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 <net/sock.h>
  42 #include <net/flow_dissector.h>
  43 #include <linux/errno.h>
  44 #include <linux/timer.h>
  45 #include <linux/uaccess.h>
  46 #include <asm/unaligned.h>
  47 #include <asm/cmpxchg.h>
  48 #include <linux/filter.h>
  49 #include <linux/ratelimit.h>
  50 #include <linux/seccomp.h>
  51 #include <linux/if_vlan.h>
  52 #include <linux/bpf.h>
  53 #include <net/sch_generic.h>
  54 #include <net/cls_cgroup.h>
  55 #include <net/dst_metadata.h>
  56 #include <net/dst.h>
  57 #include <net/sock_reuseport.h>
  58 #include <net/busy_poll.h>
  59 #include <net/tcp.h>
  60 #include <net/xfrm.h>
  61 #include <linux/bpf_trace.h>
  62 #include <net/xdp_sock.h>
  63 #include <linux/inetdevice.h>
  64 #include <net/ip_fib.h>
  65 #include <net/flow.h>
  66 #include <net/arp.h>
  67 #include <net/ipv6.h>
  68 #include <linux/seg6_local.h>
  69 #include <net/seg6.h>
  70 #include <net/seg6_local.h>
  71
  72 /**
  73  *      sk_filter_trim_cap - run a packet through a socket filter
  74  *      @sk: sock associated with &sk_buff
  75  *      @skb: buffer to filter
  76  *      @cap: limit on how short the eBPF program may trim the packet
  77  *
  78  * Run the eBPF program and then cut skb->data to correct size returned by
  79  * the program. If pkt_len is 0 we toss packet. If skb->len is smaller
  80  * than pkt_len we keep whole skb->data. This is the socket level
  81  * wrapper to BPF_PROG_RUN. It returns 0 if the packet should
  82  * be accepted or -EPERM if the packet should be tossed.
  83  *
  84  */
  85 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
  86 {
  87         int err;
  88         struct sk_filter *filter;
  89
  90         /*
  91          * If the skb was allocated from pfmemalloc reserves, only
  92          * allow SOCK_MEMALLOC sockets to use it as this socket is
  93          * helping free memory
  94          */
  95         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) {
  96                 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP);
  97                 return -ENOMEM;
  98         }
  99         err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb);
 100         if (err)
 101                 return err;
 102
 103         err = security_sock_rcv_skb(sk, skb);
 104         if (err)
 105                 return err;
 106
 107         rcu_read_lock();
 108         filter = rcu_dereference(sk->sk_filter);
 109         if (filter) {
 110                 struct sock *save_sk = skb->sk;
 111                 unsigned int pkt_len;
 112
 113                 skb->sk = sk;
 114                 pkt_len = bpf_prog_run_save_cb(filter->prog, skb);
 115                 skb->sk = save_sk;
 116                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
 117         }
 118         rcu_read_unlock();
 119
 120         return err;
 121 }
 122 EXPORT_SYMBOL(sk_filter_trim_cap);
 123
 124 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb)
 125 {
 126         return skb_get_poff(skb);
 127 }
 128
 129 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x)
 130 {
 131         struct nlattr *nla;
 132
 133         if (skb_is_nonlinear(skb))
 134                 return 0;
 135
 136         if (skb->len < sizeof(struct nlattr))
 137                 return 0;
 138
 139         if (a > skb->len - sizeof(struct nlattr))
 140                 return 0;
 141
 142         nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x);
 143         if (nla)
 144                 return (void *) nla - (void *) skb->data;
 145
 146         return 0;
 147 }
 148
 149 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x)
 150 {
 151         struct nlattr *nla;
 152
 153         if (skb_is_nonlinear(skb))
 154                 return 0;
 155
 156         if (skb->len < sizeof(struct nlattr))
 157                 return 0;
 158
 159         if (a > skb->len - sizeof(struct nlattr))
 160                 return 0;
 161
 162         nla = (struct nlattr *) &skb->data[a];
 163         if (nla->nla_len > skb->len - a)
 164                 return 0;
 165
 166         nla = nla_find_nested(nla, x);
 167         if (nla)
 168                 return (void *) nla - (void *) skb->data;
 169
 170         return 0;
 171 }
 172
 173 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *,
 174            data, int, headlen, int, offset)
 175 {
 176         u8 tmp, *ptr;
 177         const int len = sizeof(tmp);
 178
 179         if (offset >= 0) {
 180                 if (headlen - offset >= len)
 181                         return *(u8 *)(data + offset);
 182                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 183                         return tmp;
 184         } else {
 185                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 186                 if (likely(ptr))
 187                         return *(u8 *)ptr;
 188         }
 189
 190         return -EFAULT;
 191 }
 192
 193 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb,
 194            int, offset)
 195 {
 196         return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len,
 197                                          offset);
 198 }
 199
 200 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *,
 201            data, int, headlen, int, offset)
 202 {
 203         u16 tmp, *ptr;
 204         const int len = sizeof(tmp);
 205
 206         if (offset >= 0) {
 207                 if (headlen - offset >= len)
 208                         return get_unaligned_be16(data + offset);
 209                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 210                         return be16_to_cpu(tmp);
 211         } else {
 212                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 213                 if (likely(ptr))
 214                         return get_unaligned_be16(ptr);
 215         }
 216
 217         return -EFAULT;
 218 }
 219
 220 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb,
 221            int, offset)
 222 {
 223         return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len,
 224                                           offset);
 225 }
 226
 227 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *,
 228            data, int, headlen, int, offset)
 229 {
 230         u32 tmp, *ptr;
 231         const int len = sizeof(tmp);
 232
 233         if (likely(offset >= 0)) {
 234                 if (headlen - offset >= len)
 235                         return get_unaligned_be32(data + offset);
 236                 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp)))
 237                         return be32_to_cpu(tmp);
 238         } else {
 239                 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len);
 240                 if (likely(ptr))
 241                         return get_unaligned_be32(ptr);
 242         }
 243
 244         return -EFAULT;
 245 }
 246
 247 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb,
 248            int, offset)
 249 {
 250         return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len,
 251                                           offset);
 252 }
 253
 254 BPF_CALL_0(bpf_get_raw_cpu_id)
 255 {
 256         return raw_smp_processor_id();
 257 }
 258
 259 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = {
 260         .func           = bpf_get_raw_cpu_id,
 261         .gpl_only       = false,
 262         .ret_type       = RET_INTEGER,
 263 };
 264
 265 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg,
 266                               struct bpf_insn *insn_buf)
 267 {
 268         struct bpf_insn *insn = insn_buf;
 269
 270         switch (skb_field) {
 271         case SKF_AD_MARK:
 272                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4);
 273
 274                 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg,
 275                                       offsetof(struct sk_buff, mark));
 276                 break;
 277
 278         case SKF_AD_PKTTYPE:
 279                 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET());
 280                 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX);
 281 #ifdef __BIG_ENDIAN_BITFIELD
 282                 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5);
 283 #endif
 284                 break;
 285
 286         case SKF_AD_QUEUE:
 287                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2);
 288
 289                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 290                                       offsetof(struct sk_buff, queue_mapping));
 291                 break;
 292
 293         case SKF_AD_VLAN_TAG:
 294         case SKF_AD_VLAN_TAG_PRESENT:
 295                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2);
 296                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
 297
 298                 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */
 299                 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg,
 300                                       offsetof(struct sk_buff, vlan_tci));
 301                 if (skb_field == SKF_AD_VLAN_TAG) {
 302                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg,
 303                                                 ~VLAN_TAG_PRESENT);
 304                 } else {
 305                         /* dst_reg >>= 12 */
 306                         *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 12);
 307                         /* dst_reg &= 1 */
 308                         *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1);
 309                 }
 310                 break;
 311         }
 312
 313         return insn - insn_buf;
 314 }
 315
 316 static bool convert_bpf_extensions(struct sock_filter *fp,
 317                                    struct bpf_insn **insnp)
 318 {
 319         struct bpf_insn *insn = *insnp;
 320         u32 cnt;
 321
 322         switch (fp->k) {
 323         case SKF_AD_OFF + SKF_AD_PROTOCOL:
 324                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2);
 325
 326                 /* A = *(u16 *) (CTX + offsetof(protocol)) */
 327                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 328                                       offsetof(struct sk_buff, protocol));
 329                 /* A = ntohs(A) [emitting a nop or swap16] */
 330                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 331                 break;
 332
 333         case SKF_AD_OFF + SKF_AD_PKTTYPE:
 334                 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn);
 335                 insn += cnt - 1;
 336                 break;
 337
 338         case SKF_AD_OFF + SKF_AD_IFINDEX:
 339         case SKF_AD_OFF + SKF_AD_HATYPE:
 340                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4);
 341                 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2);
 342
 343                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
 344                                       BPF_REG_TMP, BPF_REG_CTX,
 345                                       offsetof(struct sk_buff, dev));
 346                 /* if (tmp != 0) goto pc + 1 */
 347                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1);
 348                 *insn++ = BPF_EXIT_INSN();
 349                 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX)
 350                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP,
 351                                             offsetof(struct net_device, ifindex));
 352                 else
 353                         *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP,
 354                                             offsetof(struct net_device, type));
 355                 break;
 356
 357         case SKF_AD_OFF + SKF_AD_MARK:
 358                 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn);
 359                 insn += cnt - 1;
 360                 break;
 361
 362         case SKF_AD_OFF + SKF_AD_RXHASH:
 363                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4);
 364
 365                 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX,
 366                                     offsetof(struct sk_buff, hash));
 367                 break;
 368
 369         case SKF_AD_OFF + SKF_AD_QUEUE:
 370                 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn);
 371                 insn += cnt - 1;
 372                 break;
 373
 374         case SKF_AD_OFF + SKF_AD_VLAN_TAG:
 375                 cnt = convert_skb_access(SKF_AD_VLAN_TAG,
 376                                          BPF_REG_A, BPF_REG_CTX, insn);
 377                 insn += cnt - 1;
 378                 break;
 379
 380         case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT:
 381                 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT,
 382                                          BPF_REG_A, BPF_REG_CTX, insn);
 383                 insn += cnt - 1;
 384                 break;
 385
 386         case SKF_AD_OFF + SKF_AD_VLAN_TPID:
 387                 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2);
 388
 389                 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */
 390                 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX,
 391                                       offsetof(struct sk_buff, vlan_proto));
 392                 /* A = ntohs(A) [emitting a nop or swap16] */
 393                 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16);
 394                 break;
 395
 396         case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 397         case SKF_AD_OFF + SKF_AD_NLATTR:
 398         case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 399         case SKF_AD_OFF + SKF_AD_CPU:
 400         case SKF_AD_OFF + SKF_AD_RANDOM:
 401                 /* arg1 = CTX */
 402                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 403                 /* arg2 = A */
 404                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A);
 405                 /* arg3 = X */
 406                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X);
 407                 /* Emit call(arg1=CTX, arg2=A, arg3=X) */
 408                 switch (fp->k) {
 409                 case SKF_AD_OFF + SKF_AD_PAY_OFFSET:
 410                         *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset);
 411                         break;
 412                 case SKF_AD_OFF + SKF_AD_NLATTR:
 413                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr);
 414                         break;
 415                 case SKF_AD_OFF + SKF_AD_NLATTR_NEST:
 416                         *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest);
 417                         break;
 418                 case SKF_AD_OFF + SKF_AD_CPU:
 419                         *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id);
 420                         break;
 421                 case SKF_AD_OFF + SKF_AD_RANDOM:
 422                         *insn = BPF_EMIT_CALL(bpf_user_rnd_u32);
 423                         bpf_user_rnd_init_once();
 424                         break;
 425                 }
 426                 break;
 427
 428         case SKF_AD_OFF + SKF_AD_ALU_XOR_X:
 429                 /* A ^= X */
 430                 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X);
 431                 break;
 432
 433         default:
 434                 /* This is just a dummy call to avoid letting the compiler
 435                  * evict __bpf_call_base() as an optimization. Placed here
 436                  * where no-one bothers.
 437                  */
 438                 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0);
 439                 return false;
 440         }
 441
 442         *insnp = insn;
 443         return true;
 444 }
 445
 446 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp)
 447 {
 448         const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS);
 449         int size = bpf_size_to_bytes(BPF_SIZE(fp->code));
 450         bool endian = BPF_SIZE(fp->code) == BPF_H ||
 451                       BPF_SIZE(fp->code) == BPF_W;
 452         bool indirect = BPF_MODE(fp->code) == BPF_IND;
 453         const int ip_align = NET_IP_ALIGN;
 454         struct bpf_insn *insn = *insnp;
 455         int offset = fp->k;
 456
 457         if (!indirect &&
 458             ((unaligned_ok && offset >= 0) ||
 459              (!unaligned_ok && offset >= 0 &&
 460               offset + ip_align >= 0 &&
 461               offset + ip_align % size == 0))) {
 462                 bool ldx_off_ok = offset <= S16_MAX;
 463
 464                 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H);
 465                 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset);
 466                 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP,
 467                                       size, 2 + endian + (!ldx_off_ok * 2));
 468                 if (ldx_off_ok) {
 469                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 470                                               BPF_REG_D, offset);
 471                 } else {
 472                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D);
 473                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset);
 474                         *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A,
 475                                               BPF_REG_TMP, 0);
 476                 }
 477                 if (endian)
 478                         *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8);
 479                 *insn++ = BPF_JMP_A(8);
 480         }
 481
 482         *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX);
 483         *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D);
 484         *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H);
 485         if (!indirect) {
 486                 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset);
 487         } else {
 488                 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X);
 489                 if (fp->k)
 490                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset);
 491         }
 492
 493         switch (BPF_SIZE(fp->code)) {
 494         case BPF_B:
 495                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8);
 496                 break;
 497         case BPF_H:
 498                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16);
 499                 break;
 500         case BPF_W:
 501                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32);
 502                 break;
 503         default:
 504                 return false;
 505         }
 506
 507         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2);
 508         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 509         *insn   = BPF_EXIT_INSN();
 510
 511         *insnp = insn;
 512         return true;
 513 }
 514
 515 /**
 516  *      bpf_convert_filter - convert filter program
 517  *      @prog: the user passed filter program
 518  *      @len: the length of the user passed filter program
 519  *      @new_prog: allocated 'struct bpf_prog' or NULL
 520  *      @new_len: pointer to store length of converted program
 521  *      @seen_ld_abs: bool whether we've seen ld_abs/ind
 522  *
 523  * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn'
 524  * style extended BPF (eBPF).
 525  * Conversion workflow:
 526  *
 527  * 1) First pass for calculating the new program length:
 528  *   bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs)
 529  *
 530  * 2) 2nd pass to remap in two passes: 1st pass finds new
 531  *    jump offsets, 2nd pass remapping:
 532  *   bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs)
 533  */
 534 static int bpf_convert_filter(struct sock_filter *prog, int len,
 535                               struct bpf_prog *new_prog, int *new_len,
 536                               bool *seen_ld_abs)
 537 {
 538         int new_flen = 0, pass = 0, target, i, stack_off;
 539         struct bpf_insn *new_insn, *first_insn = NULL;
 540         struct sock_filter *fp;
 541         int *addrs = NULL;
 542         u8 bpf_src;
 543
 544         BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK);
 545         BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG);
 546
 547         if (len <= 0 || len > BPF_MAXINSNS)
 548                 return -EINVAL;
 549
 550         if (new_prog) {
 551                 first_insn = new_prog->insnsi;
 552                 addrs = kcalloc(len, sizeof(*addrs),
 553                                 GFP_KERNEL | __GFP_NOWARN);
 554                 if (!addrs)
 555                         return -ENOMEM;
 556         }
 557
 558 do_pass:
 559         new_insn = first_insn;
 560         fp = prog;
 561
 562         /* Classic BPF related prologue emission. */
 563         if (new_prog) {
 564                 /* Classic BPF expects A and X to be reset first. These need
 565                  * to be guaranteed to be the first two instructions.
 566                  */
 567                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 568                 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X);
 569
 570                 /* All programs must keep CTX in callee saved BPF_REG_CTX.
 571                  * In eBPF case it's done by the compiler, here we need to
 572                  * do this ourself. Initial CTX is present in BPF_REG_ARG1.
 573                  */
 574                 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1);
 575                 if (*seen_ld_abs) {
 576                         /* For packet access in classic BPF, cache skb->data
 577                          * in callee-saved BPF R8 and skb->len - skb->data_len
 578                          * (headlen) in BPF R9. Since classic BPF is read-only
 579                          * on CTX, we only need to cache it once.
 580                          */
 581                         *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
 582                                                   BPF_REG_D, BPF_REG_CTX,
 583                                                   offsetof(struct sk_buff, data));
 584                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX,
 585                                                   offsetof(struct sk_buff, len));
 586                         *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX,
 587                                                   offsetof(struct sk_buff, data_len));
 588                         *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP);
 589                 }
 590         } else {
 591                 new_insn += 3;
 592         }
 593
 594         for (i = 0; i < len; fp++, i++) {
 595                 struct bpf_insn tmp_insns[32] = { };
 596                 struct bpf_insn *insn = tmp_insns;
 597
 598                 if (addrs)
 599                         addrs[i] = new_insn - first_insn;
 600
 601                 switch (fp->code) {
 602                 /* All arithmetic insns and skb loads map as-is. */
 603                 case BPF_ALU | BPF_ADD | BPF_X:
 604                 case BPF_ALU | BPF_ADD | BPF_K:
 605                 case BPF_ALU | BPF_SUB | BPF_X:
 606                 case BPF_ALU | BPF_SUB | BPF_K:
 607                 case BPF_ALU | BPF_AND | BPF_X:
 608                 case BPF_ALU | BPF_AND | BPF_K:
 609                 case BPF_ALU | BPF_OR | BPF_X:
 610                 case BPF_ALU | BPF_OR | BPF_K:
 611                 case BPF_ALU | BPF_LSH | BPF_X:
 612                 case BPF_ALU | BPF_LSH | BPF_K:
 613                 case BPF_ALU | BPF_RSH | BPF_X:
 614                 case BPF_ALU | BPF_RSH | BPF_K:
 615                 case BPF_ALU | BPF_XOR | BPF_X:
 616                 case BPF_ALU | BPF_XOR | BPF_K:
 617                 case BPF_ALU | BPF_MUL | BPF_X:
 618                 case BPF_ALU | BPF_MUL | BPF_K:
 619                 case BPF_ALU | BPF_DIV | BPF_X:
 620                 case BPF_ALU | BPF_DIV | BPF_K:
 621                 case BPF_ALU | BPF_MOD | BPF_X:
 622                 case BPF_ALU | BPF_MOD | BPF_K:
 623                 case BPF_ALU | BPF_NEG:
 624                 case BPF_LD | BPF_ABS | BPF_W:
 625                 case BPF_LD | BPF_ABS | BPF_H:
 626                 case BPF_LD | BPF_ABS | BPF_B:
 627                 case BPF_LD | BPF_IND | BPF_W:
 628                 case BPF_LD | BPF_IND | BPF_H:
 629                 case BPF_LD | BPF_IND | BPF_B:
 630                         /* Check for overloaded BPF extension and
 631                          * directly convert it if found, otherwise
 632                          * just move on with mapping.
