_len, 0, new_len - old_len);
2928         return ret;
2929 }
2930
2931 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2932 {
2933         return __skb_trim_rcsum(skb, new_len);
2934 }
2935
2936 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2937                                         u64 flags)
2938 {
2939         u32 max_len = __bpf_skb_max_len(skb);
2940         u32 min_len = __bpf_skb_min_len(skb);
2941         int ret;
2942
2943         if (unlikely(flags || new_len > max_len || new_len < min_len))
2944                 return -EINVAL;
2945         if (skb->encapsulation)
2946                 return -ENOTSUPP;
2947
2948         /* The basic idea of this helper is that it's performing the
2949          * needed work to either grow or trim an skb, and eBPF program
2950          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2951          * bpf_lX_csum_replace() and others rather than passing a raw
2952          * buffer here. This one is a slow path helper and intended
2953          * for replies with control messages.
2954          *
2955          * Like in bpf_skb_change_proto(), we want to keep this rather
2956          * minimal and without protocol specifics so that we are able
2957          * to separate concerns as in bpf_skb_store_bytes() should only
2958          * be the one responsible for writing buffers.
2959          *
2960          * It's really expected to be a slow path operation here for
2961          * control message replies, so we're implicitly linearizing,
2962          * uncloning and drop offloads from the skb by this.
2963          */
2964         ret = __bpf_try_make_writable(skb, skb->len);
2965         if (!ret) {
2966                 if (new_len > skb->len)
2967                         ret = bpf_skb_grow_rcsum(skb, new_len);
2968                 else if (new_len < skb->len)
2969                         ret = bpf_skb_trim_rcsum(skb, new_len);
2970                 if (!ret && skb_is_gso(skb))
2971                         skb_gso_reset(skb);
2972         }
2973         return ret;
2974 }
2975
2976 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2977            u64, flags)
2978 {
2979         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2980
2981         bpf_compute_data_pointers(skb);
2982         return ret;
2983 }
2984
2985 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2986         .func           = bpf_skb_change_tail,
2987         .gpl_only       = false,
2988         .ret_type       = RET_INTEGER,
2989         .arg1_type      = ARG_PTR_TO_CTX,
2990         .arg2_type      = ARG_ANYTHING,
2991         .arg3_type      = ARG_ANYTHING,
2992 };
2993
2994 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2995            u64, flags)
2996 {
2997         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2998
2999         bpf_compute_data_end_sk_skb(skb);
3000         return ret;
3001 }
3002
3003 static const struct bpf_func_proto sk_skb_change_tail_proto = {
3004         .func           = sk_skb_change_tail,
3005         .gpl_only       = false,
3006         .ret_type       = RET_INTEGER,
3007         .arg1_type      = ARG_PTR_TO_CTX,
3008         .arg2_type      = ARG_ANYTHING,
3009         .arg3_type      = ARG_ANYTHING,
3010 };
3011
3012 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3013                                         u64 flags)
3014 {
3015         u32 max_len = __bpf_skb_max_len(skb);
3016         u32 new_len = skb->len + head_room;
3017         int ret;
3018
3019         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3020                      new_len < skb->len))
3021                 return -EINVAL;
3022
3023         ret = skb_cow(skb, head_room);
3024         if (likely(!ret)) {
3025                 /* Idea for this helper is that we currently only
3026                  * allow to expand on mac header. This means that
3027                  * skb->protocol network header, etc, stay as is.
3028                  * Compared to bpf_skb_change_tail(), we're more
3029                  * flexible due to not needing to linearize or
3030                  * reset GSO. Intention for this helper is to be
3031                  * used by an L3 skb that needs to push mac header
3032                  * for redirection into L2 device.
3033                  */
3034                 __skb_push(skb, head_room);
3035                 memset(skb->data, 0, head_room);
3036                 skb_reset_mac_header(skb);
3037         }
3038
3039         return ret;
3040 }
3041
3042 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3043            u64, flags)
3044 {
3045         int ret = __bpf_skb_change_head(skb, head_room, flags);
3046
3047         bpf_compute_data_pointers(skb);
3048         return ret;
3049 }
3050
3051 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3052         .func           = bpf_skb_change_head,
3053         .gpl_only       = false,
3054         .ret_type       = RET_INTEGER,
3055         .arg1_type      = ARG_PTR_TO_CTX,
3056         .arg2_type      = ARG_ANYTHING,
3057         .arg3_type      = ARG_ANYTHING,
3058 };
3059
3060 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3061            u64, flags)
3062 {
3063         int ret = __bpf_skb_change_head(skb, head_room, flags);
3064
3065         bpf_compute_data_end_sk_skb(skb);
3066         return ret;
3067 }
3068
3069 static const struct bpf_func_proto sk_skb_change_head_proto = {
3070         .func           = sk_skb_change_head,
3071         .gpl_only       = false,
3072         .ret_type       = RET_INTEGER,
3073         .arg1_type      = ARG_PTR_TO_CTX,
3074         .arg2_type      = ARG_ANYTHING,
3075         .arg3_type      = ARG_ANYTHING,
3076 };
3077 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3078 {
3079         return xdp_data_meta_unsupported(xdp) ? 0 :
3080                xdp->data - xdp->data_meta;
3081 }
3082
3083 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3084 {
3085         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3086         unsigned long metalen = xdp_get_metalen(xdp);
3087         void *data_start = xdp_frame_end + metalen;
3088         void *data = xdp->data + offset;
3089
3090         if (unlikely(data < data_start ||
3091                      data > xdp->data_end - ETH_HLEN))
3092                 return -EINVAL;
3093
3094         if (metalen)
3095                 memmove(xdp->data_meta + offset,
3096                         xdp->data_meta, metalen);
3097         xdp->data_meta += offset;
3098         xdp->data = data;
3099
3100         return 0;
3101 }
3102
3103 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3104         .func           = bpf_xdp_adjust_head,
3105         .gpl_only       = false,
3106         .ret_type       = RET_INTEGER,
3107         .arg1_type      = ARG_PTR_TO_CTX,
3108         .arg2_type      = ARG_ANYTHING,
3109 };
3110
3111 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3112 {
3113         void *data_end = xdp->data_end + offset;
3114
3115         /* only shrinking is allowed for now. */
3116         if (unlikely(offset >= 0))
3117                 return -EINVAL;
3118
3119         if (unlikely(data_end < xdp->data + ETH_HLEN))
3120                 return -EINVAL;
3121
3122         xdp->data_end = data_end;
3123
3124         return 0;
3125 }
3126
3127 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3128         .func           = bpf_xdp_adjust_tail,
3129         .gpl_only       = false,
3130         .ret_type       = RET_INTEGER,
3131         .arg1_type      = ARG_PTR_TO_CTX,
3132         .arg2_type      = ARG_ANYTHING,
3133 };
3134
3135 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3136 {
3137         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3138         void *meta = xdp->data_meta + offset;
3139         unsigned long metalen = xdp->data - meta;
3140
3141         if (xdp_data_meta_unsupported(xdp))
3142                 return -ENOTSUPP;
3143         if (unlikely(meta < xdp_frame_end ||
3144                      meta > xdp->data))
3145                 return -EINVAL;
3146         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3147                      (metalen > 32)))
3148                 return -EACCES;
3149
3150         xdp->data_meta = meta;
3151
3152         return 0;
3153 }
3154
3155 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3156         .func           = bpf_xdp_adjust_meta,
3157         .gpl_only       = false,
3158         .ret_type       = RET_INTEGER,
3159         .arg1_type      = ARG_PTR_TO_CTX,
3160         .arg2_type      = ARG_ANYTHING,
3161 };
3162
3163 static int __bpf_tx_xdp(struct net_device *dev,
3164                         struct bpf_map *map,
3165                         struct xdp_buff *xdp,
3166                         u32 index)
3167 {
3168         struct xdp_frame *xdpf;
3169         int err, sent;
3170
3171         if (!dev->netdev_ops->ndo_xdp_xmit) {
3172                 return -EOPNOTSUPP;
3173         }
3174
3175         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3176         if (unlikely(err))
3177                 return err;
3178
3179         xdpf = convert_to_xdp_frame(xdp);
3180         if (unlikely(!xdpf))
3181                 return -EOVERFLOW;
3182
3183         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3184         if (sent <= 0)
3185                 return sent;
3186         return 0;
3187 }
3188
3189 static noinline int
3190 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp,
3191                      struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri)
3192 {
3193         struct net_device *fwd;
3194         u32 index = ri->ifindex;
3195         int err;
3196
3197         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3198         ri->ifindex = 0;
3199         if (unlikely(!fwd)) {
3200                 err = -EINVAL;
3201                 goto err;
3202         }
3203
3204         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3205         if (unlikely(err))
3206                 goto err;
3207
3208         _trace_xdp_redirect(dev, xdp_prog, index);
3209         return 0;
3210 err:
3211         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3212         return err;
3213 }
3214
3215 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3216                             struct bpf_map *map,
3217                             struct xdp_buff *xdp,
3218                             u32 index)
3219 {
3220         int err;
3221
3222         switch (map->map_type) {
3223         case BPF_MAP_TYPE_DEVMAP: {
3224                 struct bpf_dtab_netdev *dst = fwd;
3225
3226                 err = dev_map_enqueue(dst, xdp, dev_rx);
3227                 if (unlikely(err))
3228                         return err;
3229                 __dev_map_insert_ctx(map, index);
3230                 break;
3231         }
3232         case BPF_MAP_TYPE_CPUMAP: {
3233                 struct bpf_cpu_map_entry *rcpu = fwd;
3234
3235                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3236                 if (unlikely(err))
3237                         return err;
3238                 __cpu_map_insert_ctx(map, index);
3239                 break;
3240         }
3241         case BPF_MAP_TYPE_XSKMAP: {
3242                 struct xdp_sock *xs = fwd;
3243
3244                 err = __xsk_map_redirect(map, xdp, xs);
3245                 return err;
3246         }
3247         default:
3248                 break;
3249         }
3250         return 0;
3251 }
3252
3253 void xdp_do_flush_map(void)
3254 {
3255         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3256         struct bpf_map *map = ri->map_to_flush;
3257
3258         ri->map_to_flush = NULL;
3259         if (map) {
3260                 switch (map->map_type) {
3261                 case BPF_MAP_TYPE_DEVMAP:
3262                         __dev_map_flush(map);
3263                         break;
3264                 case BPF_MAP_TYPE_CPUMAP:
3265                         __cpu_map_flush(map);
3266                         break;
3267                 case BPF_MAP_TYPE_XSKMAP:
3268                         __xsk_map_flush(map);
3269                         break;
3270                 default:
3271                         break;
3272                 }
3273         }
3274 }
3275 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3276
3277 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3278 {
3279         switch (map->map_type) {
3280         case BPF_MAP_TYPE_DEVMAP:
3281                 return __dev_map_lookup_elem(map, index);
3282         case BPF_MAP_TYPE_CPUMAP:
3283                 return __cpu_map_lookup_elem(map, index);
3284         case BPF_MAP_TYPE_XSKMAP:
3285                 return __xsk_map_lookup_elem(map, index);
3286         default:
3287                 return NULL;
3288         }
3289 }
3290
3291 void bpf_clear_redirect_map(struct bpf_map *map)
3292 {
3293         struct bpf_redirect_info *ri;
3294         int cpu;
3295
3296         for_each_possible_cpu(cpu) {
3297                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3298                 /* Avoid polluting remote cacheline due to writes if
3299                  * not needed. Once we pass this test, we need the
3300                  * cmpxchg() to make sure it hasn't been changed in
3301                  * the meantime by remote CPU.
3302                  */
3303                 if (unlikely(READ_ONCE(ri->map) == map))
3304                         cmpxchg(&ri->map, map, NULL);
3305         }
3306 }
3307
3308 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3309                                struct bpf_prog *xdp_prog, struct bpf_map *map,
3310                                struct bpf_redirect_info *ri)
3311 {
3312         u32 index = ri->ifindex;
3313         void *fwd = NULL;
3314         int err;
3315
3316         ri->ifindex = 0;
3317         WRITE_ONCE(ri->map, NULL);
3318
3319         fwd = __xdp_map_lookup_elem(map, index);
3320         if (unlikely(!fwd)) {
3321                 err = -EINVAL;
3322                 goto err;
3323         }
3324         if (ri->map_to_flush && unlikely(ri->map_to_flush != map))
3325                 xdp_do_flush_map();
3326
3327         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3328         if (unlikely(err))
3329                 goto err;
3330
3331         ri->map_to_flush = map;
3332         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3333         return 0;
3334 err:
3335         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3336         return err;
3337 }
3338
3339 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3340                     struct bpf_prog *xdp_prog)
3341 {
3342         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3343         struct bpf_map *map = READ_ONCE(ri->map);
3344
3345         if (likely(map))
3346                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri);
3347
3348         return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri);
3349 }
3350 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3351
3352 static int xdp_do_generic_redirect_map(struct net_device *dev,
3353                                        struct sk_buff *skb,
3354                                        struct xdp_buff *xdp,
3355                                        struct bpf_prog *xdp_prog,
3356                                        struct bpf_map *map)
3357 {
3358         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3359         u32 index = ri->ifindex;
3360         void *fwd = NULL;
3361         int err = 0;
3362
3363         ri->ifindex = 0;
3364         WRITE_ONCE(ri->map, NULL);
3365
3366         fwd = __xdp_map_lookup_elem(map, index);
3367         if (unlikely(!fwd)) {
3368                 err = -EINVAL;
3369                 goto err;
3370         }
3371
3372         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3373                 struct bpf_dtab_netdev *dst = fwd;
3374
3375                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3376                 if (unlikely(err))
3377                         goto err;
3378         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3379                 struct xdp_sock *xs = fwd;
3380
3381                 err = xsk_generic_rcv(xs, xdp);
3382                 if (err)
3383                         goto err;
3384                 consume_skb(skb);
3385         } else {
3386                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3387                 err = -EBADRQC;
3388                 goto err;
3389         }
3390
3391         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3392         return 0;
3393 err:
3394         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3395         return err;
3396 }
3397
3398 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3399                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3400 {
3401         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3402         struct bpf_map *map = READ_ONCE(ri->map);
3403         u32 index = ri->ifindex;
3404         struct net_device *fwd;
3405         int err = 0;
3406
3407         if (map)
3408                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3409                                                    map);
3410         ri->ifindex = 0;
3411         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3412         if (unlikely(!fwd)) {
3413                 err = -EINVAL;
3414                 goto err;
3415         }
3416
3417         err = xdp_ok_fwd_dev(fwd, skb->len);
3418         if (unlikely(err))
3419                 goto err;
3420
3421         skb->dev = fwd;
3422         _trace_xdp_redirect(dev, xdp_prog, index);
3423         generic_xdp_tx(skb, xdp_prog);
3424         return 0;
3425 err:
3426         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3427         return err;
3428 }
3429 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3430
3431 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3432 {
3433         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3434
3435         if (unlikely(flags))
3436                 return XDP_ABORTED;
3437
3438         ri->ifindex = ifindex;
3439         ri->flags = flags;
3440         WRITE_ONCE(ri->map, NULL);
3441
3442         return XDP_REDIRECT;
3443 }
3444
3445 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3446         .func           = bpf_xdp_redirect,
3447         .gpl_only       = false,
3448         .ret_type       = RET_INTEGER,
3449         .arg1_type      = ARG_ANYTHING,
3450         .arg2_type      = ARG_ANYTHING,
3451 };
3452
3453 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3454            u64, flags)
3455 {
3456         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3457
3458         if (unlikely(flags))
3459                 return XDP_ABORTED;
3460
3461         ri->ifindex = ifindex;
3462         ri->flags = flags;
3463         WRITE_ONCE(ri->map, map);
3464
3465         return XDP_REDIRECT;
3466 }
3467
3468 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3469         .func           = bpf_xdp_redirect_map,
3470         .gpl_only       = false,
3471         .ret_type       = RET_INTEGER,
3472         .arg1_type      = ARG_CONST_MAP_PTR,
3473         .arg2_type      = ARG_ANYTHING,
3474         .arg3_type      = ARG_ANYTHING,
3475 };
3476
3477 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3478                                   unsigned long off, unsigned long len)
3479 {
3480         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3481
3482         if (unlikely(!ptr))
3483                 return len;
3484         if (ptr != dst_buff)
3485                 memcpy(dst_buff, ptr, len);
3486
3487         return 0;
3488 }
3489
3490 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3491            u64, flags, void *, meta, u64, meta_size)
3492 {
3493         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3494
3495         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3496                 return -EINVAL;
3497         if (unlikely(skb_size > skb->len))
3498                 return -EFAULT;
3499
3500         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3501                                 bpf_skb_copy);
3502 }
3503
3504 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3505         .func           = bpf_skb_event_output,
3506         .gpl_only       = true,
3507         .ret_type       = RET_INTEGER,
3508         .arg1_type      = ARG_PTR_TO_CTX,
3509         .arg2_type      = ARG_CONST_MAP_PTR,
3510         .arg3_type      = ARG_ANYTHING,
3511         .arg4_type      = ARG_PTR_TO_MEM,
3512         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3513 };
3514
3515 static unsigned short bpf_tunnel_key_af(u64 flags)
3516 {
3517         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3518 }
3519
3520 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3521            u32, size, u64, flags)
3522 {
3523         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3524         u8 compat[sizeof(struct bpf_tunnel_key)];
3525         void *to_orig = to;
3526         int err;
3527
3528         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3529                 err = -EINVAL;
3530                 goto err_clear;
3531         }
3532         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3533                 err = -EPROTO;
3534                 goto err_clear;
3535         }
3536         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3537                 err = -EINVAL;
3538                 switch (size) {
3539                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3540                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3541                         goto set_compat;
3542                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3543                         /* Fixup deprecated structure layouts here, so we have
3544                          * a common path later on.