 633                          */
 634                         if (BPF_CLASS(fp->code) == BPF_LD &&
 635                             BPF_MODE(fp->code) == BPF_ABS &&
 636                             convert_bpf_extensions(fp, &insn))
 637                                 break;
 638                         if (BPF_CLASS(fp->code) == BPF_LD &&
 639                             convert_bpf_ld_abs(fp, &insn)) {
 640                                 *seen_ld_abs = true;
 641                                 break;
 642                         }
 643
 644                         if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) ||
 645                             fp->code == (BPF_ALU | BPF_MOD | BPF_X)) {
 646                                 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X);
 647                                 /* Error with exception code on div/mod by 0.
 648                                  * For cBPF programs, this was always return 0.
 649                                  */
 650                                 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2);
 651                                 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A);
 652                                 *insn++ = BPF_EXIT_INSN();
 653                         }
 654
 655                         *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k);
 656                         break;
 657
 658                 /* Jump transformation cannot use BPF block macros
 659                  * everywhere as offset calculation and target updates
 660                  * require a bit more work than the rest, i.e. jump
 661                  * opcodes map as-is, but offsets need adjustment.
 662                  */
 663
 664 #define BPF_EMIT_JMP                                                    \
 665         do {                                                            \
 666                 const s32 off_min = S16_MIN, off_max = S16_MAX;         \
 667                 s32 off;                                                \
 668                                                                         \
 669                 if (target >= len || target < 0)                        \
 670                         goto err;                                       \
 671                 off = addrs ? addrs[target] - addrs[i] - 1 : 0;         \
 672                 /* Adjust pc relative offset for 2nd or 3rd insn. */    \
 673                 off -= insn - tmp_insns;                                \
 674                 /* Reject anything not fitting into insn->off. */       \
 675                 if (off < off_min || off > off_max)                     \
 676                         goto err;                                       \
 677                 insn->off = off;                                        \
 678         } while (0)
 679
 680                 case BPF_JMP | BPF_JA:
 681                         target = i + fp->k + 1;
 682                         insn->code = fp->code;
 683                         BPF_EMIT_JMP;
 684                         break;
 685
 686                 case BPF_JMP | BPF_JEQ | BPF_K:
 687                 case BPF_JMP | BPF_JEQ | BPF_X:
 688                 case BPF_JMP | BPF_JSET | BPF_K:
 689                 case BPF_JMP | BPF_JSET | BPF_X:
 690                 case BPF_JMP | BPF_JGT | BPF_K:
 691                 case BPF_JMP | BPF_JGT | BPF_X:
 692                 case BPF_JMP | BPF_JGE | BPF_K:
 693                 case BPF_JMP | BPF_JGE | BPF_X:
 694                         if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) {
 695                                 /* BPF immediates are signed, zero extend
 696                                  * immediate into tmp register and use it
 697                                  * in compare insn.
 698                                  */
 699                                 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k);
 700
 701                                 insn->dst_reg = BPF_REG_A;
 702                                 insn->src_reg = BPF_REG_TMP;
 703                                 bpf_src = BPF_X;
 704                         } else {
 705                                 insn->dst_reg = BPF_REG_A;
 706                                 insn->imm = fp->k;
 707                                 bpf_src = BPF_SRC(fp->code);
 708                                 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0;
 709                         }
 710
 711                         /* Common case where 'jump_false' is next insn. */
 712                         if (fp->jf == 0) {
 713                                 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 714                                 target = i + fp->jt + 1;
 715                                 BPF_EMIT_JMP;
 716                                 break;
 717                         }
 718
 719                         /* Convert some jumps when 'jump_true' is next insn. */
 720                         if (fp->jt == 0) {
 721                                 switch (BPF_OP(fp->code)) {
 722                                 case BPF_JEQ:
 723                                         insn->code = BPF_JMP | BPF_JNE | bpf_src;
 724                                         break;
 725                                 case BPF_JGT:
 726                                         insn->code = BPF_JMP | BPF_JLE | bpf_src;
 727                                         break;
 728                                 case BPF_JGE:
 729                                         insn->code = BPF_JMP | BPF_JLT | bpf_src;
 730                                         break;
 731                                 default:
 732                                         goto jmp_rest;
 733                                 }
 734
 735                                 target = i + fp->jf + 1;
 736                                 BPF_EMIT_JMP;
 737                                 break;
 738                         }
 739 jmp_rest:
 740                         /* Other jumps are mapped into two insns: Jxx and JA. */
 741                         target = i + fp->jt + 1;
 742                         insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src;
 743                         BPF_EMIT_JMP;
 744                         insn++;
 745
 746                         insn->code = BPF_JMP | BPF_JA;
 747                         target = i + fp->jf + 1;
 748                         BPF_EMIT_JMP;
 749                         break;
 750
 751                 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */
 752                 case BPF_LDX | BPF_MSH | BPF_B: {
 753                         struct sock_filter tmp = {
 754                                 .code   = BPF_LD | BPF_ABS | BPF_B,
 755                                 .k      = fp->k,
 756                         };
 757
 758                         *seen_ld_abs = true;
 759
 760                         /* X = A */
 761                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 762                         /* A = BPF_R0 = *(u8 *) (skb->data + K) */
 763                         convert_bpf_ld_abs(&tmp, &insn);
 764                         insn++;
 765                         /* A &= 0xf */
 766                         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf);
 767                         /* A <<= 2 */
 768                         *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2);
 769                         /* tmp = X */
 770                         *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X);
 771                         /* X = A */
 772                         *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 773                         /* A = tmp */
 774                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP);
 775                         break;
 776                 }
 777                 /* RET_K is remaped into 2 insns. RET_A case doesn't need an
 778                  * extra mov as BPF_REG_0 is already mapped into BPF_REG_A.
 779                  */
 780                 case BPF_RET | BPF_A:
 781                 case BPF_RET | BPF_K:
 782                         if (BPF_RVAL(fp->code) == BPF_K)
 783                                 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0,
 784                                                         0, fp->k);
 785                         *insn = BPF_EXIT_INSN();
 786                         break;
 787
 788                 /* Store to stack. */
 789                 case BPF_ST:
 790                 case BPF_STX:
 791                         stack_off = fp->k * 4  + 4;
 792                         *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) ==
 793                                             BPF_ST ? BPF_REG_A : BPF_REG_X,
 794                                             -stack_off);
 795                         /* check_load_and_stores() verifies that classic BPF can
 796                          * load from stack only after write, so tracking
 797                          * stack_depth for ST|STX insns is enough
 798                          */
 799                         if (new_prog && new_prog->aux->stack_depth < stack_off)
 800                                 new_prog->aux->stack_depth = stack_off;
 801                         break;
 802
 803                 /* Load from stack. */
 804                 case BPF_LD | BPF_MEM:
 805                 case BPF_LDX | BPF_MEM:
 806                         stack_off = fp->k * 4  + 4;
 807                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD  ?
 808                                             BPF_REG_A : BPF_REG_X, BPF_REG_FP,
 809                                             -stack_off);
 810                         break;
 811
 812                 /* A = K or X = K */
 813                 case BPF_LD | BPF_IMM:
 814                 case BPF_LDX | BPF_IMM:
 815                         *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ?
 816                                               BPF_REG_A : BPF_REG_X, fp->k);
 817                         break;
 818
 819                 /* X = A */
 820                 case BPF_MISC | BPF_TAX:
 821                         *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A);
 822                         break;
 823
 824                 /* A = X */
 825                 case BPF_MISC | BPF_TXA:
 826                         *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X);
 827                         break;
 828
 829                 /* A = skb->len or X = skb->len */
 830                 case BPF_LD | BPF_W | BPF_LEN:
 831                 case BPF_LDX | BPF_W | BPF_LEN:
 832                         *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ?
 833                                             BPF_REG_A : BPF_REG_X, BPF_REG_CTX,
 834                                             offsetof(struct sk_buff, len));
 835                         break;
 836
 837                 /* Access seccomp_data fields. */
 838                 case BPF_LDX | BPF_ABS | BPF_W:
 839                         /* A = *(u32 *) (ctx + K) */
 840                         *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k);
 841                         break;
 842
 843                 /* Unknown instruction. */
 844                 default:
 845                         goto err;
 846                 }
 847
 848                 insn++;
 849                 if (new_prog)
 850                         memcpy(new_insn, tmp_insns,
 851                                sizeof(*insn) * (insn - tmp_insns));
 852                 new_insn += insn - tmp_insns;
 853         }
 854
 855         if (!new_prog) {
 856                 /* Only calculating new length. */
 857                 *new_len = new_insn - first_insn;
 858                 if (*seen_ld_abs)
 859                         *new_len += 4; /* Prologue bits. */
 860                 return 0;
 861         }
 862
 863         pass++;
 864         if (new_flen != new_insn - first_insn) {
 865                 new_flen = new_insn - first_insn;
 866                 if (pass > 2)
 867                         goto err;
 868                 goto do_pass;
 869         }
 870
 871         kfree(addrs);
 872         BUG_ON(*new_len != new_flen);
 873         return 0;
 874 err:
 875         kfree(addrs);
 876         return -EINVAL;
 877 }
 878
 879 /* Security:
 880  *
 881  * As we dont want to clear mem[] array for each packet going through
 882  * __bpf_prog_run(), we check that filter loaded by user never try to read
 883  * a cell if not previously written, and we check all branches to be sure
 884  * a malicious user doesn't try to abuse us.
 885  */
 886 static int check_load_and_stores(const struct sock_filter *filter, int flen)
 887 {
 888         u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */
 889         int pc, ret = 0;
 890
 891         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 892
 893         masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL);
 894         if (!masks)
 895                 return -ENOMEM;
 896
 897         memset(masks, 0xff, flen * sizeof(*masks));
 898
 899         for (pc = 0; pc < flen; pc++) {
 900                 memvalid &= masks[pc];
 901
 902                 switch (filter[pc].code) {
 903                 case BPF_ST:
 904                 case BPF_STX:
 905                         memvalid |= (1 << filter[pc].k);
 906                         break;
 907                 case BPF_LD | BPF_MEM:
 908                 case BPF_LDX | BPF_MEM:
 909                         if (!(memvalid & (1 << filter[pc].k))) {
 910                                 ret = -EINVAL;
 911                                 goto error;
 912                         }
 913                         break;
 914                 case BPF_JMP | BPF_JA:
 915                         /* A jump must set masks on target */
 916                         masks[pc + 1 + filter[pc].k] &= memvalid;
 917                         memvalid = ~0;
 918                         break;
 919                 case BPF_JMP | BPF_JEQ | BPF_K:
 920                 case BPF_JMP | BPF_JEQ | BPF_X:
 921                 case BPF_JMP | BPF_JGE | BPF_K:
 922                 case BPF_JMP | BPF_JGE | BPF_X:
 923                 case BPF_JMP | BPF_JGT | BPF_K:
 924                 case BPF_JMP | BPF_JGT | BPF_X:
 925                 case BPF_JMP | BPF_JSET | BPF_K:
 926                 case BPF_JMP | BPF_JSET | BPF_X:
 927                         /* A jump must set masks on targets */
 928                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 929                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 930                         memvalid = ~0;
 931                         break;
 932                 }
 933         }
 934 error:
 935         kfree(masks);
 936         return ret;
 937 }
 938
 939 static bool chk_code_allowed(u16 code_to_probe)
 940 {
 941         static const bool codes[] = {
 942                 /* 32 bit ALU operations */
 943                 [BPF_ALU | BPF_ADD | BPF_K] = true,
 944                 [BPF_ALU | BPF_ADD | BPF_X] = true,
 945                 [BPF_ALU | BPF_SUB | BPF_K] = true,
 946                 [BPF_ALU | BPF_SUB | BPF_X] = true,
 947                 [BPF_ALU | BPF_MUL | BPF_K] = true,
 948                 [BPF_ALU | BPF_MUL | BPF_X] = true,
 949                 [BPF_ALU | BPF_DIV | BPF_K] = true,
 950                 [BPF_ALU | BPF_DIV | BPF_X] = true,
 951                 [BPF_ALU | BPF_MOD | BPF_K] = true,
 952                 [BPF_ALU | BPF_MOD | BPF_X] = true,
 953                 [BPF_ALU | BPF_AND | BPF_K] = true,
 954                 [BPF_ALU | BPF_AND | BPF_X] = true,
 955                 [BPF_ALU | BPF_OR | BPF_K] = true,
 956                 [BPF_ALU | BPF_OR | BPF_X] = true,
 957                 [BPF_ALU | BPF_XOR | BPF_K] = true,
 958                 [BPF_ALU | BPF_XOR | BPF_X] = true,
 959                 [BPF_ALU | BPF_LSH | BPF_K] = true,
 960                 [BPF_ALU | BPF_LSH | BPF_X] = true,
 961                 [BPF_ALU | BPF_RSH | BPF_K] = true,
 962                 [BPF_ALU | BPF_RSH | BPF_X] = true,
 963                 [BPF_ALU | BPF_NEG] = true,
 964                 /* Load instructions */
 965                 [BPF_LD | BPF_W | BPF_ABS] = true,
 966                 [BPF_LD | BPF_H | BPF_ABS] = true,
 967                 [BPF_LD | BPF_B | BPF_ABS] = true,
 968                 [BPF_LD | BPF_W | BPF_LEN] = true,
 969                 [BPF_LD | BPF_W | BPF_IND] = true,
 970                 [BPF_LD | BPF_H | BPF_IND] = true,
 971                 [BPF_LD | BPF_B | BPF_IND] = true,
 972                 [BPF_LD | BPF_IMM] = true,
 973                 [BPF_LD | BPF_MEM] = true,
 974                 [BPF_LDX | BPF_W | BPF_LEN] = true,
 975                 [BPF_LDX | BPF_B | BPF_MSH] = true,
 976                 [BPF_LDX | BPF_IMM] = true,
 977                 [BPF_LDX | BPF_MEM] = true,
 978                 /* Store instructions */
 979                 [BPF_ST] = true,
 980                 [BPF_STX] = true,
 981                 /* Misc instructions */
 982                 [BPF_MISC | BPF_TAX] = true,
 983                 [BPF_MISC | BPF_TXA] = true,
 984                 /* Return instructions */
 985                 [BPF_RET | BPF_K] = true,
 986                 [BPF_RET | BPF_A] = true,
 987                 /* Jump instructions */
 988                 [BPF_JMP | BPF_JA] = true,
 989                 [BPF_JMP | BPF_JEQ | BPF_K] = true,
 990                 [BPF_JMP | BPF_JEQ | BPF_X] = true,
 991                 [BPF_JMP | BPF_JGE | BPF_K] = true,
 992                 [BPF_JMP | BPF_JGE | BPF_X] = true,
 993                 [BPF_JMP | BPF_JGT | BPF_K] = true,
 994                 [BPF_JMP | BPF_JGT | BPF_X] = true,
 995                 [BPF_JMP | BPF_JSET | BPF_K] = true,
 996                 [BPF_JMP | BPF_JSET | BPF_X] = true,
 997         };
 998
 999         if (code_to_probe >= ARRAY_SIZE(codes))
1000                 return false;
1001
1002         return codes[code_to_probe];
1003 }
1004
1005 static bool bpf_check_basics_ok(const struct sock_filter *filter,
1006                                 unsigned int flen)
1007 {
1008         if (filter == NULL)
1009                 return false;
1010         if (flen == 0 || flen > BPF_MAXINSNS)
1011                 return false;
1012
1013         return true;
1014 }
1015
1016 /**
1017  *      bpf_check_classic - verify socket filter code
1018  *      @filter: filter to verify
1019  *      @flen: length of filter
1020  *
1021  * Check the user's filter code. If we let some ugly
1022  * filter code slip through kaboom! The filter must contain
1023  * no references or jumps that are out of range, no illegal
1024  * instructions, and must end with a RET instruction.
1025  *
1026  * All jumps are forward as they are not signed.
1027  *
1028  * Returns 0 if the rule set is legal or -EINVAL if not.
1029  */
1030 static int bpf_check_classic(const struct sock_filter *filter,
1031                              unsigned int flen)
1032 {
1033         bool anc_found;
1034         int pc;
1035
1036         /* Check the filter code now */
1037         for (pc = 0; pc < flen; pc++) {
1038                 const struct sock_filter *ftest = &filter[pc];
1039
1040                 /* May we actually operate on this code? */
1041                 if (!chk_code_allowed(ftest->code))
1042                         return -EINVAL;
1043
1044                 /* Some instructions need special checks */
1045                 switch (ftest->code) {
1046                 case BPF_ALU | BPF_DIV | BPF_K:
1047                 case BPF_ALU | BPF_MOD | BPF_K:
1048                         /* Check for division by zero */
1049                         if (ftest->k == 0)
1050                                 return -EINVAL;
1051                         break;
1052                 case BPF_ALU | BPF_LSH | BPF_K:
1053                 case BPF_ALU | BPF_RSH | BPF_K:
1054                         if (ftest->k >= 32)
1055                                 return -EINVAL;
1056                         break;
1057                 case BPF_LD | BPF_MEM:
1058                 case BPF_LDX | BPF_MEM:
1059                 case BPF_ST:
1060                 case BPF_STX:
1061                         /* Check for invalid memory addresses */
1062                         if (ftest->k >= BPF_MEMWORDS)
1063                                 return -EINVAL;
1064                         break;
1065                 case BPF_JMP | BPF_JA:
1066                         /* Note, the large ftest->k might cause loops.