3545                          */
3546                         if (ip_tunnel_info_af(info) != AF_INET)
3547                                 goto err_clear;
3548 set_compat:
3549                         to = (struct bpf_tunnel_key *)compat;
3550                         break;
3551                 default:
3552                         goto err_clear;
3553                 }
3554         }
3555
3556         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3557         to->tunnel_tos = info->key.tos;
3558         to->tunnel_ttl = info->key.ttl;
3559         to->tunnel_ext = 0;
3560
3561         if (flags & BPF_F_TUNINFO_IPV6) {
3562                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3563                        sizeof(to->remote_ipv6));
3564                 to->tunnel_label = be32_to_cpu(info->key.label);
3565         } else {
3566                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3567                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3568                 to->tunnel_label = 0;
3569         }
3570
3571         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3572                 memcpy(to_orig, to, size);
3573
3574         return 0;
3575 err_clear:
3576         memset(to_orig, 0, size);
3577         return err;
3578 }
3579
3580 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3581         .func           = bpf_skb_get_tunnel_key,
3582         .gpl_only       = false,
3583         .ret_type       = RET_INTEGER,
3584         .arg1_type      = ARG_PTR_TO_CTX,
3585         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3586         .arg3_type      = ARG_CONST_SIZE,
3587         .arg4_type      = ARG_ANYTHING,
3588 };
3589
3590 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3591 {
3592         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3593         int err;
3594
3595         if (unlikely(!info ||
3596                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3597                 err = -ENOENT;
3598                 goto err_clear;
3599         }
3600         if (unlikely(size < info->options_len)) {
3601                 err = -ENOMEM;
3602                 goto err_clear;
3603         }
3604
3605         ip_tunnel_info_opts_get(to, info);
3606         if (size > info->options_len)
3607                 memset(to + info->options_len, 0, size - info->options_len);
3608
3609         return info->options_len;
3610 err_clear:
3611         memset(to, 0, size);
3612         return err;
3613 }
3614
3615 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3616         .func           = bpf_skb_get_tunnel_opt,
3617         .gpl_only       = false,
3618         .ret_type       = RET_INTEGER,
3619         .arg1_type      = ARG_PTR_TO_CTX,
3620         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3621         .arg3_type      = ARG_CONST_SIZE,
3622 };
3623
3624 static struct metadata_dst __percpu *md_dst;
3625
3626 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3627            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3628 {
3629         struct metadata_dst *md = this_cpu_ptr(md_dst);
3630         u8 compat[sizeof(struct bpf_tunnel_key)];
3631         struct ip_tunnel_info *info;
3632
3633         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3634                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3635                 return -EINVAL;
3636         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3637                 switch (size) {
3638                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3639                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3640                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3641                         /* Fixup deprecated structure layouts here, so we have
3642                          * a common path later on.
3643                          */
3644                         memcpy(compat, from, size);
3645                         memset(compat + size, 0, sizeof(compat) - size);
3646                         from = (const struct bpf_tunnel_key *) compat;
3647                         break;
3648                 default:
3649                         return -EINVAL;
3650                 }
3651         }
3652         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3653                      from->tunnel_ext))
3654                 return -EINVAL;
3655
3656         skb_dst_drop(skb);
3657         dst_hold((struct dst_entry *) md);
3658         skb_dst_set(skb, (struct dst_entry *) md);
3659
3660         info = &md->u.tun_info;
3661         memset(info, 0, sizeof(*info));
3662         info->mode = IP_TUNNEL_INFO_TX;
3663
3664         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3665         if (flags & BPF_F_DONT_FRAGMENT)
3666                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3667         if (flags & BPF_F_ZERO_CSUM_TX)
3668                 info->key.tun_flags &= ~TUNNEL_CSUM;
3669         if (flags & BPF_F_SEQ_NUMBER)
3670                 info->key.tun_flags |= TUNNEL_SEQ;
3671
3672         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3673         info->key.tos = from->tunnel_tos;
3674         info->key.ttl = from->tunnel_ttl;
3675
3676         if (flags & BPF_F_TUNINFO_IPV6) {
3677                 info->mode |= IP_TUNNEL_INFO_IPV6;
3678                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3679                        sizeof(from->remote_ipv6));
3680                 info->key.label = cpu_to_be32(from->tunnel_label) &
3681                                   IPV6_FLOWLABEL_MASK;
3682         } else {
3683                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3684         }
3685
3686         return 0;
3687 }
3688
3689 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3690         .func           = bpf_skb_set_tunnel_key,
3691         .gpl_only       = false,
3692         .ret_type       = RET_INTEGER,
3693         .arg1_type      = ARG_PTR_TO_CTX,
3694         .arg2_type      = ARG_PTR_TO_MEM,
3695         .arg3_type      = ARG_CONST_SIZE,
3696         .arg4_type      = ARG_ANYTHING,
3697 };
3698
3699 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3700            const u8 *, from, u32, size)
3701 {
3702         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3703         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3704
3705         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3706                 return -EINVAL;
3707         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3708                 return -ENOMEM;
3709
3710         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3711
3712         return 0;
3713 }
3714
3715 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3716         .func           = bpf_skb_set_tunnel_opt,
3717         .gpl_only       = false,
3718         .ret_type       = RET_INTEGER,
3719         .arg1_type      = ARG_PTR_TO_CTX,
3720         .arg2_type      = ARG_PTR_TO_MEM,
3721         .arg3_type      = ARG_CONST_SIZE,
3722 };
3723
3724 static const struct bpf_func_proto *
3725 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3726 {
3727         if (!md_dst) {
3728                 struct metadata_dst __percpu *tmp;
3729
3730                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3731                                                 METADATA_IP_TUNNEL,
3732                                                 GFP_KERNEL);
3733                 if (!tmp)
3734                         return NULL;
3735                 if (cmpxchg(&md_dst, NULL, tmp))
3736                         metadata_dst_free_percpu(tmp);
3737         }
3738
3739         switch (which) {
3740         case BPF_FUNC_skb_set_tunnel_key:
3741                 return &bpf_skb_set_tunnel_key_proto;
3742         case BPF_FUNC_skb_set_tunnel_opt:
3743                 return &bpf_skb_set_tunnel_opt_proto;
3744         default:
3745                 return NULL;
3746         }
3747 }
3748
3749 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3750            u32, idx)
3751 {
3752         struct bpf_array *array = container_of(map, struct bpf_array, map);
3753         struct cgroup *cgrp;
3754         struct sock *sk;
3755
3756         sk = skb_to_full_sk(skb);
3757         if (!sk || !sk_fullsock(sk))
3758                 return -ENOENT;
3759         if (unlikely(idx >= array->map.max_entries))
3760                 return -E2BIG;
3761
3762         cgrp = READ_ONCE(array->ptrs[idx]);
3763         if (unlikely(!cgrp))
3764                 return -EAGAIN;
3765
3766         return sk_under_cgroup_hierarchy(sk, cgrp);
3767 }
3768
3769 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3770         .func           = bpf_skb_under_cgroup,
3771         .gpl_only       = false,
3772         .ret_type       = RET_INTEGER,
3773         .arg1_type      = ARG_PTR_TO_CTX,
3774         .arg2_type      = ARG_CONST_MAP_PTR,
3775         .arg3_type      = ARG_ANYTHING,
3776 };
3777
3778 #ifdef CONFIG_SOCK_CGROUP_DATA
3779 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3780 {
3781         struct sock *sk = skb_to_full_sk(skb);
3782         struct cgroup *cgrp;
3783
3784         if (!sk || !sk_fullsock(sk))
3785                 return 0;
3786
3787         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3788         return cgrp->kn->id.id;
3789 }
3790
3791 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3792         .func           = bpf_skb_cgroup_id,
3793         .gpl_only       = false,
3794         .ret_type       = RET_INTEGER,
3795         .arg1_type      = ARG_PTR_TO_CTX,
3796 };
3797
3798 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3799            ancestor_level)
3800 {
3801         struct sock *sk = skb_to_full_sk(skb);
3802         struct cgroup *ancestor;
3803         struct cgroup *cgrp;
3804
3805         if (!sk || !sk_fullsock(sk))
3806                 return 0;
3807
3808         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3809         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3810         if (!ancestor)
3811                 return 0;
3812
3813         return ancestor->kn->id.id;
3814 }
3815
3816 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3817         .func           = bpf_skb_ancestor_cgroup_id,
3818         .gpl_only       = false,
3819         .ret_type       = RET_INTEGER,
3820         .arg1_type      = ARG_PTR_TO_CTX,
3821         .arg2_type      = ARG_ANYTHING,
3822 };
3823 #endif
3824
3825 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3826                                   unsigned long off, unsigned long len)
3827 {
3828         memcpy(dst_buff, src_buff + off, len);
3829         return 0;
3830 }
3831
3832 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3833            u64, flags, void *, meta, u64, meta_size)
3834 {
3835         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3836
3837         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3838                 return -EINVAL;
3839         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3840                 return -EFAULT;
3841
3842         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3843                                 xdp_size, bpf_xdp_copy);
3844 }
3845
3846 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3847         .func           = bpf_xdp_event_output,
3848         .gpl_only       = true,
3849         .ret_type       = RET_INTEGER,
3850         .arg1_type      = ARG_PTR_TO_CTX,
3851         .arg2_type      = ARG_CONST_MAP_PTR,
3852         .arg3_type      = ARG_ANYTHING,
3853         .arg4_type      = ARG_PTR_TO_MEM,
3854         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3855 };
3856
3857 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3858 {
3859         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3860 }
3861
3862 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3863         .func           = bpf_get_socket_cookie,
3864         .gpl_only       = false,
3865         .ret_type       = RET_INTEGER,
3866         .arg1_type      = ARG_PTR_TO_CTX,
3867 };
3868
3869 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
3870 {
3871         return sock_gen_cookie(ctx->sk);
3872 }
3873
3874 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
3875         .func           = bpf_get_socket_cookie_sock_addr,
3876         .gpl_only       = false,
3877         .ret_type       = RET_INTEGER,
3878         .arg1_type      = ARG_PTR_TO_CTX,
3879 };
3880
3881 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
3882 {
3883         return sock_gen_cookie(ctx->sk);
3884 }
3885
3886 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
3887         .func           = bpf_get_socket_cookie_sock_ops,
3888         .gpl_only       = false,
3889         .ret_type       = RET_INTEGER,
3890         .arg1_type      = ARG_PTR_TO_CTX,
3891 };
3892
3893 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3894 {
3895         struct sock *sk = sk_to_full_sk(skb->sk);
3896         kuid_t kuid;
3897
3898         if (!sk || !sk_fullsock(sk))
3899                 return overflowuid;
3900         kuid = sock_net_uid(sock_net(sk), sk);
3901         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3902 }
3903
3904 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3905         .func           = bpf_get_socket_uid,
3906         .gpl_only       = false,
3907         .ret_type       = RET_INTEGER,
3908         .arg1_type      = ARG_PTR_TO_CTX,
3909 };
3910
3911 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3912            int, level, int, optname, char *, optval, int, optlen)
3913 {
3914         struct sock *sk = bpf_sock->sk;
3915         int ret = 0;
3916         int val;
3917
3918         if (!sk_fullsock(sk))
3919                 return -EINVAL;
3920
3921         if (level == SOL_SOCKET) {
3922                 if (optlen != sizeof(int))
3923                         return -EINVAL;
3924                 val = *((int *)optval);
3925
3926                 /* Only some socketops are supported */
3927                 switch (optname) {
3928                 case SO_RCVBUF:
3929                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3930                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3931                         break;
3932                 case SO_SNDBUF:
3933                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3934                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3935                         break;
3936                 case SO_MAX_PACING_RATE: /* 32bit version */
3937                         sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val;
3938                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3939                                                  sk->sk_max_pacing_rate);
3940                         break;
3941                 case SO_PRIORITY:
3942                         sk->sk_priority = val;
3943                         break;
3944                 case SO_RCVLOWAT:
3945                         if (val < 0)
3946                                 val = INT_MAX;
3947                         sk->sk_rcvlowat = val ? : 1;
3948                         break;
3949                 case SO_MARK:
3950                         sk->sk_mark = val;
3951                         break;
3952                 default:
3953                         ret = -EINVAL;
3954                 }
3955 #ifdef CONFIG_INET
3956         } else if (level == SOL_IP) {
3957                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3958                         return -EINVAL;
3959
3960                 val = *((int *)optval);
3961                 /* Only some options are supported */
3962                 switch (optname) {
3963                 case IP_TOS:
3964                         if (val < -1 || val > 0xff) {
3965                                 ret = -EINVAL;
3966                         } else {
3967                                 struct inet_sock *inet = inet_sk(sk);
3968
3969                                 if (val == -1)
3970                                         val = 0;
3971                                 inet->tos = val;
3972                         }
3973                         break;
3974                 default:
3975                         ret = -EINVAL;
3976                 }
3977 #if IS_ENABLED(CONFIG_IPV6)
3978         } else if (level == SOL_IPV6) {
3979                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3980                         return -EINVAL;
3981
3982                 val = *((int *)optval);
3983                 /* Only some options are supported */
3984                 switch (optname) {
3985                 case IPV6_TCLASS:
3986                         if (val < -1 || val > 0xff) {
3987                                 ret = -EINVAL;
3988                         } else {
3989                                 struct ipv6_pinfo *np = inet6_sk(sk);
3990
3991                                 if (val == -1)
3992                                         val = 0;
3993                                 np->tclass = val;
3994                         }
3995                         break;
3996                 default:
3997                         ret = -EINVAL;
3998                 }
3999 #endif
4000         } else if (level == SOL_TCP &&
4001                    sk->sk_prot->setsockopt == tcp_setsockopt) {
4002                 if (optname == TCP_CONGESTION) {
4003                         char name[TCP_CA_NAME_MAX];
4004                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
4005
4006                         strncpy(name, optval, min_t(long, optlen,
4007                                                     TCP_CA_NAME_MAX-1));
4008                         name[TCP_CA_NAME_MAX-1] = 0;
4009                         ret = tcp_set_congestion_control(sk, name, false,
4010                                                          reinit);
4011                 } else {
4012                         struct tcp_sock *tp = tcp_sk(sk);
4013
4014                         if (optlen != sizeof(int))
4015                                 return -EINVAL;
4016
4017                         val = *((int *)optval);
4018                         /* Only some options are supported */
4019                         switch (optname) {
4020                         case TCP_BPF_IW:
4021                                 if (val <= 0 || tp->data_segs_out > 0)
4022                                         ret = -EINVAL;
4023                                 else
4024                                         tp->snd_cwnd = val;
4025                                 break;
4026                         case TCP_BPF_SNDCWND_CLAMP:
4027                                 if (val <= 0) {
4028                                         ret = -EINVAL;
4029                                 } else {
4030                                         tp->snd_cwnd_clamp = val;
4031                                         tp->snd_ssthresh = val;
4032                                 }
4033                                 break;
4034                         case TCP_SAVE_SYN:
4035                                 if (val < 0 || val > 1)
4036                                         ret = -EINVAL;
4037                                 else
4038                                         tp->save_syn = val;
4039                                 break;
4040                         default:
4041                                 ret = -EINVAL;
4042                         }
4043                 }
4044 #endif
4045         } else {
4046                 ret = -EINVAL;
4047         }
4048         return ret;
4049 }
4050
4051 static const struct bpf_func_proto bpf_setsockopt_proto = {
4052         .func           = bpf_setsockopt,
4053         .gpl_only       = false,
4054         .ret_type       = RET_INTEGER,
4055         .arg1_type      = ARG_PTR_TO_CTX,
4056         .arg2_type      = ARG_ANYTHING,
4057         .arg3_type      = ARG_ANYTHING,
4058         .arg4_type      = ARG_PTR_TO_MEM,
4059         .arg5_type      = ARG_CONST_SIZE,
4060 };
4061
4062 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4063            int, level, int, optname, char *, optval, int, optlen)
4064 {
4065         struct sock *sk = bpf_sock->sk;
4066
4067         if (!sk_fullsock(sk))
4068                 goto err_clear;
4069 #ifdef CONFIG_INET
4070         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4071                 struct inet_connection_sock *icsk;
4072                 struct tcp_sock *tp;
4073
4074                 switch (optname) {
4075                 case TCP_CONGESTION:
4076                         icsk = inet_csk(sk);
4077
4078                         if (!icsk->icsk_ca_ops || optlen <= 1)
4079                                 goto err_clear;
4080                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4081                         optval[optlen - 1] = 0;
4082                         break;
4083                 case TCP_SAVED_SYN:
4084                         tp = tcp_sk(sk);
4085
4086                         if (optlen <= 0 || !tp->saved_syn ||
4087                             optlen > tp->saved_syn[0])
4088                                 goto err_clear;
4089                         memcpy(optval, tp->saved_syn + 1, optlen);
4090                         break;
4091                 default:
4092                         goto err_clear;
4093                 }
4094         } else if (level == SOL_IP) {
4095                 struct inet_sock *inet = inet_sk(sk);
4096
4097                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4098                         goto err_clear;
4099
4100                 /* Only some options are supported */
4101                 switch (optname) {
4102                 case IP_TOS:
4103                         *((int *)optval) = (int)inet->tos;
4104                         break;
4105                 default:
4106                         goto err_clear;
4107                 }
4108 #if IS_ENABLED(CONFIG_IPV6)
4109         } else if (level == SOL_IPV6) {
4110                 struct ipv6_pinfo *np = inet6_sk(sk);
4111
4112                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4113                         goto err_clear;
4114
4115                 /* Only some options are supported */
4116                 switch (optname) {
4117                 case IPV6_TCLASS:
4118                         *((int *)optval) = (int)np->tclass;
4119                         break;
4120                 default:
4121                         goto err_clear;
4122                 }
4123 #endif
4124         } else {
4125                 goto err_clear;
4126         }
4127         return 0;
4128 #endif
4129 err_clear:
4130         memset(optval, 0, optlen);
4131         return -EINVAL;
4132 }
4133
4134 static const struct bpf_func_proto bpf_getsockopt_proto = {
4135         .func           = bpf_getsockopt,
4136         .gpl_only       = false,
4137         .ret_type       = RET_INTEGER,
4138         .arg1_type      = ARG_PTR_TO_CTX,
4139         .arg2_type      = ARG_ANYTHING,
4140         .arg3_type      = ARG_ANYTHING,
4141         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4142         .arg5_type      = ARG_CONST_SIZE,
4143 };
4144
4145 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4146            int, argval)
4147 {
4148         struct sock *sk = bpf_sock->sk;
4149         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4150
4151         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4152                 return -EINVAL;
4153
4154         if (val)
4155                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4156
4157         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4158 }
4159
4160 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4161         .func           = bpf_sock_ops_cb_flags_set,
4162         .gpl_only       = false,
4163         .ret_type       = RET_INTEGER,
4164         .arg1_type      = ARG_PTR_TO_CTX,
4165         .arg2_type      = ARG_ANYTHING,
4166 };
4167
4168 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4169 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4170
4171 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4172            int, addr_len)
4173 {
4174 #ifdef CONFIG_INET
4175         struct sock *sk = ctx->sk;
4176         int err;
4177
4178         /* Binding to port can be expensive so it's prohibited in the helper.
4179          * Only binding to IP is supported.
4180          */
4181         err = -EINVAL;
4182         if (addr->sa_family == AF_INET) {
4183                 if (addr_len < sizeof(struct sockaddr_in))
4184                         return err;
4185                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4186                         return err;
4187                 return __inet_bind(sk, addr, addr_len, true, false);
4188 #if IS_ENABLED(CONFIG_IPV6)
4189         } else if (addr->sa_family == AF_INET6) {
4190                 if (addr_len < SIN6_LEN_RFC2133)
4191                         return err;
4192                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4193                         return err;
4194                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4195                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4196                  */
4197                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4198 #endif /* CONFIG_IPV6 */
4199         }
4200 #endif /* CONFIG_INET */
4201
4202         return -EAFNOSUPPORT;
4203 }
4204
4205 static const struct bpf_func_proto bpf_bind_proto = {
4206         .func           = bpf_bind,
4207         .gpl_only       = false,
4208         .ret_type       = RET_INTEGER,
4209         .arg1_type      = ARG_PTR_TO_CTX,
4210         .arg2_type      = ARG_PTR_TO_MEM,
4211         .arg3_type      = ARG_CONST_SIZE,
4212 };
4213
4214 #ifdef CONFIG_XFRM
4215 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4216            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4217 {
4218         const struct sec_path *sp = skb_sec_path(skb);
4219         const struct xfrm_state *x;
4220
4221         if (!sp || unlikely(index >= sp->len || flags))
4222                 goto err_clear;
4223
4224         x = sp->xvec[index];
4225
4226         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4227                 goto err_clear;
4228
4229         to->reqid = x->props.reqid;
4230         to->spi = x->id.spi;
4231         to->family = x->props.family;
4232         to->ext = 0;
4233
4234         if (to->family == AF_INET6) {
4235                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4236                        sizeof(to->remote_ipv6));
4237         } else {
4238                 to->remote_ipv4 = x->props.saddr.a4;
4239                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4240         }
4241
4242         return 0;
4243 err_clear:
4244         memset(to, 0, size);
4245         return -EINVAL;
4246 }
4247
4248 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4249         .func           = bpf_skb_get_xfrm_state,
4250         .gpl_only       = false,
4251         .ret_type       = RET_INTEGER,
4252         .arg1_type      = ARG_PTR_TO_CTX,
4253         .arg2_type      = ARG_ANYTHING,
4254         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4255         .arg4_type      = ARG_CONST_SIZE,
4256         .arg5_type      = ARG_ANYTHING,
4257 };
4258 #endif
4259
4260 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4261 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4262                                   const struct neighbour *neigh,
4263                                   const struct net_device *dev)
4264 {
4265         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4266         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4267         params->h_vlan_TCI = 0;
4268         params->h_vlan_proto = 0;
4269         params->ifindex = dev->ifindex;
4270
4271         return 0;
4272 }
4273 #endif
4274
4275 #if IS_ENABLED(CONFIG_INET)
4276 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4277                                u32 flags, bool check_mtu)
4278 {
4279         struct in_device *in_dev;
4280         struct neighbour *neigh;
4281         struct net_device *dev;
4282         struct fib_result res;
4283         struct fib_nh *nh;
4284         struct flowi4 fl4;
4285         int err;
4286         u32 mtu;
4287
4288         dev = dev_get_by_index_rcu(net, params->ifindex);
4289         if (unlikely(!dev))
4290                 return -ENODEV;
4291
4292         /* verify forwarding is enabled on this interface */
4293         in_dev = __in_dev_get_rcu(dev);
4294         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4295                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4296
4297         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4298                 fl4.flowi4_iif = 1;
4299                 fl4.flowi4_oif = params->ifindex;
4300         } else {
4301                 fl4.flowi4_iif = params->ifindex;
4302                 fl4.flowi4_oif = 0;
4303         }
4304         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4305         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4306         fl4.flowi4_flags = 0;
4307
4308         fl4.flowi4_proto = params->l4_protocol;
4309         fl4.daddr = params->ipv4_dst;
4310         fl4.saddr = params->ipv4_src;
4311         fl4.fl4_sport = params->sport;
4312         fl4.fl4_dport = params->dport;
4313
4314         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4315                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4316                 struct fib_table *tb;
4317
4318                 tb = fib_get_table(net, tbid);
4319                 if (unlikely(!tb))
4320                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4321
4322                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4323         } else {
4324                 fl4.flowi4_mark = 0;
4325                 fl4.flowi4_secid = 0;
4326                 fl4.flowi4_tun_key.tun_id = 0;
4327                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4328
4329                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4330         }
4331
4332         if (err) {
4333                 /* map fib lookup errors to RTN_ type */
4334                 if (err == -EINVAL)
4335                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4336                 if (err == -EHOSTUNREACH)
4337                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4338                 if (err == -EACCES)
4339                         return BPF_FIB_LKUP_RET_PROHIBIT;
4340
4341                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4342         }
4343
4344         if (res.type != RTN_UNICAST)
4345                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4346
4347         if (res.fi->fib_nhs > 1)
4348                 fib_select_path(net, &res, &fl4, NULL);
4349
4350         if (check_mtu) {
4351                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4352                 if (params->tot_len > mtu)
4353                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4354         }
4355
4356         nh = &res.fi->fib_nh[res.nh_sel];
4357
4358         /* do not handle lwt encaps right now */
4359         if (nh->nh_lwtstate)
4360                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4361
4362         dev = nh->nh_dev;
4363         if (nh->nh_gw)
4364                 params->ipv4_dst = nh->nh_gw;
4365
4366         params->rt_metric = res.fi->fib_priority;
4367
4368         /* xdp and cls_bpf programs are run in RCU-bh so
4369          * rcu_read_lock_bh is not needed here
4370          */
4371         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4372         if (!neigh)
4373                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4374
4375         return bpf_fib_set_fwd_params(params, neigh, dev);
4376 }
4377 #endif
4378
4379 #if IS_ENABLED(CONFIG_IPV6)
4380 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4381                                u32 flags, bool check_mtu)
4382 {
4383         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4384         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4385         struct neighbour *neigh;
4386         struct net_device *dev;
4387         struct inet6_dev *idev;
4388         struct fib6_info *f6i;
4389         struct flowi6 fl6;
4390         int strict = 0;
4391         int oif;
4392         u32 mtu;
4393
4394         /* link local addresses are never forwarded */
4395         if (rt6_need_strict(dst) || rt6_need_strict(src))
4396                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4397
4398         dev = dev_get_by_index_rcu(net, params->ifindex);
4399         if (unlikely(!dev))
4400                 return -ENODEV;
4401
4402         idev = __in6_dev_get_safely(dev);
4403         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4404                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4405
4406         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4407                 fl6.flowi6_iif = 1;
4408                 oif = fl6.flowi6_oif = params->ifindex;
4409         } else {
4410                 oif = fl6.flowi6_iif = params->ifindex;
4411                 fl6.flowi6_oif = 0;
4412                 strict = RT6_LOOKUP_F_HAS_SADDR;
4413         }
4414         fl6.flowlabel = params->flowinfo;
4415         fl6.flowi6_scope = 0;
4416         fl6.flowi6_flags = 0;
4417         fl6.mp_hash = 0;
4418
4419         fl6.flowi6_proto = params->l4_protocol;
4420         fl6.daddr = *dst;
4421         fl6.saddr = *src;
4422         fl6.fl6_sport = params->sport;
4423         fl6.fl6_dport = params->dport;
4424
4425         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4426                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4427                 struct fib6_table *tb;
4428
4429                 tb = ipv6_stub->fib6_get_table(net, tbid);
4430                 if (unlikely(!tb))
4431                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4432
4433                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4434         } else {
4435                 fl6.flowi6_mark = 0;
4436                 fl6.flowi6_secid = 0;
4437                 fl6.flowi6_tun_key.tun_id = 0;
4438                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4439
4440                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4441         }
4442
4443         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4444                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4445
4446         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4447                 switch (f6i->fib6_type) {
4448                 case RTN_BLACKHOLE:
4449                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4450                 case RTN_UNREACHABLE:
4451                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4452                 case RTN_PROHIBIT:
4453                         return BPF_FIB_LKUP_RET_PROHIBIT;
4454                 default:
4455                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4456                 }
4457         }
4458
4459         if (f6i->fib6_type != RTN_UNICAST)
4460                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4461
4462         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4463                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4464                                                        fl6.flowi6_oif, NULL,
4465                                                        strict);
4466
4467         if (check_mtu) {
4468                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4469                 if (params->tot_len > mtu)
4470                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4471         }
4472
4473         if (f6i->fib6_nh.nh_lwtstate)
4474                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4475
4476         if (f6i->fib6_flags & RTF_GATEWAY)
4477                 *dst = f6i->fib6_nh.nh_gw;
4478
4479         dev = f6i->fib6_nh.nh_dev;
4480         params->rt_metric = f6i->fib6_metric;
4481
4482         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4483          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4484          * because we need to get nd_tbl via the stub
4485          */
4486         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4487                                       ndisc_hashfn, dst, dev);
4488         if (!neigh)
4489                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4490
4491         return bpf_fib_set_fwd_params(params, neigh, dev);
4492 }
4493 #endif
4494
4495 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4496            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4497 {
4498         if (plen < sizeof(*params))
4499                 return -EINVAL;
4500
4501         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4502                 return -EINVAL;
4503
4504         switch (params->family) {
4505 #if IS_ENABLED(CONFIG_INET)
4506         case AF_INET:
4507                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4508                                            flags, true);
4509 #endif
4510 #if IS_ENABLED(CONFIG_IPV6)
4511         case AF_INET6:
4512                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4513                                            flags, true);
4514 #endif
4515         }
4516         return -EAFNOSUPPORT;
4517 }
4518
4519 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4520         .func           = bpf_xdp_fib_lookup,
4521         .gpl_only       = true,
4522         .ret_type       = RET_INTEGER,
4523         .arg1_type      = ARG_PTR_TO_CTX,
4524         .arg2_type      = ARG_PTR_TO_MEM,
4525         .arg3_type      = ARG_CONST_SIZE,
4526         .arg4_type      = ARG_ANYTHING,
4527 };
4528
4529 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4530            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4531 {
4532         struct net *net = dev_net(skb->dev);
4533         int rc = -EAFNOSUPPORT;
4534
4535         if (plen < sizeof(*params))
4536                 return -EINVAL;
4537
4538         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4539                 return -EINVAL;
4540
4541         switch (params->family) {
4542 #if IS_ENABLED(CONFIG_INET)
4543         case AF_INET:
4544                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4545                 break;
4546 #endif
4547 #if IS_ENABLED(CONFIG_IPV6)
4548         case AF_INET6:
4549                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4550                 break;
4551 #endif
4552         }
4553
4554         if (!rc) {
4555                 struct net_device *dev;
4556
4557                 dev = dev_get_by_index_rcu(net, params->ifindex);
4558                 if (!is_skb_forwardable(dev, skb))
4559                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4560         }
4561
4562         return rc;
4563 }
4564
4565 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4566         .func           = bpf_skb_fib_lookup,
4567         .gpl_only       = true,
4568         .ret_type       = RET_INTEGER,
4569         .arg1_type      = ARG_PTR_TO_CTX,
4570         .arg2_type      = ARG_PTR_TO_MEM,
4571         .arg3_type      = ARG_CONST_SIZE,
4572         .arg4_type      = ARG_ANYTHING,
4573 };
4574
4575 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4576 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4577 {
4578         int err;
4579         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4580
4581         if (!seg6_validate_srh(srh, len))
4582                 return -EINVAL;
4583
4584         switch (type) {
4585         case BPF_LWT_ENCAP_SEG6_INLINE:
4586                 if (skb->protocol != htons(ETH_P_IPV6))
4587                         return -EBADMSG;
4588
4589                 err = seg6_do_srh_inline(skb, srh);
4590                 break;
4591         case BPF_LWT_ENCAP_SEG6:
4592                 skb_reset_inner_headers(skb);
4593                 skb->encapsulation = 1;
4594                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4595                 break;
4596         default:
4597                 return -EINVAL;
4598         }
4599
4600         bpf_compute_data_pointers(skb);
4601         if (err)
4602                 return err;
4603
4604         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4605         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4606
4607         return seg6_lookup_nexthop(skb, NULL, 0);
4608 }
4609 #endif /* CONFIG_IPV6_SEG6_BPF */
4610
4611 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4612            u32, len)
4613 {
4614         switch (type) {
4615 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4616         case BPF_LWT_ENCAP_SEG6:
4617         case BPF_LWT_ENCAP_SEG6_INLINE:
4618                 return bpf_push_seg6_encap(skb, type, hdr, len);
4619 #endif
4620         default:
4621                 return -EINVAL;
4622         }
4623 }
4624
4625 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4626         .func           = bpf_lwt_push_encap,
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 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4636 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4637            const void *, from, u32, len)
4638 {
4639         struct seg6_bpf_srh_state *srh_state =
4640                 this_cpu_ptr(&seg6_bpf_srh_states);
4641         struct ipv6_sr_hdr *srh = srh_state->srh;
4642         void *srh_tlvs, *srh_end, *ptr;
4643         int srhoff = 0;
4644
4645         if (srh == NULL)
4646                 return -EINVAL;
4647
4648         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4649         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4650
4651         ptr = skb->data + offset;
4652         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4653                 srh_state->valid = false;
4654         else if (ptr < (void *)&srh->flags ||
4655                  ptr + len > (void *)&srh->segments)
4656                 return -EFAULT;
4657
4658         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4659                 return -EFAULT;
4660         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4661                 return -EINVAL;
4662         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4663
4664         memcpy(skb->data + offset, from, len);
4665         return 0;
4666 }
4667
4668 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4669         .func           = bpf_lwt_seg6_store_bytes,
4670         .gpl_only       = false,
4671         .ret_type       = RET_INTEGER,
4672         .arg1_type      = ARG_PTR_TO_CTX,
4673         .arg2_type      = ARG_ANYTHING,
4674         .arg3_type      = ARG_PTR_TO_MEM,
4675         .arg4_type      = ARG_CONST_SIZE
4676 };
4677
4678 static void bpf_update_srh_state(struct sk_buff *skb)
4679 {
4680         struct seg6_bpf_srh_state *srh_state =
4681                 this_cpu_ptr(&seg6_bpf_srh_states);
4682         int srhoff = 0;
4683
4684         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4685                 srh_state->srh = NULL;
4686         } else {
4687                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4688                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4689                 srh_state->valid = true;
4690         }
4691 }
4692
4693 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4694            u32, action, void *, param, u32, param_len)
4695 {
4696         struct seg6_bpf_srh_state *srh_state =
4697                 this_cpu_ptr(&seg6_bpf_srh_states);
4698         int hdroff = 0;
4699         int err;
4700
4701         switch (action) {
4702         case SEG6_LOCAL_ACTION_END_X:
4703                 if (!seg6_bpf_has_valid_srh(skb))
4704                         return -EBADMSG;
4705                 if (param_len != sizeof(struct in6_addr))
4706                         return -EINVAL;
4707                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4708         case SEG6_LOCAL_ACTION_END_T:
4709                 if (!seg6_bpf_has_valid_srh(skb))
4710                         return -EBADMSG;
4711                 if (param_len != sizeof(int))
4712                         return -EINVAL;
4713                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4714         case SEG6_LOCAL_ACTION_END_DT6:
4715                 if (!seg6_bpf_has_valid_srh(skb))
4716                         return -EBADMSG;
4717                 if (param_len != sizeof(int))
4718                         return -EINVAL;
4719
4720                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4721                         return -EBADMSG;
4722                 if (!pskb_pull(skb, hdroff))
4723                         return -EBADMSG;
4724
4725                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4726                 skb_reset_network_header(skb);
4727                 skb_reset_transport_header(skb);
4728                 skb->encapsulation = 0;
4729
4730                 bpf_compute_data_pointers(skb);
4731                 bpf_update_srh_state(skb);
4732                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4733         case SEG6_LOCAL_ACTION_END_B6:
4734                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4735                         return -EBADMSG;
4736                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4737                                           param, param_len);
4738                 if (!err)
4739                         bpf_update_srh_state(skb);
4740
4741                 return err;
4742         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4743                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4744                         return -EBADMSG;
4745                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4746                                           param, param_len);
4747                 if (!err)
4748                         bpf_update_srh_state(skb);
4749
4750                 return err;
4751         default:
4752                 return -EINVAL;
4753         }
4754 }
4755
4756 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4757         .func           = bpf_lwt_seg6_action,
4758         .gpl_only       = false,
4759         .ret_type       = RET_INTEGER,
4760         .arg1_type      = ARG_PTR_TO_CTX,
4761         .arg2_type      = ARG_ANYTHING,
4762         .arg3_type      = ARG_PTR_TO_MEM,
4763         .arg4_type      = ARG_CONST_SIZE
4764 };
4765
4766 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4767            s32, len)
4768 {
4769         struct seg6_bpf_srh_state *srh_state =
4770                 this_cpu_ptr(&seg6_bpf_srh_states);
4771         struct ipv6_sr_hdr *srh = srh_state->srh;
4772         void *srh_end, *srh_tlvs, *ptr;
4773         struct ipv6hdr *hdr;
4774         int srhoff = 0;
4775         int ret;
4776
4777         if (unlikely(srh == NULL))
4778                 return -EINVAL;
4779
4780         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4781                         ((srh->first_segment + 1) << 4));
4782         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4783                         srh_state->hdrlen);
4784         ptr = skb->data + offset;
4785
4786         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4787                 return -EFAULT;
4788         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4789                 return -EFAULT;
4790
4791         if (len > 0) {
4792                 ret = skb_cow_head(skb, len);
4793                 if (unlikely(ret < 0))
4794                         return ret;
4795
4796                 ret = bpf_skb_net_hdr_push(skb, offset, len);
4797         } else {
4798                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4799         }
4800
4801         bpf_compute_data_pointers(skb);
4802         if (unlikely(ret < 0))
4803                 return ret;
4804
4805         hdr = (struct ipv6hdr *)skb->data;
4806         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4807
4808         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4809                 return -EINVAL;
4810         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4811         srh_state->hdrlen += len;
4812         srh_state->valid = false;
4813         return 0;
4814 }
4815
4816 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4817         .func           = bpf_lwt_seg6_adjust_srh,
4818         .gpl_only       = false,
4819         .ret_type       = RET_INTEGER,
4820         .arg1_type      = ARG_PTR_TO_CTX,
4821         .arg2_type      = ARG_ANYTHING,
4822         .arg3_type      = ARG_ANYTHING,
4823 };
4824 #endif /* CONFIG_IPV6_SEG6_BPF */
4825
4826 #ifdef CONFIG_INET
4827 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple,
4828                               struct sk_buff *skb, u8 family, u8 proto)
4829 {
4830         bool refcounted = false;
4831         struct sock *sk = NULL;
4832         int dif = 0;
4833
4834         if (skb->dev)
4835                 dif = skb->dev->ifindex;
4836
4837         if (family == AF_INET) {
4838                 __be32 src4 = tuple->ipv4.saddr;
4839                 __be32 dst4 = tuple->ipv4.daddr;
4840                 int sdif = inet_sdif(skb);
4841
4842                 if (proto == IPPROTO_TCP)
4843                         sk = __inet_lookup(net, &tcp_hashinfo, skb, 0,
4844                                            src4, tuple->ipv4.sport,
4845                                            dst4, tuple->ipv4.dport,
4846                                            dif, sdif, &refcounted);
4847                 else
4848                         sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport,
4849                                                dst4, tuple->ipv4.dport,
4850                                                dif, sdif, &udp_table, skb);
4851 #if IS_ENABLED(CONFIG_IPV6)
4852         } else {
4853                 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr;
4854                 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr;
4855                 u16 hnum = ntohs(tuple->ipv6.dport);
4856                 int sdif = inet6_sdif(skb);
4857
4858                 if (proto == IPPROTO_TCP)
4859                         sk = __inet6_lookup(net, &tcp_hashinfo, skb, 0,
4860                                             src6, tuple->ipv6.sport,
4861                                             dst6, hnum,
4862                                             dif, sdif, &refcounted);
4863                 else if (likely(ipv6_bpf_stub))
4864                         sk = ipv6_bpf_stub->udp6_lib_lookup(net,
4865                                                             src6, tuple->ipv6.sport,
4866                                                             dst6, hnum,
4867                                                             dif, sdif,
4868                                                             &udp_table, skb);
4869 #endif
4870         }
4871
4872         if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) {
4873                 WARN_ONCE(1, "Found non-RCU, unreferenced socket!");
4874                 sk = NULL;
4875         }
4876         return sk;
4877 }
4878
4879 /* bpf_sk_lookup performs the core lookup for different types of sockets,
4880  * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE.