1067                          * Compare this with conditional jumps below,
1068                          * where offsets are limited. --ANK (981016)
1069                          */
1070                         if (ftest->k >= (unsigned int)(flen - pc - 1))
1071                                 return -EINVAL;
1072                         break;
1073                 case BPF_JMP | BPF_JEQ | BPF_K:
1074                 case BPF_JMP | BPF_JEQ | BPF_X:
1075                 case BPF_JMP | BPF_JGE | BPF_K:
1076                 case BPF_JMP | BPF_JGE | BPF_X:
1077                 case BPF_JMP | BPF_JGT | BPF_K:
1078                 case BPF_JMP | BPF_JGT | BPF_X:
1079                 case BPF_JMP | BPF_JSET | BPF_K:
1080                 case BPF_JMP | BPF_JSET | BPF_X:
1081                         /* Both conditionals must be safe */
1082                         if (pc + ftest->jt + 1 >= flen ||
1083                             pc + ftest->jf + 1 >= flen)
1084                                 return -EINVAL;
1085                         break;
1086                 case BPF_LD | BPF_W | BPF_ABS:
1087                 case BPF_LD | BPF_H | BPF_ABS:
1088                 case BPF_LD | BPF_B | BPF_ABS:
1089                         anc_found = false;
1090                         if (bpf_anc_helper(ftest) & BPF_ANC)
1091                                 anc_found = true;
1092                         /* Ancillary operation unknown or unsupported */
1093                         if (anc_found == false && ftest->k >= SKF_AD_OFF)
1094                                 return -EINVAL;
1095                 }
1096         }
1097
1098         /* Last instruction must be a RET code */
1099         switch (filter[flen - 1].code) {
1100         case BPF_RET | BPF_K:
1101         case BPF_RET | BPF_A:
1102                 return check_load_and_stores(filter, flen);
1103         }
1104
1105         return -EINVAL;
1106 }
1107
1108 static int bpf_prog_store_orig_filter(struct bpf_prog *fp,
1109                                       const struct sock_fprog *fprog)
1110 {
1111         unsigned int fsize = bpf_classic_proglen(fprog);
1112         struct sock_fprog_kern *fkprog;
1113
1114         fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL);
1115         if (!fp->orig_prog)
1116                 return -ENOMEM;
1117
1118         fkprog = fp->orig_prog;
1119         fkprog->len = fprog->len;
1120
1121         fkprog->filter = kmemdup(fp->insns, fsize,
1122                                  GFP_KERNEL | __GFP_NOWARN);
1123         if (!fkprog->filter) {
1124                 kfree(fp->orig_prog);
1125                 return -ENOMEM;
1126         }
1127
1128         return 0;
1129 }
1130
1131 static void bpf_release_orig_filter(struct bpf_prog *fp)
1132 {
1133         struct sock_fprog_kern *fprog = fp->orig_prog;
1134
1135         if (fprog) {
1136                 kfree(fprog->filter);
1137                 kfree(fprog);
1138         }
1139 }
1140
1141 static void __bpf_prog_release(struct bpf_prog *prog)
1142 {
1143         if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) {
1144                 bpf_prog_put(prog);
1145         } else {
1146                 bpf_release_orig_filter(prog);
1147                 bpf_prog_free(prog);
1148         }
1149 }
1150
1151 static void __sk_filter_release(struct sk_filter *fp)
1152 {
1153         __bpf_prog_release(fp->prog);
1154         kfree(fp);
1155 }
1156
1157 /**
1158  *      sk_filter_release_rcu - Release a socket filter by rcu_head
1159  *      @rcu: rcu_head that contains the sk_filter to free
1160  */
1161 static void sk_filter_release_rcu(struct rcu_head *rcu)
1162 {
1163         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
1164
1165         __sk_filter_release(fp);
1166 }
1167
1168 /**
1169  *      sk_filter_release - release a socket filter
1170  *      @fp: filter to remove
1171  *
1172  *      Remove a filter from a socket and release its resources.
1173  */
1174 static void sk_filter_release(struct sk_filter *fp)
1175 {
1176         if (refcount_dec_and_test(&fp->refcnt))
1177                 call_rcu(&fp->rcu, sk_filter_release_rcu);
1178 }
1179
1180 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1181 {
1182         u32 filter_size = bpf_prog_size(fp->prog->len);
1183
1184         atomic_sub(filter_size, &sk->sk_omem_alloc);
1185         sk_filter_release(fp);
1186 }
1187
1188 /* try to charge the socket memory if there is space available
1189  * return true on success
1190  */
1191 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1192 {
1193         u32 filter_size = bpf_prog_size(fp->prog->len);
1194
1195         /* same check as in sock_kmalloc() */
1196         if (filter_size <= sysctl_optmem_max &&
1197             atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) {
1198                 atomic_add(filter_size, &sk->sk_omem_alloc);
1199                 return true;
1200         }
1201         return false;
1202 }
1203
1204 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1205 {
1206         if (!refcount_inc_not_zero(&fp->refcnt))
1207                 return false;
1208
1209         if (!__sk_filter_charge(sk, fp)) {
1210                 sk_filter_release(fp);
1211                 return false;
1212         }
1213         return true;
1214 }
1215
1216 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp)
1217 {
1218         struct sock_filter *old_prog;
1219         struct bpf_prog *old_fp;
1220         int err, new_len, old_len = fp->len;
1221         bool seen_ld_abs = false;
1222
1223         /* We are free to overwrite insns et al right here as it
1224          * won't be used at this point in time anymore internally
1225          * after the migration to the internal BPF instruction
1226          * representation.
1227          */
1228         BUILD_BUG_ON(sizeof(struct sock_filter) !=
1229                      sizeof(struct bpf_insn));
1230
1231         /* Conversion cannot happen on overlapping memory areas,
1232          * so we need to keep the user BPF around until the 2nd
1233          * pass. At this time, the user BPF is stored in fp->insns.
1234          */
1235         old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter),
1236                            GFP_KERNEL | __GFP_NOWARN);
1237         if (!old_prog) {
1238                 err = -ENOMEM;
1239                 goto out_err;
1240         }
1241
1242         /* 1st pass: calculate the new program length. */
1243         err = bpf_convert_filter(old_prog, old_len, NULL, &new_len,
1244                                  &seen_ld_abs);
1245         if (err)
1246                 goto out_err_free;
1247
1248         /* Expand fp for appending the new filter representation. */
1249         old_fp = fp;
1250         fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0);
1251         if (!fp) {
1252                 /* The old_fp is still around in case we couldn't
1253                  * allocate new memory, so uncharge on that one.
1254                  */
1255                 fp = old_fp;
1256                 err = -ENOMEM;
1257                 goto out_err_free;
1258         }
1259
1260         fp->len = new_len;
1261
1262         /* 2nd pass: remap sock_filter insns into bpf_insn insns. */
1263         err = bpf_convert_filter(old_prog, old_len, fp, &new_len,
1264                                  &seen_ld_abs);
1265         if (err)
1266                 /* 2nd bpf_convert_filter() can fail only if it fails
1267                  * to allocate memory, remapping must succeed. Note,
1268                  * that at this time old_fp has already been released
1269                  * by krealloc().
1270                  */
1271                 goto out_err_free;
1272
1273         fp = bpf_prog_select_runtime(fp, &err);
1274         if (err)
1275                 goto out_err_free;
1276
1277         kfree(old_prog);
1278         return fp;
1279
1280 out_err_free:
1281         kfree(old_prog);
1282 out_err:
1283         __bpf_prog_release(fp);
1284         return ERR_PTR(err);
1285 }
1286
1287 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp,
1288                                            bpf_aux_classic_check_t trans)
1289 {
1290         int err;
1291
1292         fp->bpf_func = NULL;
1293         fp->jited = 0;
1294
1295         err = bpf_check_classic(fp->insns, fp->len);
1296         if (err) {
1297                 __bpf_prog_release(fp);
1298                 return ERR_PTR(err);
1299         }
1300
1301         /* There might be additional checks and transformations
1302          * needed on classic filters, f.e. in case of seccomp.
1303          */
1304         if (trans) {
1305                 err = trans(fp->insns, fp->len);
1306                 if (err) {
1307                         __bpf_prog_release(fp);
1308                         return ERR_PTR(err);
1309                 }
1310         }
1311
1312         /* Probe if we can JIT compile the filter and if so, do
1313          * the compilation of the filter.
1314          */
1315         bpf_jit_compile(fp);
1316
1317         /* JIT compiler couldn't process this filter, so do the
1318          * internal BPF translation for the optimized interpreter.
1319          */
1320         if (!fp->jited)
1321                 fp = bpf_migrate_filter(fp);
1322
1323         return fp;
1324 }
1325
1326 /**
1327  *      bpf_prog_create - create an unattached filter
1328  *      @pfp: the unattached filter that is created
1329  *      @fprog: the filter program
1330  *
1331  * Create a filter independent of any socket. We first run some
1332  * sanity checks on it to make sure it does not explode on us later.
1333  * If an error occurs or there is insufficient memory for the filter
1334  * a negative errno code is returned. On success the return is zero.
1335  */
1336 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog)
1337 {
1338         unsigned int fsize = bpf_classic_proglen(fprog);
1339         struct bpf_prog *fp;
1340
1341         /* Make sure new filter is there and in the right amounts. */
1342         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1343                 return -EINVAL;
1344
1345         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1346         if (!fp)
1347                 return -ENOMEM;
1348
1349         memcpy(fp->insns, fprog->filter, fsize);
1350
1351         fp->len = fprog->len;
1352         /* Since unattached filters are not copied back to user
1353          * space through sk_get_filter(), we do not need to hold
1354          * a copy here, and can spare us the work.
1355          */
1356         fp->orig_prog = NULL;
1357
1358         /* bpf_prepare_filter() already takes care of freeing
1359          * memory in case something goes wrong.
1360          */
1361         fp = bpf_prepare_filter(fp, NULL);
1362         if (IS_ERR(fp))
1363                 return PTR_ERR(fp);
1364
1365         *pfp = fp;
1366         return 0;
1367 }
1368 EXPORT_SYMBOL_GPL(bpf_prog_create);
1369
1370 /**
1371  *      bpf_prog_create_from_user - create an unattached filter from user buffer
1372  *      @pfp: the unattached filter that is created
1373  *      @fprog: the filter program
1374  *      @trans: post-classic verifier transformation handler
1375  *      @save_orig: save classic BPF program
1376  *
1377  * This function effectively does the same as bpf_prog_create(), only
1378  * that it builds up its insns buffer from user space provided buffer.
1379  * It also allows for passing a bpf_aux_classic_check_t handler.
1380  */
1381 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog,
1382                               bpf_aux_classic_check_t trans, bool save_orig)
1383 {
1384         unsigned int fsize = bpf_classic_proglen(fprog);
1385         struct bpf_prog *fp;
1386         int err;
1387
1388         /* Make sure new filter is there and in the right amounts. */
1389         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1390                 return -EINVAL;
1391
1392         fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1393         if (!fp)
1394                 return -ENOMEM;
1395
1396         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
1397                 __bpf_prog_free(fp);
1398                 return -EFAULT;
1399         }
1400
1401         fp->len = fprog->len;
1402         fp->orig_prog = NULL;
1403
1404         if (save_orig) {
1405                 err = bpf_prog_store_orig_filter(fp, fprog);
1406                 if (err) {
1407                         __bpf_prog_free(fp);
1408                         return -ENOMEM;
1409                 }
1410         }
1411
1412         /* bpf_prepare_filter() already takes care of freeing
1413          * memory in case something goes wrong.
1414          */
1415         fp = bpf_prepare_filter(fp, trans);
1416         if (IS_ERR(fp))
1417                 return PTR_ERR(fp);
1418
1419         *pfp = fp;
1420         return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user);
1423
1424 void bpf_prog_destroy(struct bpf_prog *fp)
1425 {
1426         __bpf_prog_release(fp);
1427 }
1428 EXPORT_SYMBOL_GPL(bpf_prog_destroy);
1429
1430 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk)
1431 {
1432         struct sk_filter *fp, *old_fp;
1433
1434         fp = kmalloc(sizeof(*fp), GFP_KERNEL);
1435         if (!fp)
1436                 return -ENOMEM;
1437
1438         fp->prog = prog;
1439
1440         if (!__sk_filter_charge(sk, fp)) {
1441                 kfree(fp);
1442                 return -ENOMEM;
1443         }
1444         refcount_set(&fp->refcnt, 1);
1445
1446         old_fp = rcu_dereference_protected(sk->sk_filter,
1447                                            lockdep_sock_is_held(sk));
1448         rcu_assign_pointer(sk->sk_filter, fp);
1449
1450         if (old_fp)
1451                 sk_filter_uncharge(sk, old_fp);
1452
1453         return 0;
1454 }
1455
1456 static
1457 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk)
1458 {
1459         unsigned int fsize = bpf_classic_proglen(fprog);
1460         struct bpf_prog *prog;
1461         int err;
1462
1463         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1464                 return ERR_PTR(-EPERM);
1465
1466         /* Make sure new filter is there and in the right amounts. */
1467         if (!bpf_check_basics_ok(fprog->filter, fprog->len))
1468                 return ERR_PTR(-EINVAL);
1469
1470         prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0);
1471         if (!prog)
1472                 return ERR_PTR(-ENOMEM);
1473
1474         if (copy_from_user(prog->insns, fprog->filter, fsize)) {
1475                 __bpf_prog_free(prog);
1476                 return ERR_PTR(-EFAULT);
1477         }
1478
1479         prog->len = fprog->len;
1480
1481         err = bpf_prog_store_orig_filter(prog, fprog);
1482         if (err) {
1483                 __bpf_prog_free(prog);
1484                 return ERR_PTR(-ENOMEM);
1485         }
1486
1487         /* bpf_prepare_filter() already takes care of freeing
1488          * memory in case something goes wrong.
1489          */
1490         return bpf_prepare_filter(prog, NULL);
1491 }
1492
1493 /**
1494  *      sk_attach_filter - attach a socket filter
1495  *      @fprog: the filter program
1496  *      @sk: the socket to use
1497  *
1498  * Attach the user's filter code. We first run some sanity checks on
1499  * it to make sure it does not explode on us later. If an error
1500  * occurs or there is insufficient memory for the filter a negative
1501  * errno code is returned. On success the return is zero.
1502  */
1503 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1504 {
1505         struct bpf_prog *prog = __get_filter(fprog, sk);
1506         int err;
1507
1508         if (IS_ERR(prog))
1509                 return PTR_ERR(prog);
1510
1511         err = __sk_attach_prog(prog, sk);
1512         if (err < 0) {
1513                 __bpf_prog_release(prog);
1514                 return err;
1515         }
1516
1517         return 0;
1518 }
1519 EXPORT_SYMBOL_GPL(sk_attach_filter);
1520
1521 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk)
1522 {
1523         struct bpf_prog *prog = __get_filter(fprog, sk);
1524         int err;
1525
1526         if (IS_ERR(prog))
1527                 return PTR_ERR(prog);
1528
1529         if (bpf_prog_size(prog->len) > sysctl_optmem_max)
1530                 err = -ENOMEM;
1531         else
1532                 err = reuseport_attach_prog(sk, prog);
1533
1534         if (err)
1535                 __bpf_prog_release(prog);
1536
1537         return err;
1538 }
1539
1540 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk)
1541 {
1542         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1543                 return ERR_PTR(-EPERM);
1544
1545         return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1546 }
1547
1548 int sk_attach_bpf(u32 ufd, struct sock *sk)
1549 {
1550         struct bpf_prog *prog = __get_bpf(ufd, sk);
1551         int err;
1552
1553         if (IS_ERR(prog))
1554                 return PTR_ERR(prog);
1555
1556         err = __sk_attach_prog(prog, sk);
1557         if (err < 0) {
1558                 bpf_prog_put(prog);
1559                 return err;
1560         }
1561
1562         return 0;
1563 }
1564
1565 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk)
1566 {
1567         struct bpf_prog *prog;
1568         int err;
1569
1570         if (sock_flag(sk, SOCK_FILTER_LOCKED))
1571                 return -EPERM;
1572
1573         prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER);
1574         if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL)
1575                 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT);
1576         if (IS_ERR(prog))
1577                 return PTR_ERR(prog);
1578
1579         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) {
1580                 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER
1581                  * bpf prog (e.g. sockmap).  It depends on the
1582                  * limitation imposed by bpf_prog_load().
1583                  * Hence, sysctl_optmem_max is not checked.