4881  * Returns the socket as an 'unsigned long' to simplify the casting in the
4882  * callers to satisfy BPF_CALL declarations.
4883  */
4884 static unsigned long
4885 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len,
4886               u8 proto, u64 netns_id, u64 flags)
4887 {
4888         struct net *caller_net;
4889         struct sock *sk = NULL;
4890         u8 family = AF_UNSPEC;
4891         struct net *net;
4892
4893         family = len == sizeof(tuple->ipv4) ? AF_INET : AF_INET6;
4894         if (unlikely(family == AF_UNSPEC || netns_id > U32_MAX || flags))
4895                 goto out;
4896
4897         if (skb->dev)
4898                 caller_net = dev_net(skb->dev);
4899         else
4900                 caller_net = sock_net(skb->sk);
4901         if (netns_id) {
4902                 net = get_net_ns_by_id(caller_net, netns_id);
4903                 if (unlikely(!net))
4904                         goto out;
4905                 sk = sk_lookup(net, tuple, skb, family, proto);
4906                 put_net(net);
4907         } else {
4908                 net = caller_net;
4909                 sk = sk_lookup(net, tuple, skb, family, proto);
4910         }
4911
4912         if (sk)
4913                 sk = sk_to_full_sk(sk);
4914 out:
4915         return (unsigned long) sk;
4916 }
4917
4918 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb,
4919            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
4920 {
4921         return bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, netns_id, flags);
4922 }
4923
4924 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = {
4925         .func           = bpf_sk_lookup_tcp,
4926         .gpl_only       = false,
4927         .pkt_access     = true,
4928         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
4929         .arg1_type      = ARG_PTR_TO_CTX,
4930         .arg2_type      = ARG_PTR_TO_MEM,
4931         .arg3_type      = ARG_CONST_SIZE,
4932         .arg4_type      = ARG_ANYTHING,
4933         .arg5_type      = ARG_ANYTHING,
4934 };
4935
4936 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb,
4937            struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags)
4938 {
4939         return bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, netns_id, flags);
4940 }
4941
4942 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = {
4943         .func           = bpf_sk_lookup_udp,
4944         .gpl_only       = false,
4945         .pkt_access     = true,
4946         .ret_type       = RET_PTR_TO_SOCKET_OR_NULL,
4947         .arg1_type      = ARG_PTR_TO_CTX,
4948         .arg2_type      = ARG_PTR_TO_MEM,
4949         .arg3_type      = ARG_CONST_SIZE,
4950         .arg4_type      = ARG_ANYTHING,
4951         .arg5_type      = ARG_ANYTHING,
4952 };
4953
4954 BPF_CALL_1(bpf_sk_release, struct sock *, sk)
4955 {
4956         if (!sock_flag(sk, SOCK_RCU_FREE))
4957                 sock_gen_put(sk);
4958         return 0;
4959 }
4960
4961 static const struct bpf_func_proto bpf_sk_release_proto = {
4962         .func           = bpf_sk_release,
4963         .gpl_only       = false,
4964         .ret_type       = RET_INTEGER,
4965         .arg1_type      = ARG_PTR_TO_SOCKET,
4966 };
4967 #endif /* CONFIG_INET */
4968
4969 bool bpf_helper_changes_pkt_data(void *func)
4970 {
4971         if (func == bpf_skb_vlan_push ||
4972             func == bpf_skb_vlan_pop ||
4973             func == bpf_skb_store_bytes ||
4974             func == bpf_skb_change_proto ||
4975             func == bpf_skb_change_head ||
4976             func == sk_skb_change_head ||
4977             func == bpf_skb_change_tail ||
4978             func == sk_skb_change_tail ||
4979             func == bpf_skb_adjust_room ||
4980             func == bpf_skb_pull_data ||
4981             func == sk_skb_pull_data ||
4982             func == bpf_clone_redirect ||
4983             func == bpf_l3_csum_replace ||
4984             func == bpf_l4_csum_replace ||
4985             func == bpf_xdp_adjust_head ||
4986             func == bpf_xdp_adjust_meta ||
4987             func == bpf_msg_pull_data ||
4988             func == bpf_msg_push_data ||
4989             func == bpf_xdp_adjust_tail ||
4990 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4991             func == bpf_lwt_seg6_store_bytes ||
4992             func == bpf_lwt_seg6_adjust_srh ||
4993             func == bpf_lwt_seg6_action ||
4994 #endif
4995             func == bpf_lwt_push_encap)
4996                 return true;
4997
4998         return false;
4999 }
5000
5001 static const struct bpf_func_proto *
5002 bpf_base_func_proto(enum bpf_func_id func_id)
5003 {
5004         switch (func_id) {
5005         case BPF_FUNC_map_lookup_elem:
5006                 return &bpf_map_lookup_elem_proto;
5007         case BPF_FUNC_map_update_elem:
5008                 return &bpf_map_update_elem_proto;
5009         case BPF_FUNC_map_delete_elem:
5010                 return &bpf_map_delete_elem_proto;
5011         case BPF_FUNC_map_push_elem:
5012                 return &bpf_map_push_elem_proto;
5013         case BPF_FUNC_map_pop_elem:
5014                 return &bpf_map_pop_elem_proto;
5015         case BPF_FUNC_map_peek_elem:
5016                 return &bpf_map_peek_elem_proto;
5017         case BPF_FUNC_get_prandom_u32:
5018                 return &bpf_get_prandom_u32_proto;
5019         case BPF_FUNC_get_smp_processor_id:
5020                 return &bpf_get_raw_smp_processor_id_proto;
5021         case BPF_FUNC_get_numa_node_id:
5022                 return &bpf_get_numa_node_id_proto;
5023         case BPF_FUNC_tail_call:
5024                 return &bpf_tail_call_proto;
5025         case BPF_FUNC_ktime_get_ns:
5026                 return &bpf_ktime_get_ns_proto;
5027         case BPF_FUNC_trace_printk:
5028                 if (capable(CAP_SYS_ADMIN))
5029                         return bpf_get_trace_printk_proto();
5030                 /* else: fall through */
5031         default:
5032                 return NULL;
5033         }
5034 }
5035
5036 static const struct bpf_func_proto *
5037 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5038 {
5039         switch (func_id) {
5040         /* inet and inet6 sockets are created in a process
5041          * context so there is always a valid uid/gid
5042          */
5043         case BPF_FUNC_get_current_uid_gid:
5044                 return &bpf_get_current_uid_gid_proto;
5045         case BPF_FUNC_get_local_storage:
5046                 return &bpf_get_local_storage_proto;
5047         default:
5048                 return bpf_base_func_proto(func_id);
5049         }
5050 }
5051
5052 static const struct bpf_func_proto *
5053 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5054 {
5055         switch (func_id) {
5056         /* inet and inet6 sockets are created in a process
5057          * context so there is always a valid uid/gid
5058          */
5059         case BPF_FUNC_get_current_uid_gid:
5060                 return &bpf_get_current_uid_gid_proto;
5061         case BPF_FUNC_bind:
5062                 switch (prog->expected_attach_type) {
5063                 case BPF_CGROUP_INET4_CONNECT:
5064                 case BPF_CGROUP_INET6_CONNECT:
5065                         return &bpf_bind_proto;
5066                 default:
5067                         return NULL;
5068                 }
5069         case BPF_FUNC_get_socket_cookie:
5070                 return &bpf_get_socket_cookie_sock_addr_proto;
5071         case BPF_FUNC_get_local_storage:
5072                 return &bpf_get_local_storage_proto;
5073         default:
5074                 return bpf_base_func_proto(func_id);
5075         }
5076 }
5077
5078 static const struct bpf_func_proto *
5079 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5080 {
5081         switch (func_id) {
5082         case BPF_FUNC_skb_load_bytes:
5083                 return &bpf_skb_load_bytes_proto;
5084         case BPF_FUNC_skb_load_bytes_relative:
5085                 return &bpf_skb_load_bytes_relative_proto;
5086         case BPF_FUNC_get_socket_cookie:
5087                 return &bpf_get_socket_cookie_proto;
5088         case BPF_FUNC_get_socket_uid:
5089                 return &bpf_get_socket_uid_proto;
5090         default:
5091                 return bpf_base_func_proto(func_id);
5092         }
5093 }
5094
5095 static const struct bpf_func_proto *
5096 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5097 {
5098         switch (func_id) {
5099         case BPF_FUNC_get_local_storage:
5100                 return &bpf_get_local_storage_proto;
5101         default:
5102                 return sk_filter_func_proto(func_id, prog);
5103         }
5104 }
5105
5106 static const struct bpf_func_proto *
5107 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5108 {
5109         switch (func_id) {
5110         case BPF_FUNC_skb_store_bytes:
5111                 return &bpf_skb_store_bytes_proto;
5112         case BPF_FUNC_skb_load_bytes:
5113                 return &bpf_skb_load_bytes_proto;
5114         case BPF_FUNC_skb_load_bytes_relative:
5115                 return &bpf_skb_load_bytes_relative_proto;
5116         case BPF_FUNC_skb_pull_data:
5117                 return &bpf_skb_pull_data_proto;
5118         case BPF_FUNC_csum_diff:
5119                 return &bpf_csum_diff_proto;
5120         case BPF_FUNC_csum_update:
5121                 return &bpf_csum_update_proto;
5122         case BPF_FUNC_l3_csum_replace:
5123                 return &bpf_l3_csum_replace_proto;
5124         case BPF_FUNC_l4_csum_replace:
5125                 return &bpf_l4_csum_replace_proto;
5126         case BPF_FUNC_clone_redirect:
5127                 return &bpf_clone_redirect_proto;
5128         case BPF_FUNC_get_cgroup_classid:
5129                 return &bpf_get_cgroup_classid_proto;
5130         case BPF_FUNC_skb_vlan_push:
5131                 return &bpf_skb_vlan_push_proto;
5132         case BPF_FUNC_skb_vlan_pop:
5133                 return &bpf_skb_vlan_pop_proto;
5134         case BPF_FUNC_skb_change_proto:
5135                 return &bpf_skb_change_proto_proto;
5136         case BPF_FUNC_skb_change_type:
5137                 return &bpf_skb_change_type_proto;
5138         case BPF_FUNC_skb_adjust_room:
5139                 return &bpf_skb_adjust_room_proto;
5140         case BPF_FUNC_skb_change_tail:
5141                 return &bpf_skb_change_tail_proto;
5142         case BPF_FUNC_skb_get_tunnel_key:
5143                 return &bpf_skb_get_tunnel_key_proto;
5144         case BPF_FUNC_skb_set_tunnel_key:
5145                 return bpf_get_skb_set_tunnel_proto(func_id);
5146         case BPF_FUNC_skb_get_tunnel_opt:
5147                 return &bpf_skb_get_tunnel_opt_proto;
5148         case BPF_FUNC_skb_set_tunnel_opt:
5149                 return bpf_get_skb_set_tunnel_proto(func_id);
5150         case BPF_FUNC_redirect:
5151                 return &bpf_redirect_proto;
5152         case BPF_FUNC_get_route_realm:
5153                 return &bpf_get_route_realm_proto;
5154         case BPF_FUNC_get_hash_recalc:
5155                 return &bpf_get_hash_recalc_proto;
5156         case BPF_FUNC_set_hash_invalid:
5157                 return &bpf_set_hash_invalid_proto;
5158         case BPF_FUNC_set_hash:
5159                 return &bpf_set_hash_proto;
5160         case BPF_FUNC_perf_event_output:
5161                 return &bpf_skb_event_output_proto;
5162         case BPF_FUNC_get_smp_processor_id:
5163                 return &bpf_get_smp_processor_id_proto;
5164         case BPF_FUNC_skb_under_cgroup:
5165                 return &bpf_skb_under_cgroup_proto;
5166         case BPF_FUNC_get_socket_cookie:
5167                 return &bpf_get_socket_cookie_proto;
5168         case BPF_FUNC_get_socket_uid:
5169                 return &bpf_get_socket_uid_proto;
5170         case BPF_FUNC_fib_lookup:
5171                 return &bpf_skb_fib_lookup_proto;
5172 #ifdef CONFIG_XFRM
5173         case BPF_FUNC_skb_get_xfrm_state:
5174                 return &bpf_skb_get_xfrm_state_proto;
5175 #endif
5176 #ifdef CONFIG_SOCK_CGROUP_DATA
5177         case BPF_FUNC_skb_cgroup_id:
5178                 return &bpf_skb_cgroup_id_proto;
5179         case BPF_FUNC_skb_ancestor_cgroup_id:
5180                 return &bpf_skb_ancestor_cgroup_id_proto;
5181 #endif
5182 #ifdef CONFIG_INET
5183         case BPF_FUNC_sk_lookup_tcp:
5184                 return &bpf_sk_lookup_tcp_proto;
5185         case BPF_FUNC_sk_lookup_udp:
5186                 return &bpf_sk_lookup_udp_proto;
5187         case BPF_FUNC_sk_release:
5188                 return &bpf_sk_release_proto;
5189 #endif
5190         default:
5191                 return bpf_base_func_proto(func_id);
5192         }
5193 }
5194
5195 static const struct bpf_func_proto *
5196 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5197 {
5198         switch (func_id) {
5199         case BPF_FUNC_perf_event_output:
5200                 return &bpf_xdp_event_output_proto;
5201         case BPF_FUNC_get_smp_processor_id:
5202                 return &bpf_get_smp_processor_id_proto;
5203         case BPF_FUNC_csum_diff:
5204                 return &bpf_csum_diff_proto;
5205         case BPF_FUNC_xdp_adjust_head:
5206                 return &bpf_xdp_adjust_head_proto;
5207         case BPF_FUNC_xdp_adjust_meta:
5208                 return &bpf_xdp_adjust_meta_proto;
5209         case BPF_FUNC_redirect:
5210                 return &bpf_xdp_redirect_proto;
5211         case BPF_FUNC_redirect_map:
5212                 return &bpf_xdp_redirect_map_proto;
5213         case BPF_FUNC_xdp_adjust_tail:
5214                 return &bpf_xdp_adjust_tail_proto;
5215         case BPF_FUNC_fib_lookup:
5216                 return &bpf_xdp_fib_lookup_proto;
5217         default:
5218                 return bpf_base_func_proto(func_id);
5219         }
5220 }
5221
5222 const struct bpf_func_proto bpf_sock_map_update_proto __weak;
5223 const struct bpf_func_proto bpf_sock_hash_update_proto __weak;
5224
5225 static const struct bpf_func_proto *
5226 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5227 {
5228         switch (func_id) {
5229         case BPF_FUNC_setsockopt:
5230                 return &bpf_setsockopt_proto;
5231         case BPF_FUNC_getsockopt:
5232                 return &bpf_getsockopt_proto;
5233         case BPF_FUNC_sock_ops_cb_flags_set:
5234                 return &bpf_sock_ops_cb_flags_set_proto;
5235         case BPF_FUNC_sock_map_update:
5236                 return &bpf_sock_map_update_proto;
5237         case BPF_FUNC_sock_hash_update:
5238                 return &bpf_sock_hash_update_proto;
5239         case BPF_FUNC_get_socket_cookie:
5240                 return &bpf_get_socket_cookie_sock_ops_proto;
5241         case BPF_FUNC_get_local_storage:
5242                 return &bpf_get_local_storage_proto;
5243         default:
5244                 return bpf_base_func_proto(func_id);
5245         }
5246 }
5247
5248 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak;
5249 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak;
5250
5251 static const struct bpf_func_proto *
5252 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5253 {
5254         switch (func_id) {
5255         case BPF_FUNC_msg_redirect_map:
5256                 return &bpf_msg_redirect_map_proto;
5257         case BPF_FUNC_msg_redirect_hash:
5258                 return &bpf_msg_redirect_hash_proto;
5259         case BPF_FUNC_msg_apply_bytes:
5260                 return &bpf_msg_apply_bytes_proto;
5261         case BPF_FUNC_msg_cork_bytes:
5262                 return &bpf_msg_cork_bytes_proto;
5263         case BPF_FUNC_msg_pull_data:
5264                 return &bpf_msg_pull_data_proto;
5265         case BPF_FUNC_msg_push_data:
5266                 return &bpf_msg_push_data_proto;
5267         default:
5268                 return bpf_base_func_proto(func_id);
5269         }
5270 }
5271
5272 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak;
5273 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak;
5274
5275 static const struct bpf_func_proto *
5276 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5277 {
5278         switch (func_id) {
5279         case BPF_FUNC_skb_store_bytes:
5280                 return &bpf_skb_store_bytes_proto;
5281         case BPF_FUNC_skb_load_bytes:
5282                 return &bpf_skb_load_bytes_proto;
5283         case BPF_FUNC_skb_pull_data:
5284                 return &sk_skb_pull_data_proto;
5285         case BPF_FUNC_skb_change_tail:
5286                 return &sk_skb_change_tail_proto;
5287         case BPF_FUNC_skb_change_head:
5288                 return &sk_skb_change_head_proto;
5289         case BPF_FUNC_get_socket_cookie:
5290                 return &bpf_get_socket_cookie_proto;
5291         case BPF_FUNC_get_socket_uid:
5292                 return &bpf_get_socket_uid_proto;
5293         case BPF_FUNC_sk_redirect_map:
5294                 return &bpf_sk_redirect_map_proto;
5295         case BPF_FUNC_sk_redirect_hash:
5296                 return &bpf_sk_redirect_hash_proto;
5297 #ifdef CONFIG_INET
5298         case BPF_FUNC_sk_lookup_tcp:
5299                 return &bpf_sk_lookup_tcp_proto;
5300         case BPF_FUNC_sk_lookup_udp:
5301                 return &bpf_sk_lookup_udp_proto;
5302         case BPF_FUNC_sk_release:
5303                 return &bpf_sk_release_proto;
5304 #endif
5305         default:
5306                 return bpf_base_func_proto(func_id);
5307         }
5308 }
5309
5310 static const struct bpf_func_proto *
5311 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5312 {
5313         switch (func_id) {
5314         case BPF_FUNC_skb_load_bytes:
5315                 return &bpf_skb_load_bytes_proto;
5316         default:
5317                 return bpf_base_func_proto(func_id);
5318         }
5319 }
5320
5321 static const struct bpf_func_proto *
5322 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5323 {
5324         switch (func_id) {
5325         case BPF_FUNC_skb_load_bytes:
5326                 return &bpf_skb_load_bytes_proto;
5327         case BPF_FUNC_skb_pull_data:
5328                 return &bpf_skb_pull_data_proto;
5329         case BPF_FUNC_csum_diff:
5330                 return &bpf_csum_diff_proto;
5331         case BPF_FUNC_get_cgroup_classid:
5332                 return &bpf_get_cgroup_classid_proto;
5333         case BPF_FUNC_get_route_realm:
5334                 return &bpf_get_route_realm_proto;
5335         case BPF_FUNC_get_hash_recalc:
5336                 return &bpf_get_hash_recalc_proto;
5337         case BPF_FUNC_perf_event_output:
5338                 return &bpf_skb_event_output_proto;
5339         case BPF_FUNC_get_smp_processor_id:
5340                 return &bpf_get_smp_processor_id_proto;
5341         case BPF_FUNC_skb_under_cgroup:
5342                 return &bpf_skb_under_cgroup_proto;
5343         default:
5344                 return bpf_base_func_proto(func_id);
5345         }
5346 }
5347
5348 static const struct bpf_func_proto *
5349 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5350 {
5351         switch (func_id) {