1584                  */
1585                 if ((sk->sk_type != SOCK_STREAM &&
1586                      sk->sk_type != SOCK_DGRAM) ||
1587                     (sk->sk_protocol != IPPROTO_UDP &&
1588                      sk->sk_protocol != IPPROTO_TCP) ||
1589                     (sk->sk_family != AF_INET &&
1590                      sk->sk_family != AF_INET6)) {
1591                         err = -ENOTSUPP;
1592                         goto err_prog_put;
1593                 }
1594         } else {
1595                 /* BPF_PROG_TYPE_SOCKET_FILTER */
1596                 if (bpf_prog_size(prog->len) > sysctl_optmem_max) {
1597                         err = -ENOMEM;
1598                         goto err_prog_put;
1599                 }
1600         }
1601
1602         err = reuseport_attach_prog(sk, prog);
1603 err_prog_put:
1604         if (err)
1605                 bpf_prog_put(prog);
1606
1607         return err;
1608 }
1609
1610 void sk_reuseport_prog_free(struct bpf_prog *prog)
1611 {
1612         if (!prog)
1613                 return;
1614
1615         if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT)
1616                 bpf_prog_put(prog);
1617         else
1618                 bpf_prog_destroy(prog);
1619 }
1620
1621 struct bpf_scratchpad {
1622         union {
1623                 __be32 diff[MAX_BPF_STACK / sizeof(__be32)];
1624                 u8     buff[MAX_BPF_STACK];
1625         };
1626 };
1627
1628 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
1629
1630 static inline int __bpf_try_make_writable(struct sk_buff *skb,
1631                                           unsigned int write_len)
1632 {
1633         return skb_ensure_writable(skb, write_len);
1634 }
1635
1636 static inline int bpf_try_make_writable(struct sk_buff *skb,
1637                                         unsigned int write_len)
1638 {
1639         int err = __bpf_try_make_writable(skb, write_len);
1640
1641         bpf_compute_data_pointers(skb);
1642         return err;
1643 }
1644
1645 static int bpf_try_make_head_writable(struct sk_buff *skb)
1646 {
1647         return bpf_try_make_writable(skb, skb_headlen(skb));
1648 }
1649
1650 static inline void bpf_push_mac_rcsum(struct sk_buff *skb)
1651 {
1652         if (skb_at_tc_ingress(skb))
1653                 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1654 }
1655
1656 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb)
1657 {
1658         if (skb_at_tc_ingress(skb))
1659                 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len);
1660 }
1661
1662 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset,
1663            const void *, from, u32, len, u64, flags)
1664 {
1665         void *ptr;
1666
1667         if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH)))
1668                 return -EINVAL;
1669         if (unlikely(offset > 0xffff))
1670                 return -EFAULT;
1671         if (unlikely(bpf_try_make_writable(skb, offset + len)))
1672                 return -EFAULT;
1673
1674         ptr = skb->data + offset;
1675         if (flags & BPF_F_RECOMPUTE_CSUM)
1676                 __skb_postpull_rcsum(skb, ptr, len, offset);
1677
1678         memcpy(ptr, from, len);
1679
1680         if (flags & BPF_F_RECOMPUTE_CSUM)
1681                 __skb_postpush_rcsum(skb, ptr, len, offset);
1682         if (flags & BPF_F_INVALIDATE_HASH)
1683                 skb_clear_hash(skb);
1684
1685         return 0;
1686 }
1687
1688 static const struct bpf_func_proto bpf_skb_store_bytes_proto = {
1689         .func           = bpf_skb_store_bytes,
1690         .gpl_only       = false,
1691         .ret_type       = RET_INTEGER,
1692         .arg1_type      = ARG_PTR_TO_CTX,
1693         .arg2_type      = ARG_ANYTHING,
1694         .arg3_type      = ARG_PTR_TO_MEM,
1695         .arg4_type      = ARG_CONST_SIZE,
1696         .arg5_type      = ARG_ANYTHING,
1697 };
1698
1699 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset,
1700            void *, to, u32, len)
1701 {
1702         void *ptr;
1703
1704         if (unlikely(offset > 0xffff))
1705                 goto err_clear;
1706
1707         ptr = skb_header_pointer(skb, offset, len, to);
1708         if (unlikely(!ptr))
1709                 goto err_clear;
1710         if (ptr != to)
1711                 memcpy(to, ptr, len);
1712
1713         return 0;
1714 err_clear:
1715         memset(to, 0, len);
1716         return -EFAULT;
1717 }
1718
1719 static const struct bpf_func_proto bpf_skb_load_bytes_proto = {
1720         .func           = bpf_skb_load_bytes,
1721         .gpl_only       = false,
1722         .ret_type       = RET_INTEGER,
1723         .arg1_type      = ARG_PTR_TO_CTX,
1724         .arg2_type      = ARG_ANYTHING,
1725         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1726         .arg4_type      = ARG_CONST_SIZE,
1727 };
1728
1729 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb,
1730            u32, offset, void *, to, u32, len, u32, start_header)
1731 {
1732         u8 *end = skb_tail_pointer(skb);
1733         u8 *net = skb_network_header(skb);
1734         u8 *mac = skb_mac_header(skb);
1735         u8 *ptr;
1736
1737         if (unlikely(offset > 0xffff || len > (end - mac)))
1738                 goto err_clear;
1739
1740         switch (start_header) {
1741         case BPF_HDR_START_MAC:
1742                 ptr = mac + offset;
1743                 break;
1744         case BPF_HDR_START_NET:
1745                 ptr = net + offset;
1746                 break;
1747         default:
1748                 goto err_clear;
1749         }
1750
1751         if (likely(ptr >= mac && ptr + len <= end)) {
1752                 memcpy(to, ptr, len);
1753                 return 0;
1754         }
1755
1756 err_clear:
1757         memset(to, 0, len);
1758         return -EFAULT;
1759 }
1760
1761 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = {
1762         .func           = bpf_skb_load_bytes_relative,
1763         .gpl_only       = false,
1764         .ret_type       = RET_INTEGER,
1765         .arg1_type      = ARG_PTR_TO_CTX,
1766         .arg2_type      = ARG_ANYTHING,
1767         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
1768         .arg4_type      = ARG_CONST_SIZE,
1769         .arg5_type      = ARG_ANYTHING,
1770 };
1771
1772 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len)
1773 {
1774         /* Idea is the following: should the needed direct read/write
1775          * test fail during runtime, we can pull in more data and redo
1776          * again, since implicitly, we invalidate previous checks here.
1777          *
1778          * Or, since we know how much we need to make read/writeable,
1779          * this can be done once at the program beginning for direct
1780          * access case. By this we overcome limitations of only current
1781          * headroom being accessible.
1782          */
1783         return bpf_try_make_writable(skb, len ? : skb_headlen(skb));
1784 }
1785
1786 static const struct bpf_func_proto bpf_skb_pull_data_proto = {
1787         .func           = bpf_skb_pull_data,
1788         .gpl_only       = false,
1789         .ret_type       = RET_INTEGER,
1790         .arg1_type      = ARG_PTR_TO_CTX,
1791         .arg2_type      = ARG_ANYTHING,
1792 };
1793
1794 static inline int sk_skb_try_make_writable(struct sk_buff *skb,
1795                                            unsigned int write_len)
1796 {
1797         int err = __bpf_try_make_writable(skb, write_len);
1798
1799         bpf_compute_data_end_sk_skb(skb);
1800         return err;
1801 }
1802
1803 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len)
1804 {
1805         /* Idea is the following: should the needed direct read/write
1806          * test fail during runtime, we can pull in more data and redo
1807          * again, since implicitly, we invalidate previous checks here.
1808          *
1809          * Or, since we know how much we need to make read/writeable,
1810          * this can be done once at the program beginning for direct
1811          * access case. By this we overcome limitations of only current
1812          * headroom being accessible.
1813          */
1814         return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb));
1815 }
1816
1817 static const struct bpf_func_proto sk_skb_pull_data_proto = {
1818         .func           = sk_skb_pull_data,
1819         .gpl_only       = false,
1820         .ret_type       = RET_INTEGER,
1821         .arg1_type      = ARG_PTR_TO_CTX,
1822         .arg2_type      = ARG_ANYTHING,
1823 };
1824
1825 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset,
1826            u64, from, u64, to, u64, flags)
1827 {
1828         __sum16 *ptr;
1829
1830         if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK)))
1831                 return -EINVAL;
1832         if (unlikely(offset > 0xffff || offset & 1))
1833                 return -EFAULT;
1834         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1835                 return -EFAULT;
1836
1837         ptr = (__sum16 *)(skb->data + offset);
1838         switch (flags & BPF_F_HDR_FIELD_MASK) {
1839         case 0:
1840                 if (unlikely(from != 0))
1841                         return -EINVAL;
1842
1843                 csum_replace_by_diff(ptr, to);
1844                 break;
1845         case 2:
1846                 csum_replace2(ptr, from, to);
1847                 break;
1848         case 4:
1849                 csum_replace4(ptr, from, to);
1850                 break;
1851         default:
1852                 return -EINVAL;
1853         }
1854
1855         return 0;
1856 }
1857
1858 static const struct bpf_func_proto bpf_l3_csum_replace_proto = {
1859         .func           = bpf_l3_csum_replace,
1860         .gpl_only       = false,
1861         .ret_type       = RET_INTEGER,
1862         .arg1_type      = ARG_PTR_TO_CTX,
1863         .arg2_type      = ARG_ANYTHING,
1864         .arg3_type      = ARG_ANYTHING,
1865         .arg4_type      = ARG_ANYTHING,
1866         .arg5_type      = ARG_ANYTHING,
1867 };
1868
1869 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset,
1870            u64, from, u64, to, u64, flags)
1871 {
1872         bool is_pseudo = flags & BPF_F_PSEUDO_HDR;
1873         bool is_mmzero = flags & BPF_F_MARK_MANGLED_0;
1874         bool do_mforce = flags & BPF_F_MARK_ENFORCE;
1875         __sum16 *ptr;
1876
1877         if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE |
1878                                BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK)))
1879                 return -EINVAL;
1880         if (unlikely(offset > 0xffff || offset & 1))
1881                 return -EFAULT;
1882         if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr))))
1883                 return -EFAULT;
1884
1885         ptr = (__sum16 *)(skb->data + offset);
1886         if (is_mmzero && !do_mforce && !*ptr)
1887                 return 0;
1888
1889         switch (flags & BPF_F_HDR_FIELD_MASK) {
1890         case 0:
1891                 if (unlikely(from != 0))
1892                         return -EINVAL;
1893
1894                 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo);
1895                 break;
1896         case 2:
1897                 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo);
1898                 break;
1899         case 4:
1900                 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo);
1901                 break;
1902         default:
1903                 return -EINVAL;
1904         }
1905
1906         if (is_mmzero && !*ptr)
1907                 *ptr = CSUM_MANGLED_0;
1908         return 0;
1909 }
1910
1911 static const struct bpf_func_proto bpf_l4_csum_replace_proto = {
1912         .func           = bpf_l4_csum_replace,
1913         .gpl_only       = false,
1914         .ret_type       = RET_INTEGER,
1915         .arg1_type      = ARG_PTR_TO_CTX,
1916         .arg2_type      = ARG_ANYTHING,
1917         .arg3_type      = ARG_ANYTHING,
1918         .arg4_type      = ARG_ANYTHING,
1919         .arg5_type      = ARG_ANYTHING,
1920 };
1921
1922 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
1923            __be32 *, to, u32, to_size, __wsum, seed)
1924 {
1925         struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
1926         u32 diff_size = from_size + to_size;
1927         int i, j = 0;
1928
1929         /* This is quite flexible, some examples:
1930          *
1931          * from_size == 0, to_size > 0,  seed := csum --> pushing data
1932          * from_size > 0,  to_size == 0, seed := csum --> pulling data
1933          * from_size > 0,  to_size > 0,  seed := 0    --> diffing data
1934          *
1935          * Even for diffing, from_size and to_size don't need to be equal.
1936          */
1937         if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) ||
1938                      diff_size > sizeof(sp->diff)))
1939                 return -EINVAL;
1940
1941         for (i = 0; i < from_size / sizeof(__be32); i++, j++)
1942                 sp->diff[j] = ~from[i];
1943         for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
1944                 sp->diff[j] = to[i];
1945
1946         return csum_partial(sp->diff, diff_size, seed);
1947 }
1948
1949 static const struct bpf_func_proto bpf_csum_diff_proto = {
1950         .func           = bpf_csum_diff,
1951         .gpl_only       = false,
1952         .pkt_access     = true,
1953         .ret_type       = RET_INTEGER,
1954         .arg1_type      = ARG_PTR_TO_MEM_OR_NULL,
1955         .arg2_type      = ARG_CONST_SIZE_OR_ZERO,
1956         .arg3_type      = ARG_PTR_TO_MEM_OR_NULL,
1957         .arg4_type      = ARG_CONST_SIZE_OR_ZERO,
1958         .arg5_type      = ARG_ANYTHING,
1959 };
1960
1961 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum)
1962 {
1963         /* The interface is to be used in combination with bpf_csum_diff()
1964          * for direct packet writes. csum rotation for alignment as well
1965          * as emulating csum_sub() can be done from the eBPF program.
1966          */
1967         if (skb->ip_summed == CHECKSUM_COMPLETE)
1968                 return (skb->csum = csum_add(skb->csum, csum));
1969
1970         return -ENOTSUPP;
1971 }
1972
1973 static const struct bpf_func_proto bpf_csum_update_proto = {
1974         .func           = bpf_csum_update,
1975         .gpl_only       = false,
1976         .ret_type       = RET_INTEGER,
1977         .arg1_type      = ARG_PTR_TO_CTX,
1978         .arg2_type      = ARG_ANYTHING,
1979 };
1980
1981 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb)
1982 {
1983         return dev_forward_skb(dev, skb);
1984 }
1985
1986 static inline int __bpf_rx_skb_no_mac(struct net_device *dev,
1987                                       struct sk_buff *skb)
1988 {
1989         int ret = ____dev_forward_skb(dev, skb);
1990
1991         if (likely(!ret)) {
1992                 skb->dev = dev;
1993                 ret = netif_rx(skb);
1994         }
1995
1996         return ret;
1997 }
1998
1999 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb)
2000 {
2001         int ret;
2002
2003         if (unlikely(__this_cpu_read(xmit_recursion) > XMIT_RECURSION_LIMIT)) {
2004                 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n");
2005                 kfree_skb(skb);
2006                 return -ENETDOWN;
2007         }
2008
2009         skb->dev = dev;
2010
2011         __this_cpu_inc(xmit_recursion);
2012         ret = dev_queue_xmit(skb);
2013         __this_cpu_dec(xmit_recursion);
2014
2015         return ret;
2016 }
2017
2018 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev,
2019                                  u32 flags)
2020 {
2021         /* skb->mac_len is not set on normal egress */
2022         unsigned int mlen = skb->network_header - skb->mac_header;
2023
2024         __skb_pull(skb, mlen);
2025
2026         /* At ingress, the mac header has already been pulled once.
2027          * At egress, skb_pospull_rcsum has to be done in case that
2028          * the skb is originated from ingress (i.e. a forwarded skb)
2029          * to ensure that rcsum starts at net header.
2030          */
2031         if (!skb_at_tc_ingress(skb))
2032                 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen);
2033         skb_pop_mac_header(skb);
2034         skb_reset_mac_len(skb);
2035         return flags & BPF_F_INGRESS ?
2036                __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb);
2037 }
2038
2039 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev,
2040                                  u32 flags)
2041 {
2042         /* Verify that a link layer header is carried */
2043         if (unlikely(skb->mac_header >= skb->network_header)) {
2044                 kfree_skb(skb);
2045                 return -ERANGE;
2046         }
2047
2048         bpf_push_mac_rcsum(skb);
2049         return flags & BPF_F_INGRESS ?
2050                __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb);
2051 }
2052
2053 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev,
2054                           u32 flags)
2055 {
2056         if (dev_is_mac_header_xmit(dev))
2057                 return __bpf_redirect_common(skb, dev, flags);
2058         else
2059                 return __bpf_redirect_no_mac(skb, dev, flags);
2060 }
2061
2062 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags)
2063 {
2064         struct net_device *dev;
2065         struct sk_buff *clone;
2066         int ret;
2067
2068         if (unlikely(flags & ~(BPF_F_INGRESS)))
2069                 return -EINVAL;
2070
2071         dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex);
2072         if (unlikely(!dev))
2073                 return -EINVAL;
2074
2075         clone = skb_clone(skb, GFP_ATOMIC);
2076         if (unlikely(!clone))
2077                 return -ENOMEM;
2078
2079         /* For direct write, we need to keep the invariant that the skbs
2080          * we're dealing with need to be uncloned. Should uncloning fail
2081          * here, we need to free the just generated clone to unclone once
2082          * again.