5352         case BPF_FUNC_lwt_push_encap:
5353                 return &bpf_lwt_push_encap_proto;
5354         default:
5355                 return lwt_out_func_proto(func_id, prog);
5356         }
5357 }
5358
5359 static const struct bpf_func_proto *
5360 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5361 {
5362         switch (func_id) {
5363         case BPF_FUNC_skb_get_tunnel_key:
5364                 return &bpf_skb_get_tunnel_key_proto;
5365         case BPF_FUNC_skb_set_tunnel_key:
5366                 return bpf_get_skb_set_tunnel_proto(func_id);
5367         case BPF_FUNC_skb_get_tunnel_opt:
5368                 return &bpf_skb_get_tunnel_opt_proto;
5369         case BPF_FUNC_skb_set_tunnel_opt:
5370                 return bpf_get_skb_set_tunnel_proto(func_id);
5371         case BPF_FUNC_redirect:
5372                 return &bpf_redirect_proto;
5373         case BPF_FUNC_clone_redirect:
5374                 return &bpf_clone_redirect_proto;
5375         case BPF_FUNC_skb_change_tail:
5376                 return &bpf_skb_change_tail_proto;
5377         case BPF_FUNC_skb_change_head:
5378                 return &bpf_skb_change_head_proto;
5379         case BPF_FUNC_skb_store_bytes:
5380                 return &bpf_skb_store_bytes_proto;
5381         case BPF_FUNC_csum_update:
5382                 return &bpf_csum_update_proto;
5383         case BPF_FUNC_l3_csum_replace:
5384                 return &bpf_l3_csum_replace_proto;
5385         case BPF_FUNC_l4_csum_replace:
5386                 return &bpf_l4_csum_replace_proto;
5387         case BPF_FUNC_set_hash_invalid:
5388                 return &bpf_set_hash_invalid_proto;
5389         default:
5390                 return lwt_out_func_proto(func_id, prog);
5391         }
5392 }
5393
5394 static const struct bpf_func_proto *
5395 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5396 {
5397         switch (func_id) {
5398 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5399         case BPF_FUNC_lwt_seg6_store_bytes:
5400                 return &bpf_lwt_seg6_store_bytes_proto;
5401         case BPF_FUNC_lwt_seg6_action:
5402                 return &bpf_lwt_seg6_action_proto;
5403         case BPF_FUNC_lwt_seg6_adjust_srh:
5404                 return &bpf_lwt_seg6_adjust_srh_proto;
5405 #endif
5406         default:
5407                 return lwt_out_func_proto(func_id, prog);
5408         }
5409 }
5410
5411 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5412                                     const struct bpf_prog *prog,
5413                                     struct bpf_insn_access_aux *info)
5414 {
5415         const int size_default = sizeof(__u32);
5416
5417         if (off < 0 || off >= sizeof(struct __sk_buff))
5418                 return false;
5419
5420         /* The verifier guarantees that size > 0. */
5421         if (off % size != 0)
5422                 return false;
5423
5424         switch (off) {
5425         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5426                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5427                         return false;
5428                 break;
5429         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5430         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5431         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5432         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5433         case bpf_ctx_range(struct __sk_buff, data):
5434         case bpf_ctx_range(struct __sk_buff, data_meta):
5435         case bpf_ctx_range(struct __sk_buff, data_end):
5436                 if (size != size_default)
5437                         return false;
5438                 break;
5439         case bpf_ctx_range(struct __sk_buff, flow_keys):
5440                 if (size != sizeof(struct bpf_flow_keys *))
5441                         return false;
5442                 break;
5443         default:
5444                 /* Only narrow read access allowed for now. */
5445                 if (type == BPF_WRITE) {
5446                         if (size != size_default)
5447                                 return false;
5448                 } else {
5449                         bpf_ctx_record_field_size(info, size_default);
5450                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5451                                 return false;
5452                 }
5453         }
5454
5455         return true;
5456 }
5457
5458 static bool sk_filter_is_valid_access(int off, int size,
5459                                       enum bpf_access_type type,
5460                                       const struct bpf_prog *prog,
5461                                       struct bpf_insn_access_aux *info)
5462 {
5463         switch (off) {
5464         case bpf_ctx_range(struct __sk_buff, tc_classid):
5465         case bpf_ctx_range(struct __sk_buff, data):
5466         case bpf_ctx_range(struct __sk_buff, data_meta):
5467         case bpf_ctx_range(struct __sk_buff, data_end):
5468         case bpf_ctx_range(struct __sk_buff, flow_keys):
5469         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5470                 return false;
5471         }
5472
5473         if (type == BPF_WRITE) {
5474                 switch (off) {
5475                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5476                         break;
5477                 default:
5478                         return false;
5479                 }
5480         }
5481
5482         return bpf_skb_is_valid_access(off, size, type, prog, info);
5483 }
5484
5485 static bool cg_skb_is_valid_access(int off, int size,
5486                                    enum bpf_access_type type,
5487                                    const struct bpf_prog *prog,
5488                                    struct bpf_insn_access_aux *info)
5489 {
5490         switch (off) {
5491         case bpf_ctx_range(struct __sk_buff, tc_classid):
5492         case bpf_ctx_range(struct __sk_buff, data_meta):
5493         case bpf_ctx_range(struct __sk_buff, flow_keys):
5494                 return false;
5495         case bpf_ctx_range(struct __sk_buff, data):
5496         case bpf_ctx_range(struct __sk_buff, data_end):
5497                 if (!capable(CAP_SYS_ADMIN))
5498                         return false;
5499                 break;
5500         }
5501
5502         if (type == BPF_WRITE) {
5503                 switch (off) {
5504                 case bpf_ctx_range(struct __sk_buff, mark):
5505                 case bpf_ctx_range(struct __sk_buff, priority):
5506                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5507                         break;
5508                 default:
5509                         return false;
5510                 }
5511         }
5512
5513         switch (off) {
5514         case bpf_ctx_range(struct __sk_buff, data):
5515                 info->reg_type = PTR_TO_PACKET;
5516                 break;
5517         case bpf_ctx_range(struct __sk_buff, data_end):
5518                 info->reg_type = PTR_TO_PACKET_END;
5519                 break;
5520         }
5521
5522         return bpf_skb_is_valid_access(off, size, type, prog, info);
5523 }
5524
5525 static bool lwt_is_valid_access(int off, int size,
5526                                 enum bpf_access_type type,
5527                                 const struct bpf_prog *prog,
5528                                 struct bpf_insn_access_aux *info)
5529 {
5530         switch (off) {
5531         case bpf_ctx_range(struct __sk_buff, tc_classid):
5532         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5533         case bpf_ctx_range(struct __sk_buff, data_meta):
5534         case bpf_ctx_range(struct __sk_buff, flow_keys):
5535                 return false;
5536         }
5537
5538         if (type == BPF_WRITE) {
5539                 switch (off) {
5540                 case bpf_ctx_range(struct __sk_buff, mark):
5541                 case bpf_ctx_range(struct __sk_buff, priority):
5542                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5543                         break;
5544                 default:
5545                         return false;
5546                 }
5547         }
5548
5549         switch (off) {
5550         case bpf_ctx_range(struct __sk_buff, data):
5551                 info->reg_type = PTR_TO_PACKET;
5552                 break;
5553         case bpf_ctx_range(struct __sk_buff, data_end):
5554                 info->reg_type = PTR_TO_PACKET_END;
5555                 break;
5556         }
5557
5558         return bpf_skb_is_valid_access(off, size, type, prog, info);
5559 }
5560
5561 /* Attach type specific accesses */
5562 static bool __sock_filter_check_attach_type(int off,
5563                                             enum bpf_access_type access_type,
5564                                             enum bpf_attach_type attach_type)
5565 {
5566         switch (off) {
5567         case offsetof(struct bpf_sock, bound_dev_if):
5568         case offsetof(struct bpf_sock, mark):
5569         case offsetof(struct bpf_sock, priority):
5570                 switch (attach_type) {
5571                 case BPF_CGROUP_INET_SOCK_CREATE:
5572                         goto full_access;
5573                 default:
5574                         return false;
5575                 }
5576         case bpf_ctx_range(struct bpf_sock, src_ip4):
5577                 switch (attach_type) {
5578                 case BPF_CGROUP_INET4_POST_BIND:
5579                         goto read_only;
5580                 default:
5581                         return false;
5582                 }
5583         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5584                 switch (attach_type) {
5585                 case BPF_CGROUP_INET6_POST_BIND:
5586                         goto read_only;
5587                 default:
5588                         return false;
5589                 }
5590         case bpf_ctx_range(struct bpf_sock, src_port):
5591                 switch (attach_type) {
5592                 case BPF_CGROUP_INET4_POST_BIND:
5593                 case BPF_CGROUP_INET6_POST_BIND:
5594                         goto read_only;
5595                 default:
5596                         return false;
5597                 }
5598         }
5599 read_only:
5600         return access_type == BPF_READ;
5601 full_access:
5602         return true;
5603 }
5604
5605 static bool __sock_filter_check_size(int off, int size,
5606                                      struct bpf_insn_access_aux *info)
5607 {
5608         const int size_default = sizeof(__u32);
5609
5610         switch (off) {
5611         case bpf_ctx_range(struct bpf_sock, src_ip4):
5612         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5613                 bpf_ctx_record_field_size(info, size_default);
5614                 return bpf_ctx_narrow_access_ok(off, size, size_default);
5615         }
5616
5617         return size == size_default;
5618 }
5619
5620 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type,
5621                               struct bpf_insn_access_aux *info)
5622 {
5623         if (off < 0 || off >= sizeof(struct bpf_sock))
5624                 return false;
5625         if (off % size != 0)
5626                 return false;
5627         if (!__sock_filter_check_size(off, size, info))
5628                 return false;
5629         return true;
5630 }
5631
5632 static bool sock_filter_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         if (!bpf_sock_is_valid_access(off, size, type, info))
5638                 return false;
5639         return __sock_filter_check_attach_type(off, type,
5640                                                prog->expected_attach_type);
5641 }
5642
5643 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write,
5644                              const struct bpf_prog *prog)
5645 {
5646         /* Neither direct read nor direct write requires any preliminary
5647          * action.
5648          */
5649         return 0;
5650 }
5651
5652 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5653                                 const struct bpf_prog *prog, int drop_verdict)
5654 {
5655         struct bpf_insn *insn = insn_buf;
5656
5657         if (!direct_write)
5658                 return 0;
5659
5660         /* if (!skb->cloned)
5661          *       goto start;
5662          *
5663          * (Fast-path, otherwise approximation that we might be
5664          *  a clone, do the rest in helper.)
5665          */
5666         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5667         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5668         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5669
5670         /* ret = bpf_skb_pull_data(skb, 0); */
5671         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5672         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5673         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5674                                BPF_FUNC_skb_pull_data);
5675         /* if (!ret)
5676          *      goto restore;
5677          * return TC_ACT_SHOT;
5678          */
5679         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5680         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5681         *insn++ = BPF_EXIT_INSN();
5682
5683         /* restore: */
5684         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5685         /* start: */
5686         *insn++ = prog->insnsi[0];
5687
5688         return insn - insn_buf;
5689 }
5690
5691 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5692                           struct bpf_insn *insn_buf)
5693 {
5694         bool indirect = BPF_MODE(orig->code) == BPF_IND;
5695         struct bpf_insn *insn = insn_buf;
5696
5697         /* We're guaranteed here that CTX is in R6. */
5698         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5699         if (!indirect) {
5700                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5701         } else {
5702                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5703                 if (orig->imm)
5704                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5705         }
5706
5707         switch (BPF_SIZE(orig->code)) {
5708         case BPF_B:
5709                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5710                 break;
5711         case BPF_H:
5712                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5713                 break;
5714         case BPF_W:
5715                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5716                 break;
5717         }
5718
5719         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5720         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5721         *insn++ = BPF_EXIT_INSN();
5722
5723         return insn - insn_buf;
5724 }
5725
5726 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5727                                const struct bpf_prog *prog)
5728 {
5729         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5730 }
5731
5732 static bool tc_cls_act_is_valid_access(int off, int size,
5733                                        enum bpf_access_type type,
5734                                        const struct bpf_prog *prog,
5735                                        struct bpf_insn_access_aux *info)
5736 {
5737         if (type == BPF_WRITE) {
5738                 switch (off) {
5739                 case bpf_ctx_range(struct __sk_buff, mark):
5740                 case bpf_ctx_range(struct __sk_buff, tc_index):
5741                 case bpf_ctx_range(struct __sk_buff, priority):
5742                 case bpf_ctx_range(struct __sk_buff, tc_classid):
5743                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5744                         break;
5745                 default:
5746                         return false;
5747                 }
5748         }
5749
5750         switch (off) {
5751         case bpf_ctx_range(struct __sk_buff, data):
5752                 info->reg_type = PTR_TO_PACKET;
5753                 break;
5754         case bpf_ctx_range(struct __sk_buff, data_meta):
5755                 info->reg_type = PTR_TO_PACKET_META;
5756                 break;
5757         case bpf_ctx_range(struct __sk_buff, data_end):
5758                 info->reg_type = PTR_TO_PACKET_END;
5759                 break;
5760         case bpf_ctx_range(struct __sk_buff, flow_keys):
5761         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5762                 return false;
5763         }
5764
5765         return bpf_skb_is_valid_access(off, size, type, prog, info);
5766 }
5767
5768 static bool __is_valid_xdp_access(int off, int size)
5769 {
5770         if (off < 0 || off >= sizeof(struct xdp_md))
5771                 return false;
5772         if (off % size != 0)
5773                 return false;
5774         if (size != sizeof(__u32))
5775                 return false;
5776
5777         return true;
5778 }
5779
5780 static bool xdp_is_valid_access(int off, int size,
5781                                 enum bpf_access_type type,
5782                                 const struct bpf_prog *prog,
5783                                 struct bpf_insn_access_aux *info)
5784 {
5785         if (type == BPF_WRITE) {
5786                 if (bpf_prog_is_dev_bound(prog->aux)) {
5787                         switch (off) {
5788                         case offsetof(struct xdp_md, rx_queue_index):
5789                                 return __is_valid_xdp_access(off, size);
5790                         }
5791                 }
5792                 return false;
5793         }
5794
5795         switch (off) {
5796         case offsetof(struct xdp_md, data):
5797                 info->reg_type = PTR_TO_PACKET;
5798                 break;
5799         case offsetof(struct xdp_md, data_meta):
5800                 info->reg_type = PTR_TO_PACKET_META;
5801                 break;
5802         case offsetof(struct xdp_md, data_end):
5803                 info->reg_type = PTR_TO_PACKET_END;
5804                 break;
5805         }
5806
5807         return __is_valid_xdp_access(off, size);
5808 }
5809
5810 void bpf_warn_invalid_xdp_action(u32 act)
5811 {
5812         const u32 act_max = XDP_REDIRECT;
5813
5814         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5815                   act > act_max ? "Illegal" : "Driver unsupported",
5816                   act);
5817 }
5818 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5819
5820 static bool sock_addr_is_valid_access(int off, int size,
5821                                       enum bpf_access_type type,
5822                                       const struct bpf_prog *prog,
5823                                       struct bpf_insn_access_aux *info)
5824 {
5825         const int size_default = sizeof(__u32);
5826
5827         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5828                 return false;
5829         if (off % size != 0)
5830                 return false;
5831
5832         /* Disallow access to IPv6 fields from IPv4 contex and vise
5833          * versa.
5834          */
5835         switch (off) {
5836         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5837                 switch (prog->expected_attach_type) {
5838                 case BPF_CGROUP_INET4_BIND:
5839                 case BPF_CGROUP_INET4_CONNECT:
5840                 case BPF_CGROUP_UDP4_SENDMSG:
5841                         break;
5842                 default:
5843                         return false;
5844                 }
5845                 break;
5846         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5847                 switch (prog->expected_attach_type) {
5848                 case BPF_CGROUP_INET6_BIND:
5849                 case BPF_CGROUP_INET6_CONNECT:
5850                 case BPF_CGROUP_UDP6_SENDMSG:
5851                         break;
5852                 default:
5853                         return false;
5854                 }
5855                 break;
5856         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5857                 switch (prog->expected_attach_type) {
5858                 case BPF_CGROUP_UDP4_SENDMSG:
5859                         break;
5860                 default:
5861                         return false;
5862                 }
5863                 break;
5864         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5865                                 msg_src_ip6[3]):
5866                 switch (prog->expected_attach_type) {
5867                 case BPF_CGROUP_UDP6_SENDMSG:
5868                         break;
5869                 default:
5870                         return false;
5871                 }
5872                 break;
5873         }
5874
5875         switch (off) {
5876         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5877         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5878         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5879         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5880                                 msg_src_ip6[3]):
5881                 /* Only narrow read access allowed for now. */
5882                 if (type == BPF_READ) {
5883                         bpf_ctx_record_field_size(info, size_default);
5884                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5885                                 return false;
5886                 } else {
5887                         if (size != size_default)
5888                                 return false;
5889                 }
5890                 break;
5891         case bpf_ctx_range(struct bpf_sock_addr, user_port):
5892                 if (size != size_default)
5893                         return false;
5894                 break;
5895         default:
5896                 if (type == BPF_READ) {
5897                         if (size != size_default)
5898                                 return false;
5899                 } else {
5900                         return false;
5901                 }
5902         }
5903
5904         return true;
5905 }
5906
5907 static bool sock_ops_is_valid_access(int off, int size,
5908                                      enum bpf_access_type type,
5909                                      const struct bpf_prog *prog,
5910                                      struct bpf_insn_access_aux *info)
5911 {
5912         const int size_default = sizeof(__u32);
5913
5914         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5915                 return false;
5916
5917         /* The verifier guarantees that size > 0. */
5918         if (off % size != 0)
5919                 return false;
5920
5921         if (type == BPF_WRITE) {
5922                 switch (off) {
5923                 case offsetof(struct bpf_sock_ops, reply):
5924                 case offsetof(struct bpf_sock_ops, sk_txhash):
5925                         if (size != size_default)
5926                                 return false;
5927                         break;
5928                 default:
5929                         return false;
5930                 }
5931         } else {
5932                 switch (off) {
5933                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5934                                         bytes_acked):
5935                         if (size != sizeof(__u64))
5936                                 return false;
5937                         break;
5938                 default:
5939                         if (size != size_default)
5940                                 return false;
5941                         break;
5942                 }
5943         }
5944
5945         return true;
5946 }
5947
5948 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5949                            const struct bpf_prog *prog)
5950 {
5951         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5952 }
5953
5954 static bool sk_skb_is_valid_access(int off, int size,
5955                                    enum bpf_access_type type,
5956                                    const struct bpf_prog *prog,
5957                                    struct bpf_insn_access_aux *info)
5958 {
5959         switch (off) {
5960         case bpf_ctx_range(struct __sk_buff, tc_classid):
5961         case bpf_ctx_range(struct __sk_buff, data_meta):
5962         case bpf_ctx_range(struct __sk_buff, flow_keys):
5963                 return false;
5964         }
5965
5966         if (type == BPF_WRITE) {
5967                 switch (off) {
5968                 case bpf_ctx_range(struct __sk_buff, tc_index):
5969                 case bpf_ctx_range(struct __sk_buff, priority):
5970                         break;
5971                 default:
5972                         return false;
5973                 }
5974         }
5975
5976         switch (off) {
5977         case bpf_ctx_range(struct __sk_buff, mark):
5978                 return false;
5979         case bpf_ctx_range(struct __sk_buff, data):
5980                 info->reg_type = PTR_TO_PACKET;
5981                 break;
5982         case bpf_ctx_range(struct __sk_buff, data_end):
5983                 info->reg_type = PTR_TO_PACKET_END;
5984                 break;
5985         }
5986
5987         return bpf_skb_is_valid_access(off, size, type, prog, info);
5988 }
5989
5990 static bool sk_msg_is_valid_access(int off, int size,
5991                                    enum bpf_access_type type,
5992                                    const struct bpf_prog *prog,
5993                                    struct bpf_insn_access_aux *info)
5994 {
5995         if (type == BPF_WRITE)
5996                 return false;
5997
5998         switch (off) {
5999         case offsetof(struct sk_msg_md, data):
6000                 info->reg_type = PTR_TO_PACKET;
6001                 if (size != sizeof(__u64))
6002                         return false;
6003                 break;
6004         case offsetof(struct sk_msg_md, data_end):
6005                 info->reg_type = PTR_TO_PACKET_END;
6006                 if (size != sizeof(__u64))
6007                         return false;
6008                 break;
6009         default:
6010                 if (size != sizeof(__u32))
6011                         return false;
6012         }
6013
6014         if (off < 0 || off >= sizeof(struct sk_msg_md))
6015                 return false;
6016         if (off % size != 0)
6017                 return false;
6018
6019         return true;
6020 }
6021
6022 static bool flow_dissector_is_valid_access(int off, int size,
6023                                            enum bpf_access_type type,
6024                                            const struct bpf_prog *prog,
6025                                            struct bpf_insn_access_aux *info)
6026 {
6027         if (type == BPF_WRITE) {
6028                 switch (off) {
6029                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
6030                         break;
6031                 default:
6032                         return false;
6033                 }
6034         }
6035
6036         switch (off) {
6037         case bpf_ctx_range(struct __sk_buff, data):
6038                 info->reg_type = PTR_TO_PACKET;
6039                 break;
6040         case bpf_ctx_range(struct __sk_buff, data_end):
6041                 info->reg_type = PTR_TO_PACKET_END;
6042                 break;
6043         case bpf_ctx_range(struct __sk_buff, flow_keys):
6044                 info->reg_type = PTR_TO_FLOW_KEYS;
6045                 break;
6046         case bpf_ctx_range(struct __sk_buff, tc_classid):
6047         case bpf_ctx_range(struct __sk_buff, data_meta):
6048         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
6049                 return false;
6050         }
6051
6052         return bpf_skb_is_valid_access(off, size, type, prog, info);
6053 }
6054
6055 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
6056                                   const struct bpf_insn *si,
6057                                   struct bpf_insn *insn_buf,
6058                                   struct bpf_prog *prog, u32 *target_size)
6059 {
6060         struct bpf_insn *insn = insn_buf;
6061         int off;
6062
6063         switch (si->off) {
6064         case offsetof(struct __sk_buff, len):
6065                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6066                                       bpf_target_off(struct sk_buff, len, 4,
6067                                                      target_size));
6068                 break;
6069
6070         case offsetof(struct __sk_buff, protocol):
6071                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6072                                       bpf_target_off(struct sk_buff, protocol, 2,
6073                                                      target_size));
6074                 break;
6075
6076         case offsetof(struct __sk_buff, vlan_proto):
6077                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6078                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
6079                                                      target_size));
6080                 break;
6081
6082         case offsetof(struct __sk_buff, priority):
6083                 if (type == BPF_WRITE)
6084                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6085                                               bpf_target_off(struct sk_buff, priority, 4,
6086                                                              target_size));
6087                 else
6088                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6089                                               bpf_target_off(struct sk_buff, priority, 4,
6090                                                              target_size));
6091                 break;
6092
6093         case offsetof(struct __sk_buff, ingress_ifindex):
6094                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6095                                       bpf_target_off(struct sk_buff, skb_iif, 4,
6096                                                      target_size));
6097                 break;
6098
6099         case offsetof(struct __sk_buff, ifindex):
6100                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6101                                       si->dst_reg, si->src_reg,
6102                                       offsetof(struct sk_buff, dev));
6103                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
6104                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6105                                       bpf_target_off(struct net_device, ifindex, 4,
6106                                                      target_size));
6107                 break;
6108
6109         case offsetof(struct __sk_buff, hash):
6110                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6111                                       bpf_target_off(struct sk_buff, hash, 4,
6112                                                      target_size));
6113                 break;
6114
6115         case offsetof(struct __sk_buff, mark):
6116                 if (type == BPF_WRITE)
6117                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6118                                               bpf_target_off(struct sk_buff, mark, 4,
6119                                                              target_size));
6120                 else
6121                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6122                                               bpf_target_off(struct sk_buff, mark, 4,
6123                                                              target_size));
6124                 break;
6125
6126         case offsetof(struct __sk_buff, pkt_type):
6127                 *target_size = 1;
6128                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
6129                                       PKT_TYPE_OFFSET());
6130                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
6131 #ifdef __BIG_ENDIAN_BITFIELD
6132                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
6133 #endif
6134                 break;
6135
6136         case offsetof(struct __sk_buff, queue_mapping):
6137                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6138                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
6139                                                      target_size));
6140                 break;
6141
6142         case offsetof(struct __sk_buff, vlan_present):
6143         case offsetof(struct __sk_buff, vlan_tci):
6144                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
6145
6146                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6147                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
6148                                                      target_size));
6149                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
6150                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
6151                                                 ~VLAN_TAG_PRESENT);
6152                 } else {
6153                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
6154                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
6155                 }
6156                 break;
6157
6158         case offsetof(struct __sk_buff, cb[0]) ...
6159              offsetofend(struct __sk_buff, cb[4]) - 1:
6160                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
6161                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
6162                               offsetof(struct qdisc_skb_cb, data)) %
6163                              sizeof(__u64));
6164
6165                 prog->cb_access = 1;
6166                 off  = si->off;
6167                 off -= offsetof(struct __sk_buff, cb[0]);
6168                 off += offsetof(struct sk_buff, cb);
6169                 off += offsetof(struct qdisc_skb_cb, data);
6170                 if (type == BPF_WRITE)
6171                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
6172                                               si->src_reg, off);
6173                 else
6174                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
6175                                               si->src_reg, off);
6176                 break;
6177
6178         case offsetof(struct __sk_buff, tc_classid):
6179                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
6180
6181                 off  = si->off;
6182                 off -= offsetof(struct __sk_buff, tc_classid);
6183                 off += offsetof(struct sk_buff, cb);
6184                 off += offsetof(struct qdisc_skb_cb, tc_classid);
6185                 *target_size = 2;
6186                 if (type == BPF_WRITE)
6187                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
6188                                               si->src_reg, off);
6189                 else
6190                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
6191                                               si->src_reg, off);
6192                 break;
6193
6194         case offsetof(struct __sk_buff, data):
6195                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
6196                                       si->dst_reg, si->src_reg,
6197                                       offsetof(struct sk_buff, data));
6198                 break;
6199
6200         case offsetof(struct __sk_buff, data_meta):
6201                 off  = si->off;
6202                 off -= offsetof(struct __sk_buff, data_meta);
6203                 off += offsetof(struct sk_buff, cb);
6204                 off += offsetof(struct bpf_skb_data_end, data_meta);
6205                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6206                                       si->src_reg, off);
6207                 break;
6208
6209         case offsetof(struct __sk_buff, data_end):
6210                 off  = si->off;
6211                 off -= offsetof(struct __sk_buff, data_end);
6212                 off += offsetof(struct sk_buff, cb);
6213                 off += offsetof(struct bpf_skb_data_end, data_end);
6214                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6215                                       si->src_reg, off);
6216                 break;
6217
6218         case offsetof(struct __sk_buff, tc_index):
6219 #ifdef CONFIG_NET_SCHED
6220                 if (type == BPF_WRITE)
6221                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
6222                                               bpf_target_off(struct sk_buff, tc_index, 2,
6223                                                              target_size));
6224                 else
6225                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6226                                               bpf_target_off(struct sk_buff, tc_index, 2,
6227                                                              target_size));
6228 #else
6229                 *target_size = 2;
6230                 if (type == BPF_WRITE)
6231                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
6232                 else
6233                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6234 #endif
6235                 break;
6236
6237         case offsetof(struct __sk_buff, napi_id):
6238 #if defined(CONFIG_NET_RX_BUSY_POLL)
6239                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6240                                       bpf_target_off(struct sk_buff, napi_id, 4,
6241                                                      target_size));
6242                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
6243                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6244 #else
6245                 *target_size = 4;
6246                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
6247 #endif
6248                 break;
6249         case offsetof(struct __sk_buff, family):
6250                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6251
6252                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6253                                       si->dst_reg, si->src_reg,
6254                                       offsetof(struct sk_buff, sk));
6255                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6256                                       bpf_target_off(struct sock_common,
6257                                                      skc_family,
6258                                                      2, target_size));
6259                 break;
6260         case offsetof(struct __sk_buff, remote_ip4):
6261                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6262
6263                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6264                                       si->dst_reg, si->src_reg,
6265                                       offsetof(struct sk_buff, sk));
6266                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6267                                       bpf_target_off(struct sock_common,
6268                                                      skc_daddr,
6269                                                      4, target_size));
6270                 break;
6271         case offsetof(struct __sk_buff, local_ip4):
6272                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6273                                           skc_rcv_saddr) != 4);
6274
6275                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6276                                       si->dst_reg, si->src_reg,
6277                                       offsetof(struct sk_buff, sk));
6278                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6279                                       bpf_target_off(struct sock_common,
6280                                                      skc_rcv_saddr,
6281                                                      4, target_size));
6282                 break;
6283         case offsetof(struct __sk_buff, remote_ip6[0]) ...
6284              offsetof(struct __sk_buff, remote_ip6[3]):
6285 #if IS_ENABLED(CONFIG_IPV6)
6286                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6287                                           skc_v6_daddr.s6_addr32[0]) != 4);
6288
6289                 off = si->off;
6290                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
6291
6292                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6293                                       si->dst_reg, si->src_reg,
6294                                       offsetof(struct sk_buff, sk));
6295                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6296                                       offsetof(struct sock_common,
6297                                                skc_v6_daddr.s6_addr32[0]) +
6298                                       off);
6299 #else
6300                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6301 #endif
6302                 break;
6303         case offsetof(struct __sk_buff, local_ip6[0]) ...
6304              offsetof(struct __sk_buff, local_ip6[3]):
6305 #if IS_ENABLED(CONFIG_IPV6)
6306                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6307                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6308
6309                 off = si->off;
6310                 off -= offsetof(struct __sk_buff, local_ip6[0]);
6311
6312                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6313                                       si->dst_reg, si->src_reg,
6314                                       offsetof(struct sk_buff, sk));
6315                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6316                                       offsetof(struct sock_common,
6317                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6318                                       off);
6319 #else
6320                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6321 #endif
6322                 break;
6323
6324         case offsetof(struct __sk_buff, remote_port):
6325                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6326
6327                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6328                                       si->dst_reg, si->src_reg,
6329                                       offsetof(struct sk_buff, sk));
6330                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6331                                       bpf_target_off(struct sock_common,
6332                                                      skc_dport,
6333                                                      2, target_size));
6334 #ifndef __BIG_ENDIAN_BITFIELD
6335                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6336 #endif
6337                 break;
6338
6339         case offsetof(struct __sk_buff, local_port):
6340                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6341
6342                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6343                                       si->dst_reg, si->src_reg,
6344                                       offsetof(struct sk_buff, sk));
6345                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6346                                       bpf_target_off(struct sock_common,
6347                                                      skc_num, 2, target_size));
6348                 break;
6349
6350         case offsetof(struct __sk_buff, flow_keys):
6351                 off  = si->off;
6352                 off -= offsetof(struct __sk_buff, flow_keys);
6353                 off += offsetof(struct sk_buff, cb);
6354                 off += offsetof(struct qdisc_skb_cb, flow_keys);
6355                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6356                                       si->src_reg, off);
6357                 break;
6358         }
6359
6360         return insn - insn_buf;
6361 }
6362
6363 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type,
6364                                 const struct bpf_insn *si,
6365                                 struct bpf_insn *insn_buf,
6366                                 struct bpf_prog *prog, u32 *target_size)
6367 {
6368         struct bpf_insn *insn = insn_buf;
6369         int off;
6370
6371         switch (si->off) {
6372         case offsetof(struct bpf_sock, bound_dev_if):
6373                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6374
6375                 if (type == BPF_WRITE)
6376                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6377                                         offsetof(struct sock, sk_bound_dev_if));
6378                 else
6379                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6380                                       offsetof(struct sock, sk_bound_dev_if));
6381                 break;
6382
6383         case offsetof(struct bpf_sock, mark):
6384                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6385
6386                 if (type == BPF_WRITE)
6387                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6388                                         offsetof(struct sock, sk_mark));
6389                 else
6390                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6391                                       offsetof(struct sock, sk_mark));
6392                 break;
6393
6394         case offsetof(struct bpf_sock, priority):
6395                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6396
6397                 if (type == BPF_WRITE)
6398                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6399                                         offsetof(struct sock, sk_priority));
6400                 else
6401                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6402                                       offsetof(struct sock, sk_priority));
6403                 break;
6404
6405         case offsetof(struct bpf_sock, family):
6406                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6407
6408                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6409                                       offsetof(struct sock, sk_family));
6410                 break;
6411
6412         case offsetof(struct bpf_sock, type):
6413                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6414                                       offsetof(struct sock, __sk_flags_offset));
6415                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6416                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6417                 break;
6418
6419         case offsetof(struct bpf_sock, protocol):
6420                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6421                                       offsetof(struct sock, __sk_flags_offset));
6422                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6423                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6424                 break;
6425
6426         case offsetof(struct bpf_sock, src_ip4):
6427                 *insn++ = BPF_LDX_MEM(
6428                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6429                         bpf_target_off(struct sock_common, skc_rcv_saddr,
6430                                        FIELD_SIZEOF(struct sock_common,
6431                                                     skc_rcv_saddr),
6432                                        target_size));
6433                 break;
6434
6435         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6436 #if IS_ENABLED(CONFIG_IPV6)
6437                 off = si->off;
6438                 off -= offsetof(struct bpf_sock, src_ip6[0]);
6439                 *insn++ = BPF_LDX_MEM(
6440                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6441                         bpf_target_off(
6442                                 struct sock_common,
6443                                 skc_v6_rcv_saddr.s6_addr32[0],
6444                                 FIELD_SIZEOF(struct sock_common,
6445                                              skc_v6_rcv_saddr.s6_addr32[0]),
6446                                 target_size) + off);
6447 #else
6448                 (void)off;
6449                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6450 #endif
6451                 break;
6452
6453         case offsetof(struct bpf_sock, src_port):
6454                 *insn++ = BPF_LDX_MEM(
6455                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6456                         si->dst_reg, si->src_reg,
6457                         bpf_target_off(struct sock_common, skc_num,
6458                                        FIELD_SIZEOF(struct sock_common,
6459                                                     skc_num),
6460                                        target_size));
6461                 break;
6462         }
6463
6464         return insn - insn_buf;
6465 }
6466
6467 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6468                                          const struct bpf_insn *si,
6469                                          struct bpf_insn *insn_buf,
6470                                          struct bpf_prog *prog, u32 *target_size)
6471 {
6472         struct bpf_insn *insn = insn_buf;
6473
6474         switch (si->off) {
6475         case offsetof(struct __sk_buff, ifindex):
6476                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6477                                       si->dst_reg, si->src_reg,
6478                                       offsetof(struct sk_buff, dev));
6479                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6480                                       bpf_target_off(struct net_device, ifindex, 4,
6481                                                      target_size));
6482                 break;
6483         default:
6484                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6485                                               target_size);
6486         }
6487
6488         return insn - insn_buf;
6489 }
6490
6491 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6492                                   const struct bpf_insn *si,
6493                                   struct bpf_insn *insn_buf,
6494                                   struct bpf_prog *prog, u32 *target_size)
6495 {
6496         struct bpf_insn *insn = insn_buf;
6497
6498         switch (si->off) {
6499         case offsetof(struct xdp_md, data):
6500                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6501                                       si->dst_reg, si->src_reg,
6502                                       offsetof(struct xdp_buff, data));
6503                 break;
6504         case offsetof(struct xdp_md, data_meta):
6505                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6506                                       si->dst_reg, si->src_reg,
6507                                       offsetof(struct xdp_buff, data_meta));
6508                 break;
6509         case offsetof(struct xdp_md, data_end):
6510                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6511                                       si->dst_reg, si->src_reg,
6512                                       offsetof(struct xdp_buff, data_end));
6513                 break;
6514         case offsetof(struct xdp_md, ingress_ifindex):
6515                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6516                                       si->dst_reg, si->src_reg,
6517                                       offsetof(struct xdp_buff, rxq));
6518                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6519                                       si->dst_reg, si->dst_reg,
6520                                       offsetof(struct xdp_rxq_info, dev));
6521                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6522                                       offsetof(struct net_device, ifindex));
6523                 break;
6524         case offsetof(struct xdp_md, rx_queue_index):
6525                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6526                                       si->dst_reg, si->src_reg,
6527                                       offsetof(struct xdp_buff, rxq));
6528                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6529                                       offsetof(struct xdp_rxq_info,
6530                                                queue_index));
6531                 break;
6532         }
6533
6534         return insn - insn_buf;
6535 }
6536
6537 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6538  * context Structure, F is Field in context structure that contains a pointer
6539  * to Nested Structure of type NS that has the field NF.