2083          */
2084         ret = bpf_try_make_head_writable(skb);
2085         if (unlikely(ret)) {
2086                 kfree_skb(clone);
2087                 return -ENOMEM;
2088         }
2089
2090         return __bpf_redirect(clone, dev, flags);
2091 }
2092
2093 static const struct bpf_func_proto bpf_clone_redirect_proto = {
2094         .func           = bpf_clone_redirect,
2095         .gpl_only       = false,
2096         .ret_type       = RET_INTEGER,
2097         .arg1_type      = ARG_PTR_TO_CTX,
2098         .arg2_type      = ARG_ANYTHING,
2099         .arg3_type      = ARG_ANYTHING,
2100 };
2101
2102 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
2103 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
2104
2105 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
2106 {
2107         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2108
2109         if (unlikely(flags & ~(BPF_F_INGRESS)))
2110                 return TC_ACT_SHOT;
2111
2112         ri->ifindex = ifindex;
2113         ri->flags = flags;
2114
2115         return TC_ACT_REDIRECT;
2116 }
2117
2118 int skb_do_redirect(struct sk_buff *skb)
2119 {
2120         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
2121         struct net_device *dev;
2122
2123         dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->ifindex);
2124         ri->ifindex = 0;
2125         if (unlikely(!dev)) {
2126                 kfree_skb(skb);
2127                 return -EINVAL;
2128         }
2129
2130         return __bpf_redirect(skb, dev, ri->flags);
2131 }
2132
2133 static const struct bpf_func_proto bpf_redirect_proto = {
2134         .func           = bpf_redirect,
2135         .gpl_only       = false,
2136         .ret_type       = RET_INTEGER,
2137         .arg1_type      = ARG_ANYTHING,
2138         .arg2_type      = ARG_ANYTHING,
2139 };
2140
2141 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb,
2142            struct bpf_map *, map, void *, key, u64, flags)
2143 {
2144         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2145
2146         /* If user passes invalid input drop the packet. */
2147         if (unlikely(flags & ~(BPF_F_INGRESS)))
2148                 return SK_DROP;
2149
2150         tcb->bpf.flags = flags;
2151         tcb->bpf.sk_redir = __sock_hash_lookup_elem(map, key);
2152         if (!tcb->bpf.sk_redir)
2153                 return SK_DROP;
2154
2155         return SK_PASS;
2156 }
2157
2158 static const struct bpf_func_proto bpf_sk_redirect_hash_proto = {
2159         .func           = bpf_sk_redirect_hash,
2160         .gpl_only       = false,
2161         .ret_type       = RET_INTEGER,
2162         .arg1_type      = ARG_PTR_TO_CTX,
2163         .arg2_type      = ARG_CONST_MAP_PTR,
2164         .arg3_type      = ARG_PTR_TO_MAP_KEY,
2165         .arg4_type      = ARG_ANYTHING,
2166 };
2167
2168 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb,
2169            struct bpf_map *, map, u32, key, u64, flags)
2170 {
2171         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2172
2173         /* If user passes invalid input drop the packet. */
2174         if (unlikely(flags & ~(BPF_F_INGRESS)))
2175                 return SK_DROP;
2176
2177         tcb->bpf.flags = flags;
2178         tcb->bpf.sk_redir = __sock_map_lookup_elem(map, key);
2179         if (!tcb->bpf.sk_redir)
2180                 return SK_DROP;
2181
2182         return SK_PASS;
2183 }
2184
2185 struct sock *do_sk_redirect_map(struct sk_buff *skb)
2186 {
2187         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
2188
2189         return tcb->bpf.sk_redir;
2190 }
2191
2192 static const struct bpf_func_proto bpf_sk_redirect_map_proto = {
2193         .func           = bpf_sk_redirect_map,
2194         .gpl_only       = false,
2195         .ret_type       = RET_INTEGER,
2196         .arg1_type      = ARG_PTR_TO_CTX,
2197         .arg2_type      = ARG_CONST_MAP_PTR,
2198         .arg3_type      = ARG_ANYTHING,
2199         .arg4_type      = ARG_ANYTHING,
2200 };
2201
2202 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg_buff *, msg,
2203            struct bpf_map *, map, void *, key, u64, flags)
2204 {
2205         /* If user passes invalid input drop the packet. */
2206         if (unlikely(flags & ~(BPF_F_INGRESS)))
2207                 return SK_DROP;
2208
2209         msg->flags = flags;
2210         msg->sk_redir = __sock_hash_lookup_elem(map, key);
2211         if (!msg->sk_redir)
2212                 return SK_DROP;
2213
2214         return SK_PASS;
2215 }
2216
2217 static const struct bpf_func_proto bpf_msg_redirect_hash_proto = {
2218         .func           = bpf_msg_redirect_hash,
2219         .gpl_only       = false,
2220         .ret_type       = RET_INTEGER,
2221         .arg1_type      = ARG_PTR_TO_CTX,
2222         .arg2_type      = ARG_CONST_MAP_PTR,
2223         .arg3_type      = ARG_PTR_TO_MAP_KEY,
2224         .arg4_type      = ARG_ANYTHING,
2225 };
2226
2227 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg_buff *, msg,
2228            struct bpf_map *, map, u32, key, u64, flags)
2229 {
2230         /* If user passes invalid input drop the packet. */
2231         if (unlikely(flags & ~(BPF_F_INGRESS)))
2232                 return SK_DROP;
2233
2234         msg->flags = flags;
2235         msg->sk_redir = __sock_map_lookup_elem(map, key);
2236         if (!msg->sk_redir)
2237                 return SK_DROP;
2238
2239         return SK_PASS;
2240 }
2241
2242 struct sock *do_msg_redirect_map(struct sk_msg_buff *msg)
2243 {
2244         return msg->sk_redir;
2245 }
2246
2247 static const struct bpf_func_proto bpf_msg_redirect_map_proto = {
2248         .func           = bpf_msg_redirect_map,
2249         .gpl_only       = false,
2250         .ret_type       = RET_INTEGER,
2251         .arg1_type      = ARG_PTR_TO_CTX,
2252         .arg2_type      = ARG_CONST_MAP_PTR,
2253         .arg3_type      = ARG_ANYTHING,
2254         .arg4_type      = ARG_ANYTHING,
2255 };
2256
2257 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg_buff *, msg, u32, bytes)
2258 {
2259         msg->apply_bytes = bytes;
2260         return 0;
2261 }
2262
2263 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = {
2264         .func           = bpf_msg_apply_bytes,
2265         .gpl_only       = false,
2266         .ret_type       = RET_INTEGER,
2267         .arg1_type      = ARG_PTR_TO_CTX,
2268         .arg2_type      = ARG_ANYTHING,
2269 };
2270
2271 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg_buff *, msg, u32, bytes)
2272 {
2273         msg->cork_bytes = bytes;
2274         return 0;
2275 }
2276
2277 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = {
2278         .func           = bpf_msg_cork_bytes,
2279         .gpl_only       = false,
2280         .ret_type       = RET_INTEGER,
2281         .arg1_type      = ARG_PTR_TO_CTX,
2282         .arg2_type      = ARG_ANYTHING,
2283 };
2284
2285 BPF_CALL_4(bpf_msg_pull_data,
2286            struct sk_msg_buff *, msg, u32, start, u32, end, u64, flags)
2287 {
2288         unsigned int len = 0, offset = 0, copy = 0;
2289         struct scatterlist *sg = msg->sg_data;
2290         int first_sg, last_sg, i, shift;
2291         unsigned char *p, *to, *from;
2292         int bytes = end - start;
2293         struct page *page;
2294
2295         if (unlikely(flags || end <= start))
2296                 return -EINVAL;
2297
2298         /* First find the starting scatterlist element */
2299         i = msg->sg_start;
2300         do {
2301                 len = sg[i].length;
2302                 offset += len;
2303                 if (start < offset + len)
2304                         break;
2305                 i++;
2306                 if (i == MAX_SKB_FRAGS)
2307                         i = 0;
2308         } while (i != msg->sg_end);
2309
2310         if (unlikely(start >= offset + len))
2311                 return -EINVAL;
2312
2313         if (!msg->sg_copy[i] && bytes <= len)
2314                 goto out;
2315
2316         first_sg = i;
2317
2318         /* At this point we need to linearize multiple scatterlist
2319          * elements or a single shared page. Either way we need to
2320          * copy into a linear buffer exclusively owned by BPF. Then
2321          * place the buffer in the scatterlist and fixup the original
2322          * entries by removing the entries now in the linear buffer
2323          * and shifting the remaining entries. For now we do not try
2324          * to copy partial entries to avoid complexity of running out
2325          * of sg_entry slots. The downside is reading a single byte
2326          * will copy the entire sg entry.
2327          */
2328         do {
2329                 copy += sg[i].length;
2330                 i++;
2331                 if (i == MAX_SKB_FRAGS)
2332                         i = 0;
2333                 if (bytes < copy)
2334                         break;
2335         } while (i != msg->sg_end);
2336         last_sg = i;
2337
2338         if (unlikely(copy < end - start))
2339                 return -EINVAL;
2340
2341         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC, get_order(copy));
2342         if (unlikely(!page))
2343                 return -ENOMEM;
2344         p = page_address(page);
2345         offset = 0;
2346
2347         i = first_sg;
2348         do {
2349                 from = sg_virt(&sg[i]);
2350                 len = sg[i].length;
2351                 to = p + offset;
2352
2353                 memcpy(to, from, len);
2354                 offset += len;
2355                 sg[i].length = 0;
2356                 put_page(sg_page(&sg[i]));
2357
2358                 i++;
2359                 if (i == MAX_SKB_FRAGS)
2360                         i = 0;
2361         } while (i != last_sg);
2362
2363         sg[first_sg].length = copy;
2364         sg_set_page(&sg[first_sg], page, copy, 0);
2365
2366         /* To repair sg ring we need to shift entries. If we only
2367          * had a single entry though we can just replace it and
2368          * be done. Otherwise walk the ring and shift the entries.
2369          */
2370         shift = last_sg - first_sg - 1;
2371         if (!shift)
2372                 goto out;
2373
2374         i = first_sg + 1;
2375         do {
2376                 int move_from;
2377
2378                 if (i + shift >= MAX_SKB_FRAGS)
2379                         move_from = i + shift - MAX_SKB_FRAGS;
2380                 else
2381                         move_from = i + shift;
2382
2383                 if (move_from == msg->sg_end)
2384                         break;
2385
2386                 sg[i] = sg[move_from];
2387                 sg[move_from].length = 0;
2388                 sg[move_from].page_link = 0;
2389                 sg[move_from].offset = 0;
2390
2391                 i++;
2392                 if (i == MAX_SKB_FRAGS)
2393                         i = 0;
2394         } while (1);
2395         msg->sg_end -= shift;
2396         if (msg->sg_end < 0)
2397                 msg->sg_end += MAX_SKB_FRAGS;
2398 out:
2399         msg->data = sg_virt(&sg[i]) + start - offset;
2400         msg->data_end = msg->data + bytes;
2401
2402         return 0;
2403 }
2404
2405 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2406         .func           = bpf_msg_pull_data,
2407         .gpl_only       = false,
2408         .ret_type       = RET_INTEGER,
2409         .arg1_type      = ARG_PTR_TO_CTX,
2410         .arg2_type      = ARG_ANYTHING,
2411         .arg3_type      = ARG_ANYTHING,
2412         .arg4_type      = ARG_ANYTHING,
2413 };
2414
2415 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2416 {
2417         return task_get_classid(skb);
2418 }
2419
2420 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2421         .func           = bpf_get_cgroup_classid,
2422         .gpl_only       = false,
2423         .ret_type       = RET_INTEGER,
2424         .arg1_type      = ARG_PTR_TO_CTX,
2425 };
2426
2427 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2428 {
2429         return dst_tclassid(skb);
2430 }
2431
2432 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2433         .func           = bpf_get_route_realm,
2434         .gpl_only       = false,
2435         .ret_type       = RET_INTEGER,
2436         .arg1_type      = ARG_PTR_TO_CTX,
2437 };
2438
2439 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2440 {
2441         /* If skb_clear_hash() was called due to mangling, we can
2442          * trigger SW recalculation here. Later access to hash
2443          * can then use the inline skb->hash via context directly
2444          * instead of calling this helper again.
2445          */
2446         return skb_get_hash(skb);
2447 }
2448
2449 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2450         .func           = bpf_get_hash_recalc,
2451         .gpl_only       = false,
2452         .ret_type       = RET_INTEGER,
2453         .arg1_type      = ARG_PTR_TO_CTX,
2454 };
2455
2456 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2457 {
2458         /* After all direct packet write, this can be used once for
2459          * triggering a lazy recalc on next skb_get_hash() invocation.
2460          */
2461         skb_clear_hash(skb);
2462         return 0;
2463 }
2464
2465 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2466         .func           = bpf_set_hash_invalid,
2467         .gpl_only       = false,
2468         .ret_type       = RET_INTEGER,
2469         .arg1_type      = ARG_PTR_TO_CTX,
2470 };
2471
2472 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2473 {
2474         /* Set user specified hash as L4(+), so that it gets returned
2475          * on skb_get_hash() call unless BPF prog later on triggers a
2476          * skb_clear_hash().
2477          */
2478         __skb_set_sw_hash(skb, hash, true);
2479         return 0;
2480 }
2481
2482 static const struct bpf_func_proto bpf_set_hash_proto = {
2483         .func           = bpf_set_hash,
2484         .gpl_only       = false,
2485         .ret_type       = RET_INTEGER,
2486         .arg1_type      = ARG_PTR_TO_CTX,
2487         .arg2_type      = ARG_ANYTHING,
2488 };
2489
2490 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2491            u16, vlan_tci)
2492 {
2493         int ret;
2494
2495         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2496                      vlan_proto != htons(ETH_P_8021AD)))
2497                 vlan_proto = htons(ETH_P_8021Q);
2498
2499         bpf_push_mac_rcsum(skb);
2500         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2501         bpf_pull_mac_rcsum(skb);
2502
2503         bpf_compute_data_pointers(skb);
2504         return ret;
2505 }
2506
2507 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2508         .func           = bpf_skb_vlan_push,
2509         .gpl_only       = false,
2510         .ret_type       = RET_INTEGER,
2511         .arg1_type      = ARG_PTR_TO_CTX,
2512         .arg2_type      = ARG_ANYTHING,
2513         .arg3_type      = ARG_ANYTHING,
2514 };
2515
2516 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2517 {
2518         int ret;
2519
2520         bpf_push_mac_rcsum(skb);
2521         ret = skb_vlan_pop(skb);
2522         bpf_pull_mac_rcsum(skb);
2523
2524         bpf_compute_data_pointers(skb);
2525         return ret;
2526 }
2527
2528 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2529         .func           = bpf_skb_vlan_pop,
2530         .gpl_only       = false,
2531         .ret_type       = RET_INTEGER,
2532         .arg1_type      = ARG_PTR_TO_CTX,
2533 };
2534
2535 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2536 {
2537         /* Caller already did skb_cow() with len as headroom,
2538          * so no need to do it here.
2539          */
2540         skb_push(skb, len);
2541         memmove(skb->data, skb->data + len, off);
2542         memset(skb->data + off, 0, len);
2543
2544         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2545          * needed here as it does not change the skb->csum
2546          * result for checksum complete when summing over
2547          * zeroed blocks.
2548          */
2549         return 0;
2550 }
2551
2552 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2553 {
2554         /* skb_ensure_writable() is not needed here, as we're
2555          * already working on an uncloned skb.
2556          */
2557         if (unlikely(!pskb_may_pull(skb, off + len)))
2558                 return -ENOMEM;
2559
2560         skb_postpull_rcsum(skb, skb->data + off, len);
2561         memmove(skb->data + len, skb->data, off);
2562         __skb_pull(skb, len);
2563
2564         return 0;
2565 }
2566
2567 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2568 {
2569         bool trans_same = skb->transport_header == skb->network_header;
2570         int ret;
2571
2572         /* There's no need for __skb_push()/__skb_pull() pair to
2573          * get to the start of the mac header as we're guaranteed
2574          * to always start from here under eBPF.
2575          */
2576         ret = bpf_skb_generic_push(skb, off, len);
2577         if (likely(!ret)) {
2578                 skb->mac_header -= len;
2579                 skb->network_header -= len;
2580                 if (trans_same)
2581                         skb->transport_header = skb->network_header;
2582         }
2583
2584         return ret;
2585 }
2586
2587 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2588 {
2589         bool trans_same = skb->transport_header == skb->network_header;
2590         int ret;
2591
2592         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2593         ret = bpf_skb_generic_pop(skb, off, len);
2594         if (likely(!ret)) {
2595                 skb->mac_header += len;
2596                 skb->network_header += len;
2597                 if (trans_same)
2598                         skb->transport_header = skb->network_header;
2599         }
2600
2601         return ret;
2602 }
2603
2604 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2605 {
2606         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2607         u32 off = skb_mac_header_len(skb);
2608         int ret;
2609
2610         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2611         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2612                 return -ENOTSUPP;
2613
2614         ret = skb_cow(skb, len_diff);
2615         if (unlikely(ret < 0))
2616                 return ret;
2617
2618         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2619         if (unlikely(ret < 0))
2620                 return ret;
2621
2622         if (skb_is_gso(skb)) {
2623                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2624
2625                 /* SKB_GSO_TCPV4 needs to be changed into
2626                  * SKB_GSO_TCPV6.
2627                  */
2628                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2629                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2630                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2631                 }
2632
2633                 /* Due to IPv6 header, MSS needs to be downgraded. */
2634                 skb_decrease_gso_size(shinfo, len_diff);
2635                 /* Header must be checked, and gso_segs recomputed. */
2636                 shinfo->gso_type |= SKB_GSO_DODGY;
2637                 shinfo->gso_segs = 0;
2638         }
2639
2640         skb->protocol = htons(ETH_P_IPV6);
2641         skb_clear_hash(skb);
2642
2643         return 0;
2644 }
2645
2646 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2647 {
2648         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2649         u32 off = skb_mac_header_len(skb);
2650         int ret;
2651
2652         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2653         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2654                 return -ENOTSUPP;
2655
2656         ret = skb_unclone(skb, GFP_ATOMIC);
2657         if (unlikely(ret < 0))
2658                 return ret;
2659
2660         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2661         if (unlikely(ret < 0))
2662                 return ret;
2663
2664         if (skb_is_gso(skb)) {
2665                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2666
2667                 /* SKB_GSO_TCPV6 needs to be changed into
2668                  * SKB_GSO_TCPV4.
2669                  */
2670                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2671                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2672                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2673                 }
2674
2675                 /* Due to IPv4 header, MSS can be upgraded. */
2676                 skb_increase_gso_size(shinfo, len_diff);
2677                 /* Header must be checked, and gso_segs recomputed. */
2678                 shinfo->gso_type |= SKB_GSO_DODGY;
2679                 shinfo->gso_segs = 0;
2680         }
2681
2682         skb->protocol = htons(ETH_P_IP);
2683         skb_clear_hash(skb);
2684
2685         return 0;
2686 }
2687
2688 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2689 {
2690         __be16 from_proto = skb->protocol;
2691
2692         if (from_proto == htons(ETH_P_IP) &&
2693               to_proto == htons(ETH_P_IPV6))
2694                 return bpf_skb_proto_4_to_6(skb);
2695
2696         if (from_proto == htons(ETH_P_IPV6) &&
2697               to_proto == htons(ETH_P_IP))
2698                 return bpf_skb_proto_6_to_4(skb);
2699
2700         return -ENOTSUPP;
2701 }
2702
2703 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2704            u64, flags)
2705 {
2706         int ret;
2707
2708         if (unlikely(flags))
2709                 return -EINVAL;
2710
2711         /* General idea is that this helper does the basic groundwork
2712          * needed for changing the protocol, and eBPF program fills the
2713          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2714          * and other helpers, rather than passing a raw buffer here.
2715          *
2716          * The rationale is to keep this minimal and without a need to
2717          * deal with raw packet data. F.e. even if we would pass buffers
2718          * here, the program still needs to call the bpf_lX_csum_replace()
2719          * helpers anyway. Plus, this way we keep also separation of
2720          * concerns, since f.e. bpf_skb_store_bytes() should only take
2721          * care of stores.
2722          *
2723          * Currently, additional options and extension header space are
2724          * not supported, but flags register is reserved so we can adapt
2725          * that. For offloads, we mark packet as dodgy, so that headers
2726          * need to be verified first.
2727          */
2728         ret = bpf_skb_proto_xlat(skb, proto);
2729         bpf_compute_data_pointers(skb);
2730         return ret;
2731 }
2732
2733 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2734         .func           = bpf_skb_change_proto,
2735         .gpl_only       = false,
2736         .ret_type       = RET_INTEGER,
2737         .arg1_type      = ARG_PTR_TO_CTX,
2738         .arg2_type      = ARG_ANYTHING,
2739         .arg3_type      = ARG_ANYTHING,
2740 };
2741
2742 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2743 {
2744         /* We only allow a restricted subset to be changed for now. */
2745         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2746                      !skb_pkt_type_ok(pkt_type)))
2747                 return -EINVAL;
2748
2749         skb->pkt_type = pkt_type;
2750         return 0;
2751 }
2752
2753 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2754         .func           = bpf_skb_change_type,
2755         .gpl_only       = false,
2756         .ret_type       = RET_INTEGER,
2757         .arg1_type      = ARG_PTR_TO_CTX,
2758         .arg2_type      = ARG_ANYTHING,
2759 };
2760
2761 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2762 {
2763         switch (skb->protocol) {
2764         case htons(ETH_P_IP):
2765                 return sizeof(struct iphdr);
2766         case htons(ETH_P_IPV6):
2767                 return sizeof(struct ipv6hdr);
2768         default:
2769                 return ~0U;
2770         }
2771 }
2772
2773 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2774 {
2775         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2776         int ret;
2777
2778         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2779         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2780                 return -ENOTSUPP;
2781
2782         ret = skb_cow(skb, len_diff);
2783         if (unlikely(ret < 0))
2784                 return ret;
2785
2786         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2787         if (unlikely(ret < 0))
2788                 return ret;
2789
2790         if (skb_is_gso(skb)) {
2791                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2792
2793                 /* Due to header grow, MSS needs to be downgraded. */
2794                 skb_decrease_gso_size(shinfo, len_diff);
2795                 /* Header must be checked, and gso_segs recomputed. */
2796                 shinfo->gso_type |= SKB_GSO_DODGY;
2797                 shinfo->gso_segs = 0;
2798         }
2799
2800         return 0;
2801 }
2802
2803 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2804 {
2805         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2806         int ret;
2807
2808         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2809         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2810                 return -ENOTSUPP;
2811
2812         ret = skb_unclone(skb, GFP_ATOMIC);
2813         if (unlikely(ret < 0))
2814                 return ret;
2815
2816         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2817         if (unlikely(ret < 0))
2818                 return ret;
2819
2820         if (skb_is_gso(skb)) {
2821                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2822
2823                 /* Due to header shrink, MSS can be upgraded. */
2824                 skb_increase_gso_size(shinfo, len_diff);
2825                 /* Header must be checked, and gso_segs recomputed. */
2826                 shinfo->gso_type |= SKB_GSO_DODGY;
2827                 shinfo->gso_segs = 0;
2828         }
2829
2830         return 0;
2831 }
2832
2833 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2834 {
2835         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2836                           SKB_MAX_ALLOC;
2837 }
2838
2839 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2840 {
2841         bool trans_same = skb->transport_header == skb->network_header;
2842         u32 len_cur, len_diff_abs = abs(len_diff);
2843         u32 len_min = bpf_skb_net_base_len(skb);
2844         u32 len_max = __bpf_skb_max_len(skb);
2845         __be16 proto = skb->protocol;
2846         bool shrink = len_diff < 0;
2847         int ret;
2848
2849         if (unlikely(len_diff_abs > 0xfffU))
2850                 return -EFAULT;
2851         if (unlikely(proto != htons(ETH_P_IP) &&
2852                      proto != htons(ETH_P_IPV6)))
2853                 return -ENOTSUPP;
2854
2855         len_cur = skb->len - skb_network_offset(skb);
2856         if (skb_transport_header_was_set(skb) && !trans_same)
2857                 len_cur = skb_network_header_len(skb);
2858         if ((shrink && (len_diff_abs >= len_cur ||
2859                         len_cur - len_diff_abs < len_min)) ||
2860             (!shrink && (skb->len + len_diff_abs > len_max &&
2861                          !skb_is_gso(skb))))
2862                 return -ENOTSUPP;
2863
2864         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2865                        bpf_skb_net_grow(skb, len_diff_abs);
2866
2867         bpf_compute_data_pointers(skb);
2868         return ret;
2869 }
2870
2871 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2872            u32, mode, u64, flags)
2873 {
2874         if (unlikely(flags))
2875                 return -EINVAL;
2876         if (likely(mode == BPF_ADJ_ROOM_NET))
2877                 return bpf_skb_adjust_net(skb, len_diff);
2878
2879         return -ENOTSUPP;
2880 }
2881
2882 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2883         .func           = bpf_skb_adjust_room,
2884         .gpl_only       = false,
2885         .ret_type       = RET_INTEGER,
2886         .arg1_type      = ARG_PTR_TO_CTX,
2887         .arg2_type      = ARG_ANYTHING,
2888         .arg3_type      = ARG_ANYTHING,
2889         .arg4_type      = ARG_ANYTHING,
2890 };
2891
2892 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2893 {
2894         u32 min_len = skb_network_offset(skb);
2895
2896         if (skb_transport_header_was_set(skb))
2897                 min_len = skb_transport_offset(skb);
2898         if (skb->ip_summed == CHECKSUM_PARTIAL)
2899                 min_len = skb_checksum_start_offset(skb) +
2900                           skb->csum_offset + sizeof(__sum16);
2901         return min_len;
2902 }
2903
2904 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2905 {
2906         unsigned int old_len = skb->len;
2907         int ret;
2908
2909         ret = __skb_grow_rcsum(skb, new_len);
2910         if (!ret)
2911                 memset(skb->data + old_len, 0, new_len - old_len);
2912         return ret;
2913 }
2914
2915 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2916 {
2917         return __skb_trim_rcsum(skb, new_len);
2918 }
2919
2920 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2921                                         u64 flags)
2922 {
2923         u32 max_len = __bpf_skb_max_len(skb);
2924         u32 min_len = __bpf_skb_min_len(skb);
2925         int ret;
2926
2927         if (unlikely(flags || new_len > max_len || new_len < min_len))
2928                 return -EINVAL;
2929         if (skb->encapsulation)
2930                 return -ENOTSUPP;
2931
2932         /* The basic idea of this helper is that it's performing the
2933          * needed work to either grow or trim an skb, and eBPF program
2934          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2935          * bpf_lX_csum_replace() and others rather than passing a raw
2936          * buffer here. This one is a slow path helper and intended
2937          * for replies with control messages.
2938          *
2939          * Like in bpf_skb_change_proto(), we want to keep this rather
2940          * minimal and without protocol specifics so that we are able
2941          * to separate concerns as in bpf_skb_store_bytes() should only
2942          * be the one responsible for writing buffers.
2943          *
2944          * It's really expected to be a slow path operation here for
2945          * control message replies, so we're implicitly linearizing,
2946          * uncloning and drop offloads from the skb by this.
2947          */
2948         ret = __bpf_try_make_writable(skb, skb->len);
2949         if (!ret) {
2950                 if (new_len > skb->len)
2951                         ret = bpf_skb_grow_rcsum(skb, new_len);
2952                 else if (new_len < skb->len)
2953                         ret = bpf_skb_trim_rcsum(skb, new_len);
2954                 if (!ret && skb_is_gso(skb))
2955                         skb_gso_reset(skb);
2956         }
2957         return ret;
2958 }
2959
2960 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2961            u64, flags)
2962 {
2963         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2964
2965         bpf_compute_data_pointers(skb);
2966         return ret;
2967 }
2968
2969 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2970         .func           = bpf_skb_change_tail,
2971         .gpl_only       = false,
2972         .ret_type       = RET_INTEGER,
2973         .arg1_type      = ARG_PTR_TO_CTX,
2974         .arg2_type      = ARG_ANYTHING,
2975         .arg3_type      = ARG_ANYTHING,
2976 };
2977
2978 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2979            u64, flags)
2980 {
2981         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2982
2983         bpf_compute_data_end_sk_skb(skb);
2984         return ret;
2985 }
2986
2987 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2988         .func           = sk_skb_change_tail,
2989         .gpl_only       = false,
2990         .ret_type       = RET_INTEGER,
2991         .arg1_type      = ARG_PTR_TO_CTX,
2992         .arg2_type      = ARG_ANYTHING,
2993         .arg3_type      = ARG_ANYTHING,
2994 };
2995
2996 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
2997                                         u64 flags)
2998 {
2999         u32 max_len = __bpf_skb_max_len(skb);
3000         u32 new_len = skb->len + head_room;
3001         int ret;
3002
3003         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3004                      new_len < skb->len))
3005                 return -EINVAL;
3006
3007         ret = skb_cow(skb, head_room);
3008         if (likely(!ret)) {
3009                 /* Idea for this helper is that we currently only
3010                  * allow to expand on mac header. This means that
3011                  * skb->protocol network header, etc, stay as is.
3012                  * Compared to bpf_skb_change_tail(), we're more
3013                  * flexible due to not needing to linearize or
3014                  * reset GSO. Intention for this helper is to be
3015                  * used by an L3 skb that needs to push mac header
3016                  * for redirection into L2 device.
3017                  */
3018                 __skb_push(skb, head_room);
3019                 memset(skb->data, 0, head_room);
3020                 skb_reset_mac_header(skb);
3021         }
3022
3023         return ret;
3024 }
3025
3026 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3027            u64, flags)
3028 {
3029         int ret = __bpf_skb_change_head(skb, head_room, flags);
3030
3031         bpf_compute_data_pointers(skb);
3032         return ret;
3033 }
3034
3035 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3036         .func           = bpf_skb_change_head,
3037         .gpl_only       = false,
3038         .ret_type       = RET_INTEGER,
3039         .arg1_type      = ARG_PTR_TO_CTX,
3040         .arg2_type      = ARG_ANYTHING,
3041         .arg3_type      = ARG_ANYTHING,
3042 };
3043
3044 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3045            u64, flags)
3046 {
3047         int ret = __bpf_skb_change_head(skb, head_room, flags);
3048
3049         bpf_compute_data_end_sk_skb(skb);
3050         return ret;
3051 }
3052
3053 static const struct bpf_func_proto sk_skb_change_head_proto = {
3054         .func           = sk_skb_change_head,
3055         .gpl_only       = false,
3056         .ret_type       = RET_INTEGER,
3057         .arg1_type      = ARG_PTR_TO_CTX,
3058         .arg2_type      = ARG_ANYTHING,
3059         .arg3_type      = ARG_ANYTHING,
3060 };
3061 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3062 {
3063         return xdp_data_meta_unsupported(xdp) ? 0 :
3064                xdp->data - xdp->data_meta;
3065 }
3066
3067 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3068 {
3069         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3070         unsigned long metalen = xdp_get_metalen(xdp);
3071         void *data_start = xdp_frame_end + metalen;
3072         void *data = xdp->data + offset;
3073
3074         if (unlikely(data < data_start ||
3075                      data > xdp->data_end - ETH_HLEN))
3076                 return -EINVAL;
3077
3078         if (metalen)
3079                 memmove(xdp->data_meta + offset,
3080                         xdp->data_meta, metalen);
3081         xdp->data_meta += offset;
3082         xdp->data = data;
3083
3084         return 0;
3085 }
3086
3087 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3088         .func           = bpf_xdp_adjust_head,
3089         .gpl_only       = false,
3090         .ret_type       = RET_INTEGER,
3091         .arg1_type      = ARG_PTR_TO_CTX,
3092         .arg2_type      = ARG_ANYTHING,
3093 };
3094
3095 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3096 {
3097         void *data_end = xdp->data_end + offset;
3098
3099         /* only shrinking is allowed for now. */
3100         if (unlikely(offset >= 0))
3101                 return -EINVAL;
3102
3103         if (unlikely(data_end < xdp->data + ETH_HLEN))
3104                 return -EINVAL;
3105
3106         xdp->data_end = data_end;
3107
3108         return 0;
3109 }
3110
3111 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3112         .func           = bpf_xdp_adjust_tail,
3113         .gpl_only       = false,
3114         .ret_type       = RET_INTEGER,
3115         .arg1_type      = ARG_PTR_TO_CTX,
3116         .arg2_type      = ARG_ANYTHING,
3117 };
3118
3119 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3120 {
3121         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3122         void *meta = xdp->data_meta + offset;
3123         unsigned long metalen = xdp->data - meta;
3124
3125         if (xdp_data_meta_unsupported(xdp))
3126                 return -ENOTSUPP;
3127         if (unlikely(meta < xdp_frame_end ||
3128                      meta > xdp->data))
3129                 return -EINVAL;
3130         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3131                      (metalen > 32)))
3132                 return -EACCES;
3133
3134         xdp->data_meta = meta;
3135
3136         return 0;
3137 }
3138
3139 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3140         .func           = bpf_xdp_adjust_meta,
3141         .gpl_only       = false,
3142         .ret_type       = RET_INTEGER,
3143         .arg1_type      = ARG_PTR_TO_CTX,
3144         .arg2_type      = ARG_ANYTHING,
3145 };
3146
3147 static int __bpf_tx_xdp(struct net_device *dev,
3148                         struct bpf_map *map,
3149                         struct xdp_buff *xdp,
3150                         u32 index)
3151 {
3152         struct xdp_frame *xdpf;
3153         int err, sent;
3154
3155         if (!dev->netdev_ops->ndo_xdp_xmit) {
3156                 return -EOPNOTSUPP;
3157         }
3158
3159         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3160         if (unlikely(err))
3161                 return err;
3162
3163         xdpf = convert_to_xdp_frame(xdp);
3164         if (unlikely(!xdpf))
3165                 return -EOVERFLOW;
3166
3167         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3168         if (sent <= 0)
3169                 return sent;
3170         return 0;
3171 }
3172
3173 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3174                             struct bpf_map *map,
3175                             struct xdp_buff *xdp,
3176                             u32 index)
3177 {
3178         int err;
3179
3180         switch (map->map_type) {
3181         case BPF_MAP_TYPE_DEVMAP: {
3182                 struct bpf_dtab_netdev *dst = fwd;
3183
3184                 err = dev_map_enqueue(dst, xdp, dev_rx);
3185                 if (err)
3186                         return err;
3187                 __dev_map_insert_ctx(map, index);
3188                 break;
3189         }
3190         case BPF_MAP_TYPE_CPUMAP: {
3191                 struct bpf_cpu_map_entry *rcpu = fwd;
3192
3193                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3194                 if (err)
3195                         return err;
3196                 __cpu_map_insert_ctx(map, index);
3197                 break;
3198         }
3199         case BPF_MAP_TYPE_XSKMAP: {
3200                 struct xdp_sock *xs = fwd;
3201
3202                 err = __xsk_map_redirect(map, xdp, xs);
3203                 return err;
3204         }
3205         default:
3206                 break;
3207         }
3208         return 0;
3209 }
3210
3211 void xdp_do_flush_map(void)
3212 {
3213         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3214         struct bpf_map *map = ri->map_to_flush;
3215
3216         ri->map_to_flush = NULL;
3217         if (map) {
3218                 switch (map->map_type) {
3219                 case BPF_MAP_TYPE_DEVMAP:
3220                         __dev_map_flush(map);
3221                         break;
3222                 case BPF_MAP_TYPE_CPUMAP:
3223                         __cpu_map_flush(map);
3224                         break;
3225                 case BPF_MAP_TYPE_XSKMAP:
3226                         __xsk_map_flush(map);
3227                         break;
3228                 default:
3229                         break;
3230                 }
3231         }
3232 }
3233 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3234
3235 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3236 {
3237         switch (map->map_type) {
3238         case BPF_MAP_TYPE_DEVMAP:
3239                 return __dev_map_lookup_elem(map, index);
3240         case BPF_MAP_TYPE_CPUMAP:
3241                 return __cpu_map_lookup_elem(map, index);
3242         case BPF_MAP_TYPE_XSKMAP:
3243                 return __xsk_map_lookup_elem(map, index);
3244         default:
3245                 return NULL;
3246         }
3247 }
3248
3249 static inline bool xdp_map_invalid(const struct bpf_prog *xdp_prog,
3250                                    unsigned long aux)
3251 {
3252         return (unsigned long)xdp_prog->aux != aux;
3253 }
3254
3255 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3256                                struct bpf_prog *xdp_prog)
3257 {
3258         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3259         unsigned long map_owner = ri->map_owner;
3260         struct bpf_map *map = ri->map;
3261         u32 index = ri->ifindex;
3262         void *fwd = NULL;
3263         int err;
3264
3265         ri->ifindex = 0;
3266         ri->map = NULL;
3267         ri->map_owner = 0;
3268
3269         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3270                 err = -EFAULT;
3271                 map = NULL;
3272                 goto err;
3273         }
3274
3275         fwd = __xdp_map_lookup_elem(map, index);
3276         if (!fwd) {
3277                 err = -EINVAL;
3278                 goto err;
3279         }
3280         if (ri->map_to_flush && ri->map_to_flush != map)
3281                 xdp_do_flush_map();
3282
3283         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3284         if (unlikely(err))
3285                 goto err;
3286
3287         ri->map_to_flush = map;
3288         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3289         return 0;
3290 err:
3291         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3292         return err;
3293 }
3294
3295 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3296                     struct bpf_prog *xdp_prog)
3297 {
3298         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3299         struct net_device *fwd;
3300         u32 index = ri->ifindex;
3301         int err;
3302
3303         if (ri->map)
3304                 return xdp_do_redirect_map(dev, xdp, xdp_prog);
3305
3306         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3307         ri->ifindex = 0;
3308         if (unlikely(!fwd)) {
3309                 err = -EINVAL;
3310                 goto err;
3311         }
3312
3313         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3314         if (unlikely(err))
3315                 goto err;
3316
3317         _trace_xdp_redirect(dev, xdp_prog, index);
3318         return 0;
3319 err:
3320         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3321         return err;
3322 }
3323 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3324
3325 static int xdp_do_generic_redirect_map(struct net_device *dev,
3326                                        struct sk_buff *skb,
3327                                        struct xdp_buff *xdp,
3328                                        struct bpf_prog *xdp_prog)
3329 {
3330         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3331         unsigned long map_owner = ri->map_owner;
3332         struct bpf_map *map = ri->map;
3333         u32 index = ri->ifindex;
3334         void *fwd = NULL;
3335         int err = 0;
3336
3337         ri->ifindex = 0;
3338         ri->map = NULL;
3339         ri->map_owner = 0;
3340
3341         if (unlikely(xdp_map_invalid(xdp_prog, map_owner))) {
3342                 err = -EFAULT;
3343                 map = NULL;
3344                 goto err;
3345         }
3346         fwd = __xdp_map_lookup_elem(map, index);
3347         if (unlikely(!fwd)) {
3348                 err = -EINVAL;
3349                 goto err;
3350         }
3351
3352         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3353                 struct bpf_dtab_netdev *dst = fwd;
3354
3355                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3356                 if (unlikely(err))
3357                         goto err;
3358         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3359                 struct xdp_sock *xs = fwd;
3360
3361                 err = xsk_generic_rcv(xs, xdp);
3362                 if (err)
3363                         goto err;
3364                 consume_skb(skb);
3365         } else {
3366                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3367                 err = -EBADRQC;
3368                 goto err;
3369         }
3370
3371         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3372         return 0;
3373 err:
3374         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3375         return err;
3376 }
3377
3378 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3379                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3380 {
3381         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3382         u32 index = ri->ifindex;
3383         struct net_device *fwd;
3384         int err = 0;
3385
3386         if (ri->map)
3387                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog);
3388
3389         ri->ifindex = 0;
3390         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3391         if (unlikely(!fwd)) {
3392                 err = -EINVAL;
3393                 goto err;
3394         }
3395
3396         err = xdp_ok_fwd_dev(fwd, skb->len);
3397         if (unlikely(err))
3398                 goto err;
3399
3400         skb->dev = fwd;
3401         _trace_xdp_redirect(dev, xdp_prog, index);
3402         generic_xdp_tx(skb, xdp_prog);
3403         return 0;
3404 err:
3405         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3406         return err;
3407 }
3408 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3409
3410 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3411 {
3412         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3413
3414         if (unlikely(flags))
3415                 return XDP_ABORTED;
3416
3417         ri->ifindex = ifindex;
3418         ri->flags = flags;
3419         ri->map = NULL;
3420         ri->map_owner = 0;
3421
3422         return XDP_REDIRECT;
3423 }
3424
3425 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3426         .func           = bpf_xdp_redirect,
3427         .gpl_only       = false,
3428         .ret_type       = RET_INTEGER,
3429         .arg1_type      = ARG_ANYTHING,
3430         .arg2_type      = ARG_ANYTHING,
3431 };
3432
3433 BPF_CALL_4(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, u64, flags,
3434            unsigned long, map_owner)
3435 {
3436         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3437
3438         if (unlikely(flags))
3439                 return XDP_ABORTED;
3440
3441         ri->ifindex = ifindex;
3442         ri->flags = flags;
3443         ri->map = map;
3444         ri->map_owner = map_owner;
3445
3446         return XDP_REDIRECT;
3447 }
3448
3449 /* Note, arg4 is hidden from users and populated by the verifier
3450  * with the right pointer.