6540  *
6541  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6542  * sure that SIZE is not greater than actual size of S.F.NF.
6543  *
6544  * If offset OFF is provided, the load happens from that offset relative to
6545  * offset of NF.
6546  */
6547 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
6548         do {                                                                   \
6549                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
6550                                       si->src_reg, offsetof(S, F));            \
6551                 *insn++ = BPF_LDX_MEM(                                         \
6552                         SIZE, si->dst_reg, si->dst_reg,                        \
6553                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6554                                        target_size)                            \
6555                                 + OFF);                                        \
6556         } while (0)
6557
6558 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
6559         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
6560                                              BPF_FIELD_SIZEOF(NS, NF), 0)
6561
6562 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6563  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6564  *
6565  * It doesn't support SIZE argument though since narrow stores are not
6566  * supported for now.
6567  *
6568  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6569  * "register" since two registers available in convert_ctx_access are not
6570  * enough: we can't override neither SRC, since it contains value to store, nor
6571  * DST since it contains pointer to context that may be used by later
6572  * instructions. But we need a temporary place to save pointer to nested
6573  * structure whose field we want to store to.
6574  */
6575 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
6576         do {                                                                   \
6577                 int tmp_reg = BPF_REG_9;                                       \
6578                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6579                         --tmp_reg;                                             \
6580                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6581                         --tmp_reg;                                             \
6582                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
6583                                       offsetof(S, TF));                        \
6584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
6585                                       si->dst_reg, offsetof(S, F));            \
6586                 *insn++ = BPF_STX_MEM(                                         \
6587                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
6588                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6589                                        target_size)                            \
6590                                 + OFF);                                        \
6591                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
6592                                       offsetof(S, TF));                        \
6593         } while (0)
6594
6595 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6596                                                       TF)                      \
6597         do {                                                                   \
6598                 if (type == BPF_WRITE) {                                       \
6599                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
6600                                                          TF);                  \
6601                 } else {                                                       \
6602                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
6603                                 S, NS, F, NF, SIZE, OFF);  \
6604                 }                                                              \
6605         } while (0)
6606
6607 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
6608         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
6609                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6610
6611 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6612                                         const struct bpf_insn *si,
6613                                         struct bpf_insn *insn_buf,
6614                                         struct bpf_prog *prog, u32 *target_size)
6615 {
6616         struct bpf_insn *insn = insn_buf;
6617         int off;
6618
6619         switch (si->off) {
6620         case offsetof(struct bpf_sock_addr, user_family):
6621                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6622                                             struct sockaddr, uaddr, sa_family);
6623                 break;
6624
6625         case offsetof(struct bpf_sock_addr, user_ip4):
6626                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6627                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6628                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6629                 break;
6630
6631         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6632                 off = si->off;
6633                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6634                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6635                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6636                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6637                         tmp_reg);
6638                 break;
6639
6640         case offsetof(struct bpf_sock_addr, user_port):
6641                 /* To get port we need to know sa_family first and then treat
6642                  * sockaddr as either sockaddr_in or sockaddr_in6.
6643                  * Though we can simplify since port field has same offset and
6644                  * size in both structures.
6645                  * Here we check this invariant and use just one of the
6646                  * structures if it's true.
6647                  */
6648                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6649                              offsetof(struct sockaddr_in6, sin6_port));
6650                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6651                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6652                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6653                                                      struct sockaddr_in6, uaddr,
6654                                                      sin6_port, tmp_reg);
6655                 break;
6656
6657         case offsetof(struct bpf_sock_addr, family):
6658                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6659                                             struct sock, sk, sk_family);
6660                 break;
6661
6662         case offsetof(struct bpf_sock_addr, type):
6663                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6664                         struct bpf_sock_addr_kern, struct sock, sk,
6665                         __sk_flags_offset, BPF_W, 0);
6666                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6667                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6668                 break;
6669
6670         case offsetof(struct bpf_sock_addr, protocol):
6671                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6672                         struct bpf_sock_addr_kern, struct sock, sk,
6673                         __sk_flags_offset, BPF_W, 0);
6674                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6675                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6676                                         SK_FL_PROTO_SHIFT);
6677                 break;
6678
6679         case offsetof(struct bpf_sock_addr, msg_src_ip4):
6680                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6681                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6682                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6683                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6684                 break;
6685
6686         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6687                                 msg_src_ip6[3]):
6688                 off = si->off;
6689                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6690                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6691                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6692                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6693                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6694                 break;
6695         }
6696
6697         return insn - insn_buf;
6698 }
6699
6700 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6701                                        const struct bpf_insn *si,
6702                                        struct bpf_insn *insn_buf,
6703                                        struct bpf_prog *prog,
6704                                        u32 *target_size)
6705 {
6706         struct bpf_insn *insn = insn_buf;
6707         int off;
6708
6709         switch (si->off) {
6710         case offsetof(struct bpf_sock_ops, op) ...
6711              offsetof(struct bpf_sock_ops, replylong[3]):
6712                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6713                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6714                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6715                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6716                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6717                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6718                 off = si->off;
6719                 off -= offsetof(struct bpf_sock_ops, op);
6720                 off += offsetof(struct bpf_sock_ops_kern, op);
6721                 if (type == BPF_WRITE)
6722                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6723                                               off);
6724                 else
6725                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6726                                               off);
6727                 break;
6728
6729         case offsetof(struct bpf_sock_ops, family):
6730                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6731
6732                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6733                                               struct bpf_sock_ops_kern, sk),
6734                                       si->dst_reg, si->src_reg,
6735                                       offsetof(struct bpf_sock_ops_kern, sk));
6736                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6737                                       offsetof(struct sock_common, skc_family));
6738                 break;
6739
6740         case offsetof(struct bpf_sock_ops, remote_ip4):
6741                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6742
6743                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6744                                                 struct bpf_sock_ops_kern, sk),
6745                                       si->dst_reg, si->src_reg,
6746                                       offsetof(struct bpf_sock_ops_kern, sk));
6747                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6748                                       offsetof(struct sock_common, skc_daddr));
6749                 break;
6750
6751         case offsetof(struct bpf_sock_ops, local_ip4):
6752                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6753                                           skc_rcv_saddr) != 4);
6754
6755                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6756                                               struct bpf_sock_ops_kern, sk),
6757                                       si->dst_reg, si->src_reg,
6758                                       offsetof(struct bpf_sock_ops_kern, sk));
6759                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6760                                       offsetof(struct sock_common,
6761                                                skc_rcv_saddr));
6762                 break;
6763
6764         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6765              offsetof(struct bpf_sock_ops, remote_ip6[3]):
6766 #if IS_ENABLED(CONFIG_IPV6)
6767                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6768                                           skc_v6_daddr.s6_addr32[0]) != 4);
6769
6770                 off = si->off;
6771                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6772                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6773                                                 struct bpf_sock_ops_kern, sk),
6774                                       si->dst_reg, si->src_reg,
6775                                       offsetof(struct bpf_sock_ops_kern, sk));
6776                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6777                                       offsetof(struct sock_common,
6778                                                skc_v6_daddr.s6_addr32[0]) +
6779                                       off);
6780 #else
6781                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6782 #endif
6783                 break;
6784
6785         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6786              offsetof(struct bpf_sock_ops, local_ip6[3]):
6787 #if IS_ENABLED(CONFIG_IPV6)
6788                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6789                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6790
6791                 off = si->off;
6792                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6793                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6794                                                 struct bpf_sock_ops_kern, sk),
6795                                       si->dst_reg, si->src_reg,
6796                                       offsetof(struct bpf_sock_ops_kern, sk));
6797                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6798                                       offsetof(struct sock_common,
6799                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6800                                       off);
6801 #else
6802                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6803 #endif
6804                 break;
6805
6806         case offsetof(struct bpf_sock_ops, remote_port):
6807                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6808
6809                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6810                                                 struct bpf_sock_ops_kern, sk),
6811                                       si->dst_reg, si->src_reg,
6812                                       offsetof(struct bpf_sock_ops_kern, sk));
6813                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6814                                       offsetof(struct sock_common, skc_dport));
6815 #ifndef __BIG_ENDIAN_BITFIELD
6816                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6817 #endif
6818                 break;
6819
6820         case offsetof(struct bpf_sock_ops, local_port):
6821                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6822
6823                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6824                                                 struct bpf_sock_ops_kern, sk),
6825                                       si->dst_reg, si->src_reg,
6826                                       offsetof(struct bpf_sock_ops_kern, sk));
6827                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6828                                       offsetof(struct sock_common, skc_num));
6829                 break;
6830
6831         case offsetof(struct bpf_sock_ops, is_fullsock):
6832                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6833                                                 struct bpf_sock_ops_kern,
6834                                                 is_fullsock),
6835                                       si->dst_reg, si->src_reg,
6836                                       offsetof(struct bpf_sock_ops_kern,
6837                                                is_fullsock));
6838                 break;
6839
6840         case offsetof(struct bpf_sock_ops, state):
6841                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6842
6843                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6844                                                 struct bpf_sock_ops_kern, sk),
6845                                       si->dst_reg, si->src_reg,
6846                                       offsetof(struct bpf_sock_ops_kern, sk));
6847                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6848                                       offsetof(struct sock_common, skc_state));
6849                 break;
6850
6851         case offsetof(struct bpf_sock_ops, rtt_min):
6852                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6853                              sizeof(struct minmax));
6854                 BUILD_BUG_ON(sizeof(struct minmax) <
6855                              sizeof(struct minmax_sample));
6856
6857                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6858                                                 struct bpf_sock_ops_kern, sk),
6859                                       si->dst_reg, si->src_reg,
6860                                       offsetof(struct bpf_sock_ops_kern, sk));
6861                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6862                                       offsetof(struct tcp_sock, rtt_min) +
6863                                       FIELD_SIZEOF(struct minmax_sample, t));
6864                 break;
6865
6866 /* Helper macro for adding read access to tcp_sock or sock fields. */
6867 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6868         do {                                                                  \
6869                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6870                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6871                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6872                                                 struct bpf_sock_ops_kern,     \
6873                                                 is_fullsock),                 \
6874                                       si->dst_reg, si->src_reg,               \
6875                                       offsetof(struct bpf_sock_ops_kern,      \
6876                                                is_fullsock));                 \
6877                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
6878                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6879                                                 struct bpf_sock_ops_kern, sk),\
6880                                       si->dst_reg, si->src_reg,               \
6881                                       offsetof(struct bpf_sock_ops_kern, sk));\
6882                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
6883                                                        OBJ_FIELD),            \
6884                                       si->dst_reg, si->dst_reg,               \
6885                                       offsetof(OBJ, OBJ_FIELD));              \
6886         } while (0)
6887
6888 /* Helper macro for adding write access to tcp_sock or sock fields.
6889  * The macro is called with two registers, dst_reg which contains a pointer
6890  * to ctx (context) and src_reg which contains the value that should be
6891  * stored. However, we need an additional register since we cannot overwrite
6892  * dst_reg because it may be used later in the program.
6893  * Instead we "borrow" one of the other register. We first save its value
6894  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6895  * it at the end of the macro.
6896  */
6897 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6898         do {                                                                  \
6899                 int reg = BPF_REG_9;                                          \
6900                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6901                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6902                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6903                         reg--;                                                \
6904                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6905                         reg--;                                                \
6906                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
6907                                       offsetof(struct bpf_sock_ops_kern,      \
6908                                                temp));                        \
6909                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6910                                                 struct bpf_sock_ops_kern,     \
6911                                                 is_fullsock),                 \
6912                                       reg, si->dst_reg,                       \
6913                                       offsetof(struct bpf_sock_ops_kern,      \
6914                                                is_fullsock));                 \
6915                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
6916                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6917                                                 struct bpf_sock_ops_kern, sk),\
6918                                       reg, si->dst_reg,                       \
6919                                       offsetof(struct bpf_sock_ops_kern, sk));\
6920                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
6921                                       reg, si->src_reg,                       \
6922                                       offsetof(OBJ, OBJ_FIELD));              \
6923                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
6924                                       offsetof(struct bpf_sock_ops_kern,      \
6925                                                temp));                        \
6926         } while (0)
6927
6928 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
6929         do {                                                                  \
6930                 if (TYPE == BPF_WRITE)                                        \
6931                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6932                 else                                                          \
6933                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6934         } while (0)
6935
6936         case offsetof(struct bpf_sock_ops, snd_cwnd):
6937                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6938                 break;
6939
6940         case offsetof(struct bpf_sock_ops, srtt_us):
6941                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6942                 break;
6943
6944         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6945                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6946                                    struct tcp_sock);
6947                 break;
6948
6949         case offsetof(struct bpf_sock_ops, snd_ssthresh):
6950                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6951                 break;
6952
6953         case offsetof(struct bpf_sock_ops, rcv_nxt):
6954                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6955                 break;
6956
6957         case offsetof(struct bpf_sock_ops, snd_nxt):
6958                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6959                 break;
6960
6961         case offsetof(struct bpf_sock_ops, snd_una):
6962                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6963                 break;
6964
6965         case offsetof(struct bpf_sock_ops, mss_cache):
6966                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6967                 break;
6968
6969         case offsetof(struct bpf_sock_ops, ecn_flags):
6970                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6971                 break;
6972
6973         case offsetof(struct bpf_sock_ops, rate_delivered):
6974                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6975                                    struct tcp_sock);
6976                 break;
6977
6978         case offsetof(struct bpf_sock_ops, rate_interval_us):
6979                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6980                                    struct tcp_sock);
6981                 break;
6982
6983         case offsetof(struct bpf_sock_ops, packets_out):
6984                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6985                 break;
6986
6987         case offsetof(struct bpf_sock_ops, retrans_out):
6988                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6989                 break;
6990
6991         case offsetof(struct bpf_sock_ops, total_retrans):
6992                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6993                                    struct tcp_sock);
6994                 break;
6995
6996         case offsetof(struct bpf_sock_ops, segs_in):
6997                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6998                 break;
6999
7000         case offsetof(struct bpf_sock_ops, data_segs_in):
7001                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
7002                 break;
7003
7004         case offsetof(struct bpf_sock_ops, segs_out):
7005                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
7006                 break;
7007
7008         case offsetof(struct bpf_sock_ops, data_segs_out):
7009                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
7010                                    struct tcp_sock);
7011                 break;
7012
7013         case offsetof(struct bpf_sock_ops, lost_out):
7014                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
7015                 break;
7016
7017         case offsetof(struct bpf_sock_ops, sacked_out):
7018                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
7019                 break;
7020
7021         case offsetof(struct bpf_sock_ops, sk_txhash):
7022                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
7023                                           struct sock, type);
7024                 break;
7025
7026         case offsetof(struct bpf_sock_ops, bytes_received):
7027                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
7028                                    struct tcp_sock);
7029                 break;
7030
7031         case offsetof(struct bpf_sock_ops, bytes_acked):
7032                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
7033                 break;
7034
7035         }
7036         return insn - insn_buf;
7037 }
7038
7039 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
7040                                      const struct bpf_insn *si,
7041                                      struct bpf_insn *insn_buf,
7042                                      struct bpf_prog *prog, u32 *target_size)
7043 {
7044         struct bpf_insn *insn = insn_buf;
7045         int off;
7046
7047         switch (si->off) {
7048         case offsetof(struct __sk_buff, data_end):
7049                 off  = si->off;
7050                 off -= offsetof(struct __sk_buff, data_end);
7051                 off += offsetof(struct sk_buff, cb);
7052                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
7053                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
7054                                       si->src_reg, off);
7055                 break;
7056         default:
7057                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
7058                                               target_size);
7059         }
7060
7061         return insn - insn_buf;
7062 }
7063
7064 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
7065                                      const struct bpf_insn *si,
7066                                      struct bpf_insn *insn_buf,
7067                                      struct bpf_prog *prog, u32 *target_size)
7068 {
7069         struct bpf_insn *insn = insn_buf;
7070 #if IS_ENABLED(CONFIG_IPV6)
7071         int off;
7072 #endif
7073
7074         switch (si->off) {
7075         case offsetof(struct sk_msg_md, data):
7076                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data),
7077                                       si->dst_reg, si->src_reg,
7078                                       offsetof(struct sk_msg, data));
7079                 break;
7080         case offsetof(struct sk_msg_md, data_end):
7081                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end),
7082                                       si->dst_reg, si->src_reg,
7083                                       offsetof(struct sk_msg, data_end));
7084                 break;
7085         case offsetof(struct sk_msg_md, family):
7086                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
7087
7088                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7089                                               struct sk_msg, sk),
7090                                       si->dst_reg, si->src_reg,
7091                                       offsetof(struct sk_msg, sk));
7092                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7093                                       offsetof(struct sock_common, skc_family));
7094                 break;
7095
7096         case offsetof(struct sk_msg_md, remote_ip4):
7097                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
7098
7099                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7100                                                 struct sk_msg, sk),
7101                                       si->dst_reg, si->src_reg,
7102                                       offsetof(struct sk_msg, sk));
7103                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7104                                       offsetof(struct sock_common, skc_daddr));
7105                 break;
7106
7107         case offsetof(struct sk_msg_md, local_ip4):
7108                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7109                                           skc_rcv_saddr) != 4);
7110
7111                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7112                                               struct sk_msg, sk),
7113                                       si->dst_reg, si->src_reg,
7114                                       offsetof(struct sk_msg, sk));
7115                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7116                                       offsetof(struct sock_common,
7117                                                skc_rcv_saddr));
7118                 break;
7119
7120         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
7121              offsetof(struct sk_msg_md, remote_ip6[3]):
7122 #if IS_ENABLED(CONFIG_IPV6)
7123                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7124                                           skc_v6_daddr.s6_addr32[0]) != 4);
7125
7126                 off = si->off;
7127                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
7128                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7129                                                 struct sk_msg, sk),
7130                                       si->dst_reg, si->src_reg,
7131                                       offsetof(struct sk_msg, sk));
7132                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7133                                       offsetof(struct sock_common,
7134                                                skc_v6_daddr.s6_addr32[0]) +
7135                                       off);
7136 #else
7137                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7138 #endif
7139                 break;
7140
7141         case offsetof(struct sk_msg_md, local_ip6[0]) ...