3451  */
3452 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3453         .func           = bpf_xdp_redirect_map,
3454         .gpl_only       = false,
3455         .ret_type       = RET_INTEGER,
3456         .arg1_type      = ARG_CONST_MAP_PTR,
3457         .arg2_type      = ARG_ANYTHING,
3458         .arg3_type      = ARG_ANYTHING,
3459 };
3460
3461 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3462                                   unsigned long off, unsigned long len)
3463 {
3464         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3465
3466         if (unlikely(!ptr))
3467                 return len;
3468         if (ptr != dst_buff)
3469                 memcpy(dst_buff, ptr, len);
3470
3471         return 0;
3472 }
3473
3474 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3475            u64, flags, void *, meta, u64, meta_size)
3476 {
3477         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3478
3479         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3480                 return -EINVAL;
3481         if (unlikely(skb_size > skb->len))
3482                 return -EFAULT;
3483
3484         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3485                                 bpf_skb_copy);
3486 }
3487
3488 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3489         .func           = bpf_skb_event_output,
3490         .gpl_only       = true,
3491         .ret_type       = RET_INTEGER,
3492         .arg1_type      = ARG_PTR_TO_CTX,
3493         .arg2_type      = ARG_CONST_MAP_PTR,
3494         .arg3_type      = ARG_ANYTHING,
3495         .arg4_type      = ARG_PTR_TO_MEM,
3496         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3497 };
3498
3499 static unsigned short bpf_tunnel_key_af(u64 flags)
3500 {
3501         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3502 }
3503
3504 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3505            u32, size, u64, flags)
3506 {
3507         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3508         u8 compat[sizeof(struct bpf_tunnel_key)];
3509         void *to_orig = to;
3510         int err;
3511
3512         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3513                 err = -EINVAL;
3514                 goto err_clear;
3515         }
3516         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3517                 err = -EPROTO;
3518                 goto err_clear;
3519         }
3520         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3521                 err = -EINVAL;
3522                 switch (size) {
3523                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3524                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3525                         goto set_compat;
3526                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3527                         /* Fixup deprecated structure layouts here, so we have
3528                          * a common path later on.
3529                          */
3530                         if (ip_tunnel_info_af(info) != AF_INET)
3531                                 goto err_clear;
3532 set_compat:
3533                         to = (struct bpf_tunnel_key *)compat;
3534                         break;
3535                 default:
3536                         goto err_clear;
3537                 }
3538         }
3539
3540         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3541         to->tunnel_tos = info->key.tos;
3542         to->tunnel_ttl = info->key.ttl;
3543         to->tunnel_ext = 0;
3544
3545         if (flags & BPF_F_TUNINFO_IPV6) {
3546                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3547                        sizeof(to->remote_ipv6));
3548                 to->tunnel_label = be32_to_cpu(info->key.label);
3549         } else {
3550                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3551                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3552                 to->tunnel_label = 0;
3553         }
3554
3555         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3556                 memcpy(to_orig, to, size);
3557
3558         return 0;
3559 err_clear:
3560         memset(to_orig, 0, size);
3561         return err;
3562 }
3563
3564 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3565         .func           = bpf_skb_get_tunnel_key,
3566         .gpl_only       = false,
3567         .ret_type       = RET_INTEGER,
3568         .arg1_type      = ARG_PTR_TO_CTX,
3569         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3570         .arg3_type      = ARG_CONST_SIZE,
3571         .arg4_type      = ARG_ANYTHING,
3572 };
3573
3574 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3575 {
3576         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3577         int err;
3578
3579         if (unlikely(!info ||
3580                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3581                 err = -ENOENT;
3582                 goto err_clear;
3583         }
3584         if (unlikely(size < info->options_len)) {
3585                 err = -ENOMEM;
3586                 goto err_clear;
3587         }
3588
3589         ip_tunnel_info_opts_get(to, info);
3590         if (size > info->options_len)
3591                 memset(to + info->options_len, 0, size - info->options_len);
3592
3593         return info->options_len;
3594 err_clear:
3595         memset(to, 0, size);
3596         return err;
3597 }
3598
3599 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3600         .func           = bpf_skb_get_tunnel_opt,
3601         .gpl_only       = false,
3602         .ret_type       = RET_INTEGER,
3603         .arg1_type      = ARG_PTR_TO_CTX,
3604         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3605         .arg3_type      = ARG_CONST_SIZE,
3606 };
3607
3608 static struct metadata_dst __percpu *md_dst;
3609
3610 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3611            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3612 {
3613         struct metadata_dst *md = this_cpu_ptr(md_dst);
3614         u8 compat[sizeof(struct bpf_tunnel_key)];
3615         struct ip_tunnel_info *info;
3616
3617         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3618                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3619                 return -EINVAL;
3620         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3621                 switch (size) {
3622                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3623                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3624                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3625                         /* Fixup deprecated structure layouts here, so we have
3626                          * a common path later on.
3627                          */
3628                         memcpy(compat, from, size);
3629                         memset(compat + size, 0, sizeof(compat) - size);
3630                         from = (const struct bpf_tunnel_key *) compat;
3631                         break;
3632                 default:
3633                         return -EINVAL;
3634                 }
3635         }
3636         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3637                      from->tunnel_ext))
3638                 return -EINVAL;
3639
3640         skb_dst_drop(skb);
3641         dst_hold((struct dst_entry *) md);
3642         skb_dst_set(skb, (struct dst_entry *) md);
3643
3644         info = &md->u.tun_info;
3645         memset(info, 0, sizeof(*info));
3646         info->mode = IP_TUNNEL_INFO_TX;
3647
3648         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3649         if (flags & BPF_F_DONT_FRAGMENT)
3650                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3651         if (flags & BPF_F_ZERO_CSUM_TX)
3652                 info->key.tun_flags &= ~TUNNEL_CSUM;
3653         if (flags & BPF_F_SEQ_NUMBER)
3654                 info->key.tun_flags |= TUNNEL_SEQ;
3655
3656         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3657         info->key.tos = from->tunnel_tos;
3658         info->key.ttl = from->tunnel_ttl;
3659
3660         if (flags & BPF_F_TUNINFO_IPV6) {
3661                 info->mode |= IP_TUNNEL_INFO_IPV6;
3662                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3663                        sizeof(from->remote_ipv6));
3664                 info->key.label = cpu_to_be32(from->tunnel_label) &
3665                                   IPV6_FLOWLABEL_MASK;
3666         } else {
3667                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3668         }
3669
3670         return 0;
3671 }
3672
3673 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3674         .func           = bpf_skb_set_tunnel_key,
3675         .gpl_only       = false,
3676         .ret_type       = RET_INTEGER,
3677         .arg1_type      = ARG_PTR_TO_CTX,
3678         .arg2_type      = ARG_PTR_TO_MEM,
3679         .arg3_type      = ARG_CONST_SIZE,
3680         .arg4_type      = ARG_ANYTHING,
3681 };
3682
3683 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3684            const u8 *, from, u32, size)
3685 {
3686         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3687         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3688
3689         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3690                 return -EINVAL;
3691         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3692                 return -ENOMEM;
3693
3694         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3695
3696         return 0;
3697 }
3698
3699 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3700         .func           = bpf_skb_set_tunnel_opt,
3701         .gpl_only       = false,
3702         .ret_type       = RET_INTEGER,
3703         .arg1_type      = ARG_PTR_TO_CTX,
3704         .arg2_type      = ARG_PTR_TO_MEM,
3705         .arg3_type      = ARG_CONST_SIZE,
3706 };
3707
3708 static const struct bpf_func_proto *
3709 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3710 {
3711         if (!md_dst) {
3712                 struct metadata_dst __percpu *tmp;
3713
3714                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3715                                                 METADATA_IP_TUNNEL,
3716                                                 GFP_KERNEL);
3717                 if (!tmp)
3718                         return NULL;
3719                 if (cmpxchg(&md_dst, NULL, tmp))
3720                         metadata_dst_free_percpu(tmp);
3721         }
3722
3723         switch (which) {
3724         case BPF_FUNC_skb_set_tunnel_key:
3725                 return &bpf_skb_set_tunnel_key_proto;
3726         case BPF_FUNC_skb_set_tunnel_opt:
3727                 return &bpf_skb_set_tunnel_opt_proto;
3728         default:
3729                 return NULL;
3730         }
3731 }
3732
3733 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3734            u32, idx)
3735 {
3736         struct bpf_array *array = container_of(map, struct bpf_array, map);
3737         struct cgroup *cgrp;
3738         struct sock *sk;
3739
3740         sk = skb_to_full_sk(skb);
3741         if (!sk || !sk_fullsock(sk))
3742                 return -ENOENT;
3743         if (unlikely(idx >= array->map.max_entries))
3744                 return -E2BIG;
3745
3746         cgrp = READ_ONCE(array->ptrs[idx]);
3747         if (unlikely(!cgrp))
3748                 return -EAGAIN;
3749
3750         return sk_under_cgroup_hierarchy(sk, cgrp);
3751 }
3752
3753 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3754         .func           = bpf_skb_under_cgroup,
3755         .gpl_only       = false,
3756         .ret_type       = RET_INTEGER,
3757         .arg1_type      = ARG_PTR_TO_CTX,
3758         .arg2_type      = ARG_CONST_MAP_PTR,
3759         .arg3_type      = ARG_ANYTHING,
3760 };
3761
3762 #ifdef CONFIG_SOCK_CGROUP_DATA
3763 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3764 {
3765         struct sock *sk = skb_to_full_sk(skb);
3766         struct cgroup *cgrp;
3767
3768         if (!sk || !sk_fullsock(sk))
3769                 return 0;
3770
3771         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3772         return cgrp->kn->id.id;
3773 }
3774
3775 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3776         .func           = bpf_skb_cgroup_id,
3777         .gpl_only       = false,
3778         .ret_type       = RET_INTEGER,
3779         .arg1_type      = ARG_PTR_TO_CTX,
3780 };
3781
3782 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3783            ancestor_level)
3784 {
3785         struct sock *sk = skb_to_full_sk(skb);
3786         struct cgroup *ancestor;
3787         struct cgroup *cgrp;
3788
3789         if (!sk || !sk_fullsock(sk))
3790                 return 0;
3791
3792         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3793         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3794         if (!ancestor)
3795                 return 0;
3796
3797         return ancestor->kn->id.id;
3798 }
3799
3800 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3801         .func           = bpf_skb_ancestor_cgroup_id,
3802         .gpl_only       = false,
3803         .ret_type       = RET_INTEGER,
3804         .arg1_type      = ARG_PTR_TO_CTX,
3805         .arg2_type      = ARG_ANYTHING,
3806 };
3807 #endif
3808
3809 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3810                                   unsigned long off, unsigned long len)
3811 {
3812         memcpy(dst_buff, src_buff + off, len);
3813         return 0;
3814 }
3815
3816 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3817            u64, flags, void *, meta, u64, meta_size)
3818 {
3819         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3820
3821         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3822                 return -EINVAL;
3823         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3824                 return -EFAULT;
3825
3826         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3827                                 xdp_size, bpf_xdp_copy);
3828 }
3829
3830 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3831         .func           = bpf_xdp_event_output,
3832         .gpl_only       = true,
3833         .ret_type       = RET_INTEGER,
3834         .arg1_type      = ARG_PTR_TO_CTX,
3835         .arg2_type      = ARG_CONST_MAP_PTR,
3836         .arg3_type      = ARG_ANYTHING,
3837         .arg4_type      = ARG_PTR_TO_MEM,
3838         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3839 };
3840
3841 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3842 {
3843         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3844 }
3845
3846 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3847         .func           = bpf_get_socket_cookie,
3848         .gpl_only       = false,
3849         .ret_type       = RET_INTEGER,
3850         .arg1_type      = ARG_PTR_TO_CTX,
3851 };
3852
3853 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
3854 {
3855         return sock_gen_cookie(ctx->sk);
3856 }
3857
3858 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
3859         .func           = bpf_get_socket_cookie_sock_addr,
3860         .gpl_only       = false,
3861         .ret_type       = RET_INTEGER,
3862         .arg1_type      = ARG_PTR_TO_CTX,
3863 };
3864
3865 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
3866 {
3867         return sock_gen_cookie(ctx->sk);
3868 }
3869
3870 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
3871         .func           = bpf_get_socket_cookie_sock_ops,
3872         .gpl_only       = false,
3873         .ret_type       = RET_INTEGER,
3874         .arg1_type      = ARG_PTR_TO_CTX,
3875 };
3876
3877 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3878 {
3879         struct sock *sk = sk_to_full_sk(skb->sk);
3880         kuid_t kuid;
3881
3882         if (!sk || !sk_fullsock(sk))
3883                 return overflowuid;
3884         kuid = sock_net_uid(sock_net(sk), sk);
3885         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3886 }
3887
3888 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3889         .func           = bpf_get_socket_uid,
3890         .gpl_only       = false,
3891         .ret_type       = RET_INTEGER,
3892         .arg1_type      = ARG_PTR_TO_CTX,
3893 };
3894
3895 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3896            int, level, int, optname, char *, optval, int, optlen)
3897 {
3898         struct sock *sk = bpf_sock->sk;
3899         int ret = 0;
3900         int val;
3901
3902         if (!sk_fullsock(sk))
3903                 return -EINVAL;
3904
3905         if (level == SOL_SOCKET) {
3906                 if (optlen != sizeof(int))
3907                         return -EINVAL;
3908                 val = *((int *)optval);
3909
3910                 /* Only some socketops are supported */
3911                 switch (optname) {
3912                 case SO_RCVBUF:
3913                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3914                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3915                         break;
3916                 case SO_SNDBUF:
3917                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3918                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3919                         break;
3920                 case SO_MAX_PACING_RATE:
3921                         sk->sk_max_pacing_rate = val;
3922                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3923                                                  sk->sk_max_pacing_rate);
3924                         break;
3925                 case SO_PRIORITY:
3926                         sk->sk_priority = val;
3927                         break;
3928                 case SO_RCVLOWAT:
3929                         if (val < 0)
3930                                 val = INT_MAX;
3931                         sk->sk_rcvlowat = val ? : 1;
3932                         break;
3933                 case SO_MARK:
3934                         sk->sk_mark = val;
3935                         break;
3936                 default:
3937                         ret = -EINVAL;
3938                 }
3939 #ifdef CONFIG_INET
3940         } else if (level == SOL_IP) {
3941                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3942                         return -EINVAL;
3943
3944                 val = *((int *)optval);
3945                 /* Only some options are supported */
3946                 switch (optname) {
3947                 case IP_TOS:
3948                         if (val < -1 || val > 0xff) {
3949                                 ret = -EINVAL;
3950                         } else {
3951                                 struct inet_sock *inet = inet_sk(sk);
3952
3953                                 if (val == -1)
3954                                         val = 0;
3955                                 inet->tos = val;
3956                         }
3957                         break;
3958                 default:
3959                         ret = -EINVAL;
3960                 }
3961 #if IS_ENABLED(CONFIG_IPV6)
3962         } else if (level == SOL_IPV6) {
3963                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3964                         return -EINVAL;
3965
3966                 val = *((int *)optval);
3967                 /* Only some options are supported */
3968                 switch (optname) {
3969                 case IPV6_TCLASS:
3970                         if (val < -1 || val > 0xff) {
3971                                 ret = -EINVAL;
3972                         } else {
3973                                 struct ipv6_pinfo *np = inet6_sk(sk);
3974
3975                                 if (val == -1)
3976                                         val = 0;
3977                                 np->tclass = val;
3978                         }
3979                         break;
3980                 default:
3981                         ret = -EINVAL;
3982                 }
3983 #endif
3984         } else if (level == SOL_TCP &&
3985                    sk->sk_prot->setsockopt == tcp_setsockopt) {
3986                 if (optname == TCP_CONGESTION) {
3987                         char name[TCP_CA_NAME_MAX];
3988                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3989
3990                         strncpy(name, optval, min_t(long, optlen,
3991                                                     TCP_CA_NAME_MAX-1));
3992                         name[TCP_CA_NAME_MAX-1] = 0;
3993                         ret = tcp_set_congestion_control(sk, name, false,
3994                                                          reinit);
3995                 } else {
3996                         struct tcp_sock *tp = tcp_sk(sk);
3997
3998                         if (optlen != sizeof(int))
3999                                 return -EINVAL;
4000
4001                         val = *((int *)optval);
4002                         /* Only some options are supported */
4003                         switch (optname) {
4004                         case TCP_BPF_IW:
4005                                 if (val <= 0 || tp->data_segs_out > 0)
4006                                         ret = -EINVAL;
4007                                 else
4008                                         tp->snd_cwnd = val;
4009                                 break;
4010                         case TCP_BPF_SNDCWND_CLAMP:
4011                                 if (val <= 0) {
4012                                         ret = -EINVAL;
4013                                 } else {
4014                                         tp->snd_cwnd_clamp = val;
4015                                         tp->snd_ssthresh = val;
4016                                 }
4017                                 break;
4018                         default:
4019                                 ret = -EINVAL;
4020                         }
4021                 }
4022 #endif
4023         } else {
4024                 ret = -EINVAL;
4025         }
4026         return ret;
4027 }
4028
4029 static const struct bpf_func_proto bpf_setsockopt_proto = {
4030         .func           = bpf_setsockopt,
4031         .gpl_only       = false,
4032         .ret_type       = RET_INTEGER,
4033         .arg1_type      = ARG_PTR_TO_CTX,
4034         .arg2_type      = ARG_ANYTHING,
4035         .arg3_type      = ARG_ANYTHING,
4036         .arg4_type      = ARG_PTR_TO_MEM,
4037         .arg5_type      = ARG_CONST_SIZE,
4038 };
4039
4040 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4041            int, level, int, optname, char *, optval, int, optlen)
4042 {
4043         struct sock *sk = bpf_sock->sk;
4044
4045         if (!sk_fullsock(sk))
4046                 goto err_clear;
4047
4048 #ifdef CONFIG_INET
4049         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4050                 if (optname == TCP_CONGESTION) {
4051                         struct inet_connection_sock *icsk = inet_csk(sk);
4052
4053                         if (!icsk->icsk_ca_ops || optlen <= 1)
4054                                 goto err_clear;
4055                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4056                         optval[optlen - 1] = 0;
4057                 } else {
4058                         goto err_clear;
4059                 }
4060         } else if (level == SOL_IP) {
4061                 struct inet_sock *inet = inet_sk(sk);
4062
4063                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4064                         goto err_clear;
4065
4066                 /* Only some options are supported */
4067                 switch (optname) {
4068                 case IP_TOS:
4069                         *((int *)optval) = (int)inet->tos;
4070                         break;
4071                 default:
4072                         goto err_clear;
4073                 }
4074 #if IS_ENABLED(CONFIG_IPV6)
4075         } else if (level == SOL_IPV6) {
4076                 struct ipv6_pinfo *np = inet6_sk(sk);
4077
4078                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4079                         goto err_clear;
4080
4081                 /* Only some options are supported */
4082                 switch (optname) {
4083                 case IPV6_TCLASS:
4084                         *((int *)optval) = (int)np->tclass;
4085                         break;
4086                 default:
4087                         goto err_clear;
4088                 }
4089 #endif
4090         } else {
4091                 goto err_clear;
4092         }
4093         return 0;
4094 #endif
4095 err_clear:
4096         memset(optval, 0, optlen);
4097         return -EINVAL;
4098 }
4099
4100 static const struct bpf_func_proto bpf_getsockopt_proto = {
4101         .func           = bpf_getsockopt,
4102         .gpl_only       = false,
4103         .ret_type       = RET_INTEGER,
4104         .arg1_type      = ARG_PTR_TO_CTX,
4105         .arg2_type      = ARG_ANYTHING,
4106         .arg3_type      = ARG_ANYTHING,
4107         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4108         .arg5_type      = ARG_CONST_SIZE,
4109 };
4110
4111 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4112            int, argval)
4113 {
4114         struct sock *sk = bpf_sock->sk;
4115         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4116
4117         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4118                 return -EINVAL;
4119
4120         if (val)
4121                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4122
4123         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4124 }
4125
4126 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4127         .func           = bpf_sock_ops_cb_flags_set,
4128         .gpl_only       = false,
4129         .ret_type       = RET_INTEGER,
4130         .arg1_type      = ARG_PTR_TO_CTX,
4131         .arg2_type      = ARG_ANYTHING,
4132 };
4133
4134 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4135 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4136
4137 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4138            int, addr_len)
4139 {
4140 #ifdef CONFIG_INET
4141         struct sock *sk = ctx->sk;
4142         int err;
4143
4144         /* Binding to port can be expensive so it's prohibited in the helper.