7142              offsetof(struct sk_msg_md, local_ip6[3]):
7143 #if IS_ENABLED(CONFIG_IPV6)
7144                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
7145                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
7146
7147                 off = si->off;
7148                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
7149                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7150                                                 struct sk_msg, sk),
7151                                       si->dst_reg, si->src_reg,
7152                                       offsetof(struct sk_msg, sk));
7153                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
7154                                       offsetof(struct sock_common,
7155                                                skc_v6_rcv_saddr.s6_addr32[0]) +
7156                                       off);
7157 #else
7158                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
7159 #endif
7160                 break;
7161
7162         case offsetof(struct sk_msg_md, remote_port):
7163                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
7164
7165                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7166                                                 struct sk_msg, sk),
7167                                       si->dst_reg, si->src_reg,
7168                                       offsetof(struct sk_msg, sk));
7169                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7170                                       offsetof(struct sock_common, skc_dport));
7171 #ifndef __BIG_ENDIAN_BITFIELD
7172                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
7173 #endif
7174                 break;
7175
7176         case offsetof(struct sk_msg_md, local_port):
7177                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
7178
7179                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
7180                                                 struct sk_msg, sk),
7181                                       si->dst_reg, si->src_reg,
7182                                       offsetof(struct sk_msg, sk));
7183                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
7184                                       offsetof(struct sock_common, skc_num));
7185                 break;
7186         }
7187
7188         return insn - insn_buf;
7189 }
7190
7191 const struct bpf_verifier_ops sk_filter_verifier_ops = {
7192         .get_func_proto         = sk_filter_func_proto,
7193         .is_valid_access        = sk_filter_is_valid_access,
7194         .convert_ctx_access     = bpf_convert_ctx_access,
7195         .gen_ld_abs             = bpf_gen_ld_abs,
7196 };
7197
7198 const struct bpf_prog_ops sk_filter_prog_ops = {
7199         .test_run               = bpf_prog_test_run_skb,
7200 };
7201
7202 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
7203         .get_func_proto         = tc_cls_act_func_proto,
7204         .is_valid_access        = tc_cls_act_is_valid_access,
7205         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
7206         .gen_prologue           = tc_cls_act_prologue,
7207         .gen_ld_abs             = bpf_gen_ld_abs,
7208 };
7209
7210 const struct bpf_prog_ops tc_cls_act_prog_ops = {
7211         .test_run               = bpf_prog_test_run_skb,
7212 };
7213
7214 const struct bpf_verifier_ops xdp_verifier_ops = {
7215         .get_func_proto         = xdp_func_proto,
7216         .is_valid_access        = xdp_is_valid_access,
7217         .convert_ctx_access     = xdp_convert_ctx_access,
7218         .gen_prologue           = bpf_noop_prologue,
7219 };
7220
7221 const struct bpf_prog_ops xdp_prog_ops = {
7222         .test_run               = bpf_prog_test_run_xdp,
7223 };
7224
7225 const struct bpf_verifier_ops cg_skb_verifier_ops = {
7226         .get_func_proto         = cg_skb_func_proto,
7227         .is_valid_access        = cg_skb_is_valid_access,
7228         .convert_ctx_access     = bpf_convert_ctx_access,
7229 };
7230
7231 const struct bpf_prog_ops cg_skb_prog_ops = {
7232         .test_run               = bpf_prog_test_run_skb,
7233 };
7234
7235 const struct bpf_verifier_ops lwt_in_verifier_ops = {
7236         .get_func_proto         = lwt_in_func_proto,
7237         .is_valid_access        = lwt_is_valid_access,
7238         .convert_ctx_access     = bpf_convert_ctx_access,
7239 };
7240
7241 const struct bpf_prog_ops lwt_in_prog_ops = {
7242         .test_run               = bpf_prog_test_run_skb,
7243 };
7244
7245 const struct bpf_verifier_ops lwt_out_verifier_ops = {
7246         .get_func_proto         = lwt_out_func_proto,
7247         .is_valid_access        = lwt_is_valid_access,
7248         .convert_ctx_access     = bpf_convert_ctx_access,
7249 };
7250
7251 const struct bpf_prog_ops lwt_out_prog_ops = {
7252         .test_run               = bpf_prog_test_run_skb,
7253 };
7254
7255 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
7256         .get_func_proto         = lwt_xmit_func_proto,
7257         .is_valid_access        = lwt_is_valid_access,
7258         .convert_ctx_access     = bpf_convert_ctx_access,
7259         .gen_prologue           = tc_cls_act_prologue,
7260 };
7261
7262 const struct bpf_prog_ops lwt_xmit_prog_ops = {
7263         .test_run               = bpf_prog_test_run_skb,
7264 };
7265
7266 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
7267         .get_func_proto         = lwt_seg6local_func_proto,
7268         .is_valid_access        = lwt_is_valid_access,
7269         .convert_ctx_access     = bpf_convert_ctx_access,
7270 };
7271
7272 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
7273         .test_run               = bpf_prog_test_run_skb,
7274 };
7275
7276 const struct bpf_verifier_ops cg_sock_verifier_ops = {
7277         .get_func_proto         = sock_filter_func_proto,
7278         .is_valid_access        = sock_filter_is_valid_access,
7279         .convert_ctx_access     = bpf_sock_convert_ctx_access,
7280 };
7281
7282 const struct bpf_prog_ops cg_sock_prog_ops = {
7283 };
7284
7285 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
7286         .get_func_proto         = sock_addr_func_proto,
7287         .is_valid_access        = sock_addr_is_valid_access,
7288         .convert_ctx_access     = sock_addr_convert_ctx_access,
7289 };
7290
7291 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
7292 };
7293
7294 const struct bpf_verifier_ops sock_ops_verifier_ops = {
7295         .get_func_proto         = sock_ops_func_proto,
7296         .is_valid_access        = sock_ops_is_valid_access,
7297         .convert_ctx_access     = sock_ops_convert_ctx_access,
7298 };
7299
7300 const struct bpf_prog_ops sock_ops_prog_ops = {
7301 };
7302
7303 const struct bpf_verifier_ops sk_skb_verifier_ops = {
7304         .get_func_proto         = sk_skb_func_proto,
7305         .is_valid_access        = sk_skb_is_valid_access,
7306         .convert_ctx_access     = sk_skb_convert_ctx_access,
7307         .gen_prologue           = sk_skb_prologue,
7308 };
7309
7310 const struct bpf_prog_ops sk_skb_prog_ops = {
7311 };
7312
7313 const struct bpf_verifier_ops sk_msg_verifier_ops = {
7314         .get_func_proto         = sk_msg_func_proto,
7315         .is_valid_access        = sk_msg_is_valid_access,
7316         .convert_ctx_access     = sk_msg_convert_ctx_access,
7317         .gen_prologue           = bpf_noop_prologue,
7318 };
7319
7320 const struct bpf_prog_ops sk_msg_prog_ops = {
7321 };
7322
7323 const struct bpf_verifier_ops flow_dissector_verifier_ops = {
7324         .get_func_proto         = flow_dissector_func_proto,
7325         .is_valid_access        = flow_dissector_is_valid_access,
7326         .convert_ctx_access     = bpf_convert_ctx_access,
7327 };
7328
7329 const struct bpf_prog_ops flow_dissector_prog_ops = {
7330 };
7331
7332 int sk_detach_filter(struct sock *sk)
7333 {
7334         int ret = -ENOENT;
7335         struct sk_filter *filter;
7336
7337         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7338                 return -EPERM;
7339
7340         filter = rcu_dereference_protected(sk->sk_filter,
7341                                            lockdep_sock_is_held(sk));
7342         if (filter) {
7343                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7344                 sk_filter_uncharge(sk, filter);
7345                 ret = 0;
7346         }
7347
7348         return ret;
7349 }
7350 EXPORT_SYMBOL_GPL(sk_detach_filter);
7351
7352 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7353                   unsigned int len)
7354 {
7355         struct sock_fprog_kern *fprog;
7356         struct sk_filter *filter;
7357         int ret = 0;
7358
7359         lock_sock(sk);
7360         filter = rcu_dereference_protected(sk->sk_filter,
7361                                            lockdep_sock_is_held(sk));
7362         if (!filter)
7363                 goto out;
7364
7365         /* We're copying the filter that has been originally attached,
7366          * so no conversion/decode needed anymore. eBPF programs that
7367          * have no original program cannot be dumped through this.
7368          */
7369         ret = -EACCES;
7370         fprog = filter->prog->orig_prog;
7371         if (!fprog)
7372                 goto out;
7373
7374         ret = fprog->len;
7375         if (!len)
7376                 /* User space only enquires number of filter blocks. */
7377                 goto out;
7378
7379         ret = -EINVAL;
7380         if (len < fprog->len)
7381                 goto out;
7382
7383         ret = -EFAULT;
7384         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7385                 goto out;
7386
7387         /* Instead of bytes, the API requests to return the number
7388          * of filter blocks.
7389          */
7390         ret = fprog->len;
7391 out:
7392         release_sock(sk);
7393         return ret;
7394 }
7395
7396 #ifdef CONFIG_INET
7397 struct sk_reuseport_kern {
7398         struct sk_buff *skb;
7399         struct sock *sk;
7400         struct sock *selected_sk;
7401         void *data_end;
7402         u32 hash;
7403         u32 reuseport_id;
7404         bool bind_inany;
7405 };
7406
7407 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7408                                     struct sock_reuseport *reuse,
7409                                     struct sock *sk, struct sk_buff *skb,
7410                                     u32 hash)
7411 {
7412         reuse_kern->skb = skb;
7413         reuse_kern->sk = sk;
7414         reuse_kern->selected_sk = NULL;
7415         reuse_kern->data_end = skb->data + skb_headlen(skb);
7416         reuse_kern->hash = hash;
7417         reuse_kern->reuseport_id = reuse->reuseport_id;
7418         reuse_kern->bind_inany = reuse->bind_inany;
7419 }
7420
7421 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7422                                   struct bpf_prog *prog, struct sk_buff *skb,
7423                                   u32 hash)
7424 {
7425         struct sk_reuseport_kern reuse_kern;
7426         enum sk_action action;
7427
7428         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7429         action = BPF_PROG_RUN(prog, &reuse_kern);
7430
7431         if (action == SK_PASS)
7432                 return reuse_kern.selected_sk;
7433         else
7434                 return ERR_PTR(-ECONNREFUSED);
7435 }
7436
7437 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7438            struct bpf_map *, map, void *, key, u32, flags)
7439 {
7440         struct sock_reuseport *reuse;
7441         struct sock *selected_sk;
7442
7443         selected_sk = map->ops->map_lookup_elem(map, key);
7444         if (!selected_sk)
7445                 return -ENOENT;
7446
7447         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7448         if (!reuse)
7449                 /* selected_sk is unhashed (e.g. by close()) after the
7450                  * above map_lookup_elem().  Treat selected_sk has already
7451                  * been removed from the map.
7452                  */
7453                 return -ENOENT;
7454
7455         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7456                 struct sock *sk;
7457
7458                 if (unlikely(!reuse_kern->reuseport_id))
7459                         /* There is a small race between adding the
7460                          * sk to the map and setting the
7461                          * reuse_kern->reuseport_id.
7462                          * Treat it as the sk has not been added to
7463                          * the bpf map yet.
7464                          */
7465                         return -ENOENT;
7466
7467                 sk = reuse_kern->sk;
7468                 if (sk->sk_protocol != selected_sk->sk_protocol)
7469                         return -EPROTOTYPE;
7470                 else if (sk->sk_family != selected_sk->sk_family)
7471                         return -EAFNOSUPPORT;
7472
7473                 /* Catch all. Likely bound to a different sockaddr. */
7474                 return -EBADFD;
7475         }
7476
7477         reuse_kern->selected_sk = selected_sk;
7478
7479         return 0;
7480 }
7481
7482 static const struct bpf_func_proto sk_select_reuseport_proto = {
7483         .func           = sk_select_reuseport,
7484         .gpl_only       = false,
7485         .ret_type       = RET_INTEGER,
7486         .arg1_type      = ARG_PTR_TO_CTX,
7487         .arg2_type      = ARG_CONST_MAP_PTR,
7488         .arg3_type      = ARG_PTR_TO_MAP_KEY,
7489         .arg4_type      = ARG_ANYTHING,
7490 };
7491
7492 BPF_CALL_4(sk_reuseport_load_bytes,
7493            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7494            void *, to, u32, len)
7495 {
7496         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7497 }
7498
7499 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7500         .func           = sk_reuseport_load_bytes,
7501         .gpl_only       = false,
7502         .ret_type       = RET_INTEGER,
7503         .arg1_type      = ARG_PTR_TO_CTX,
7504         .arg2_type      = ARG_ANYTHING,
7505         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7506         .arg4_type      = ARG_CONST_SIZE,
7507 };
7508
7509 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7510            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7511            void *, to, u32, len, u32, start_header)
7512 {
7513         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7514                                                len, start_header);
7515 }
7516
7517 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7518         .func           = sk_reuseport_load_bytes_relative,
7519         .gpl_only       = false,
7520         .ret_type       = RET_INTEGER,
7521         .arg1_type      = ARG_PTR_TO_CTX,
7522         .arg2_type      = ARG_ANYTHING,
7523         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7524         .arg4_type      = ARG_CONST_SIZE,
7525         .arg5_type      = ARG_ANYTHING,
7526 };
7527
7528 static const struct bpf_func_proto *
7529 sk_reuseport_func_proto(enum bpf_func_id func_id,
7530                         const struct bpf_prog *prog)
7531 {
7532         switch (func_id) {
7533         case BPF_FUNC_sk_select_reuseport:
7534                 return &sk_select_reuseport_proto;
7535         case BPF_FUNC_skb_load_bytes:
7536                 return &sk_reuseport_load_bytes_proto;
7537         case BPF_FUNC_skb_load_bytes_relative:
7538                 return &sk_reuseport_load_bytes_relative_proto;
7539         default:
7540                 return bpf_base_func_proto(func_id);
7541         }
7542 }
7543
7544 static bool
7545 sk_reuseport_is_valid_access(int off, int size,
7546                              enum bpf_access_type type,
7547                              const struct bpf_prog *prog,
7548                              struct bpf_insn_access_aux *info)
7549 {
7550         const u32 size_default = sizeof(__u32);
7551
7552         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7553             off % size || type != BPF_READ)
7554                 return false;
7555
7556         switch (off) {
7557         case offsetof(struct sk_reuseport_md, data):
7558                 info->reg_type = PTR_TO_PACKET;
7559                 return size == sizeof(__u64);
7560
7561         case offsetof(struct sk_reuseport_md, data_end):
7562                 info->reg_type = PTR_TO_PACKET_END;
7563                 return size == sizeof(__u64);
7564
7565         case offsetof(struct sk_reuseport_md, hash):
7566                 return size == size_default;
7567
7568         /* Fields that allow narrowing */
7569         case offsetof(struct sk_reuseport_md, eth_protocol):
7570                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7571                         return false;
7572                 /* fall through */
7573         case offsetof(struct sk_reuseport_md, ip_protocol):
7574         case offsetof(struct sk_reuseport_md, bind_inany):
7575         case offsetof(struct sk_reuseport_md, len):
7576                 bpf_ctx_record_field_size(info, size_default);
7577                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7578
7579         default:
7580                 return false;
7581         }
7582 }
7583
7584 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
7585         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7586                               si->dst_reg, si->src_reg,                 \
7587                               bpf_target_off(struct sk_reuseport_kern, F, \
7588                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7589                                              target_size));             \
7590         })
7591
7592 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
7593         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
7594                                     struct sk_buff,                     \
7595                                     skb,                                \
7596                                     SKB_FIELD)
7597
7598 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7599         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
7600                                              struct sock,               \
7601                                              sk,                        \
7602                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
7603
7604 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7605                                            const struct bpf_insn *si,
7606                                            struct bpf_insn *insn_buf,
7607                                            struct bpf_prog *prog,
7608                                            u32 *target_size)
7609 {
7610         struct bpf_insn *insn = insn_buf;
7611
7612         switch (si->off) {
7613         case offsetof(struct sk_reuseport_md, data):
7614                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7615                 break;
7616
7617         case offsetof(struct sk_reuseport_md, len):
7618                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7619                 break;
7620
7621         case offsetof(struct sk_reuseport_md, eth_protocol):
7622                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7623                 break;
7624
7625         case offsetof(struct sk_reuseport_md, ip_protocol):
7626                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7627                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7628                                                     BPF_W, 0);
7629                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7630                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7631                                         SK_FL_PROTO_SHIFT);
7632                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
7633                  * aware.  No further narrowing or masking is needed.
7634                  */
7635                 *target_size = 1;
7636                 break;
7637
7638         case offsetof(struct sk_reuseport_md, data_end):
7639                 SK_REUSEPORT_LOAD_FIELD(data_end);
7640                 break;
7641
7642         case offsetof(struct sk_reuseport_md, hash):
7643                 SK_REUSEPORT_LOAD_FIELD(hash);
7644                 break;
7645
7646         case offsetof(struct sk_reuseport_md, bind_inany):
7647                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
7648                 break;
7649         }
7650
7651         return insn - insn_buf;
7652 }
7653
7654 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
7655         .get_func_proto         = sk_reuseport_func_proto,
7656         .is_valid_access        = sk_reuseport_is_valid_access,
7657         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
7658 };
7659
7660 const struct bpf_prog_ops sk_reuseport_prog_ops = {
7661 };
7662 #endif /* CONFIG_INET */