4145          * Only binding to IP is supported.
4146          */
4147         err = -EINVAL;
4148         if (addr->sa_family == AF_INET) {
4149                 if (addr_len < sizeof(struct sockaddr_in))
4150                         return err;
4151                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4152                         return err;
4153                 return __inet_bind(sk, addr, addr_len, true, false);
4154 #if IS_ENABLED(CONFIG_IPV6)
4155         } else if (addr->sa_family == AF_INET6) {
4156                 if (addr_len < SIN6_LEN_RFC2133)
4157                         return err;
4158                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4159                         return err;
4160                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4161                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4162                  */
4163                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4164 #endif /* CONFIG_IPV6 */
4165         }
4166 #endif /* CONFIG_INET */
4167
4168         return -EAFNOSUPPORT;
4169 }
4170
4171 static const struct bpf_func_proto bpf_bind_proto = {
4172         .func           = bpf_bind,
4173         .gpl_only       = false,
4174         .ret_type       = RET_INTEGER,
4175         .arg1_type      = ARG_PTR_TO_CTX,
4176         .arg2_type      = ARG_PTR_TO_MEM,
4177         .arg3_type      = ARG_CONST_SIZE,
4178 };
4179
4180 #ifdef CONFIG_XFRM
4181 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4182            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4183 {
4184         const struct sec_path *sp = skb_sec_path(skb);
4185         const struct xfrm_state *x;
4186
4187         if (!sp || unlikely(index >= sp->len || flags))
4188                 goto err_clear;
4189
4190         x = sp->xvec[index];
4191
4192         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4193                 goto err_clear;
4194
4195         to->reqid = x->props.reqid;
4196         to->spi = x->id.spi;
4197         to->family = x->props.family;
4198         to->ext = 0;
4199
4200         if (to->family == AF_INET6) {
4201                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4202                        sizeof(to->remote_ipv6));
4203         } else {
4204                 to->remote_ipv4 = x->props.saddr.a4;
4205                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4206         }
4207
4208         return 0;
4209 err_clear:
4210         memset(to, 0, size);
4211         return -EINVAL;
4212 }
4213
4214 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4215         .func           = bpf_skb_get_xfrm_state,
4216         .gpl_only       = false,
4217         .ret_type       = RET_INTEGER,
4218         .arg1_type      = ARG_PTR_TO_CTX,
4219         .arg2_type      = ARG_ANYTHING,
4220         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4221         .arg4_type      = ARG_CONST_SIZE,
4222         .arg5_type      = ARG_ANYTHING,
4223 };
4224 #endif
4225
4226 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4227 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4228                                   const struct neighbour *neigh,
4229                                   const struct net_device *dev)
4230 {
4231         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4232         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4233         params->h_vlan_TCI = 0;
4234         params->h_vlan_proto = 0;
4235         params->ifindex = dev->ifindex;
4236
4237         return 0;
4238 }
4239 #endif
4240
4241 #if IS_ENABLED(CONFIG_INET)
4242 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4243                                u32 flags, bool check_mtu)
4244 {
4245         struct in_device *in_dev;
4246         struct neighbour *neigh;
4247         struct net_device *dev;
4248         struct fib_result res;
4249         struct fib_nh *nh;
4250         struct flowi4 fl4;
4251         int err;
4252         u32 mtu;
4253
4254         dev = dev_get_by_index_rcu(net, params->ifindex);
4255         if (unlikely(!dev))
4256                 return -ENODEV;
4257
4258         /* verify forwarding is enabled on this interface */
4259         in_dev = __in_dev_get_rcu(dev);
4260         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4261                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4262
4263         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4264                 fl4.flowi4_iif = 1;
4265                 fl4.flowi4_oif = params->ifindex;
4266         } else {
4267                 fl4.flowi4_iif = params->ifindex;
4268                 fl4.flowi4_oif = 0;
4269         }
4270         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4271         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4272         fl4.flowi4_flags = 0;
4273
4274         fl4.flowi4_proto = params->l4_protocol;
4275         fl4.daddr = params->ipv4_dst;
4276         fl4.saddr = params->ipv4_src;
4277         fl4.fl4_sport = params->sport;
4278         fl4.fl4_dport = params->dport;
4279
4280         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4281                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4282                 struct fib_table *tb;
4283
4284                 tb = fib_get_table(net, tbid);
4285                 if (unlikely(!tb))
4286                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4287
4288                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4289         } else {
4290                 fl4.flowi4_mark = 0;
4291                 fl4.flowi4_secid = 0;
4292                 fl4.flowi4_tun_key.tun_id = 0;
4293                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4294
4295                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4296         }
4297
4298         if (err) {
4299                 /* map fib lookup errors to RTN_ type */
4300                 if (err == -EINVAL)
4301                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4302                 if (err == -EHOSTUNREACH)
4303                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4304                 if (err == -EACCES)
4305                         return BPF_FIB_LKUP_RET_PROHIBIT;
4306
4307                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4308         }
4309
4310         if (res.type != RTN_UNICAST)
4311                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4312
4313         if (res.fi->fib_nhs > 1)
4314                 fib_select_path(net, &res, &fl4, NULL);
4315
4316         if (check_mtu) {
4317                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4318                 if (params->tot_len > mtu)
4319                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4320         }
4321
4322         nh = &res.fi->fib_nh[res.nh_sel];
4323
4324         /* do not handle lwt encaps right now */
4325         if (nh->nh_lwtstate)
4326                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4327
4328         dev = nh->nh_dev;
4329         if (nh->nh_gw)
4330                 params->ipv4_dst = nh->nh_gw;
4331
4332         params->rt_metric = res.fi->fib_priority;
4333
4334         /* xdp and cls_bpf programs are run in RCU-bh so
4335          * rcu_read_lock_bh is not needed here
4336          */
4337         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4338         if (!neigh)
4339                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4340
4341         return bpf_fib_set_fwd_params(params, neigh, dev);
4342 }
4343 #endif
4344
4345 #if IS_ENABLED(CONFIG_IPV6)
4346 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4347                                u32 flags, bool check_mtu)
4348 {
4349         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4350         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4351         struct neighbour *neigh;
4352         struct net_device *dev;
4353         struct inet6_dev *idev;
4354         struct fib6_info *f6i;
4355         struct flowi6 fl6;
4356         int strict = 0;
4357         int oif;
4358         u32 mtu;
4359
4360         /* link local addresses are never forwarded */
4361         if (rt6_need_strict(dst) || rt6_need_strict(src))
4362                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4363
4364         dev = dev_get_by_index_rcu(net, params->ifindex);
4365         if (unlikely(!dev))
4366                 return -ENODEV;
4367
4368         idev = __in6_dev_get_safely(dev);
4369         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4370                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4371
4372         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4373                 fl6.flowi6_iif = 1;
4374                 oif = fl6.flowi6_oif = params->ifindex;
4375         } else {
4376                 oif = fl6.flowi6_iif = params->ifindex;
4377                 fl6.flowi6_oif = 0;
4378                 strict = RT6_LOOKUP_F_HAS_SADDR;
4379         }
4380         fl6.flowlabel = params->flowinfo;
4381         fl6.flowi6_scope = 0;
4382         fl6.flowi6_flags = 0;
4383         fl6.mp_hash = 0;
4384
4385         fl6.flowi6_proto = params->l4_protocol;
4386         fl6.daddr = *dst;
4387         fl6.saddr = *src;
4388         fl6.fl6_sport = params->sport;
4389         fl6.fl6_dport = params->dport;
4390
4391         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4392                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4393                 struct fib6_table *tb;
4394
4395                 tb = ipv6_stub->fib6_get_table(net, tbid);
4396                 if (unlikely(!tb))
4397                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4398
4399                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4400         } else {
4401                 fl6.flowi6_mark = 0;
4402                 fl6.flowi6_secid = 0;
4403                 fl6.flowi6_tun_key.tun_id = 0;
4404                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4405
4406                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4407         }
4408
4409         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4410                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4411
4412         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4413                 switch (f6i->fib6_type) {
4414                 case RTN_BLACKHOLE:
4415                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4416                 case RTN_UNREACHABLE:
4417                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4418                 case RTN_PROHIBIT:
4419                         return BPF_FIB_LKUP_RET_PROHIBIT;
4420                 default:
4421                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4422                 }
4423         }
4424
4425         if (f6i->fib6_type != RTN_UNICAST)
4426                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4427
4428         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4429                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4430                                                        fl6.flowi6_oif, NULL,
4431                                                        strict);
4432
4433         if (check_mtu) {
4434                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4435                 if (params->tot_len > mtu)
4436                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4437         }
4438
4439         if (f6i->fib6_nh.nh_lwtstate)
4440                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4441
4442         if (f6i->fib6_flags & RTF_GATEWAY)
4443                 *dst = f6i->fib6_nh.nh_gw;
4444
4445         dev = f6i->fib6_nh.nh_dev;
4446         params->rt_metric = f6i->fib6_metric;
4447
4448         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4449          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4450          * because we need to get nd_tbl via the stub
4451          */
4452         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4453                                       ndisc_hashfn, dst, dev);
4454         if (!neigh)
4455                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4456
4457         return bpf_fib_set_fwd_params(params, neigh, dev);
4458 }
4459 #endif
4460
4461 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4462            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4463 {
4464         if (plen < sizeof(*params))
4465                 return -EINVAL;
4466
4467         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4468                 return -EINVAL;
4469
4470         switch (params->family) {
4471 #if IS_ENABLED(CONFIG_INET)
4472         case AF_INET:
4473                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4474                                            flags, true);
4475 #endif
4476 #if IS_ENABLED(CONFIG_IPV6)
4477         case AF_INET6:
4478                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4479                                            flags, true);
4480 #endif
4481         }
4482         return -EAFNOSUPPORT;
4483 }
4484
4485 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4486         .func           = bpf_xdp_fib_lookup,
4487         .gpl_only       = true,
4488         .ret_type       = RET_INTEGER,
4489         .arg1_type      = ARG_PTR_TO_CTX,
4490         .arg2_type      = ARG_PTR_TO_MEM,
4491         .arg3_type      = ARG_CONST_SIZE,
4492         .arg4_type      = ARG_ANYTHING,
4493 };
4494
4495 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4496            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4497 {
4498         struct net *net = dev_net(skb->dev);
4499         int rc = -EAFNOSUPPORT;
4500
4501         if (plen < sizeof(*params))
4502                 return -EINVAL;
4503
4504         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4505                 return -EINVAL;
4506
4507         switch (params->family) {
4508 #if IS_ENABLED(CONFIG_INET)
4509         case AF_INET:
4510                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4511                 break;
4512 #endif
4513 #if IS_ENABLED(CONFIG_IPV6)
4514         case AF_INET6:
4515                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4516                 break;
4517 #endif
4518         }
4519
4520         if (!rc) {
4521                 struct net_device *dev;
4522
4523                 dev = dev_get_by_index_rcu(net, params->ifindex);
4524                 if (!is_skb_forwardable(dev, skb))
4525                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4526         }
4527
4528         return rc;
4529 }
4530
4531 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4532         .func           = bpf_skb_fib_lookup,
4533         .gpl_only       = true,
4534         .ret_type       = RET_INTEGER,
4535         .arg1_type      = ARG_PTR_TO_CTX,
4536         .arg2_type      = ARG_PTR_TO_MEM,
4537         .arg3_type      = ARG_CONST_SIZE,
4538         .arg4_type      = ARG_ANYTHING,
4539 };
4540
4541 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4542 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4543 {
4544         int err;
4545         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4546
4547         if (!seg6_validate_srh(srh, len))
4548                 return -EINVAL;
4549
4550         switch (type) {
4551         case BPF_LWT_ENCAP_SEG6_INLINE:
4552                 if (skb->protocol != htons(ETH_P_IPV6))
4553                         return -EBADMSG;
4554
4555                 err = seg6_do_srh_inline(skb, srh);
4556                 break;
4557         case BPF_LWT_ENCAP_SEG6:
4558                 skb_reset_inner_headers(skb);
4559                 skb->encapsulation = 1;
4560                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4561                 break;
4562         default:
4563                 return -EINVAL;
4564         }
4565
4566         bpf_compute_data_pointers(skb);
4567         if (err)
4568                 return err;
4569
4570         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4571         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4572
4573         return seg6_lookup_nexthop(skb, NULL, 0);
4574 }
4575 #endif /* CONFIG_IPV6_SEG6_BPF */
4576
4577 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4578            u32, len)
4579 {
4580         switch (type) {
4581 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4582         case BPF_LWT_ENCAP_SEG6:
4583         case BPF_LWT_ENCAP_SEG6_INLINE:
4584                 return bpf_push_seg6_encap(skb, type, hdr, len);
4585 #endif
4586         default:
4587                 return -EINVAL;
4588         }
4589 }
4590
4591 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4592         .func           = bpf_lwt_push_encap,
4593         .gpl_only       = false,
4594         .ret_type       = RET_INTEGER,
4595         .arg1_type      = ARG_PTR_TO_CTX,
4596         .arg2_type      = ARG_ANYTHING,
4597         .arg3_type      = ARG_PTR_TO_MEM,
4598         .arg4_type      = ARG_CONST_SIZE
4599 };
4600
4601 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4602 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4603            const void *, from, u32, len)
4604 {
4605         struct seg6_bpf_srh_state *srh_state =
4606                 this_cpu_ptr(&seg6_bpf_srh_states);
4607         struct ipv6_sr_hdr *srh = srh_state->srh;
4608         void *srh_tlvs, *srh_end, *ptr;
4609         int srhoff = 0;
4610
4611         if (srh == NULL)
4612                 return -EINVAL;
4613
4614         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4615         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4616
4617         ptr = skb->data + offset;
4618         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4619                 srh_state->valid = false;
4620         else if (ptr < (void *)&srh->flags ||
4621                  ptr + len > (void *)&srh->segments)
4622                 return -EFAULT;
4623
4624         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4625                 return -EFAULT;
4626         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4627                 return -EINVAL;
4628         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4629
4630         memcpy(skb->data + offset, from, len);
4631         return 0;
4632 }
4633
4634 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4635         .func           = bpf_lwt_seg6_store_bytes,
4636         .gpl_only       = false,
4637         .ret_type       = RET_INTEGER,
4638         .arg1_type      = ARG_PTR_TO_CTX,
4639         .arg2_type      = ARG_ANYTHING,
4640         .arg3_type      = ARG_PTR_TO_MEM,
4641         .arg4_type      = ARG_CONST_SIZE
4642 };
4643
4644 static void bpf_update_srh_state(struct sk_buff *skb)
4645 {
4646         struct seg6_bpf_srh_state *srh_state =
4647                 this_cpu_ptr(&seg6_bpf_srh_states);
4648         int srhoff = 0;
4649
4650         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4651                 srh_state->srh = NULL;
4652         } else {
4653                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4654                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4655                 srh_state->valid = true;
4656         }
4657 }
4658
4659 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4660            u32, action, void *, param, u32, param_len)
4661 {
4662         struct seg6_bpf_srh_state *srh_state =
4663                 this_cpu_ptr(&seg6_bpf_srh_states);
4664         int hdroff = 0;
4665         int err;
4666
4667         switch (action) {
4668         case SEG6_LOCAL_ACTION_END_X:
4669                 if (!seg6_bpf_has_valid_srh(skb))
4670                         return -EBADMSG;
4671                 if (param_len != sizeof(struct in6_addr))
4672                         return -EINVAL;
4673                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4674         case SEG6_LOCAL_ACTION_END_T:
4675                 if (!seg6_bpf_has_valid_srh(skb))
4676                         return -EBADMSG;
4677                 if (param_len != sizeof(int))
4678                         return -EINVAL;
4679                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4680         case SEG6_LOCAL_ACTION_END_DT6:
4681                 if (!seg6_bpf_has_valid_srh(skb))
4682                         return -EBADMSG;
4683                 if (param_len != sizeof(int))
4684                         return -EINVAL;
4685
4686                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4687                         return -EBADMSG;
4688                 if (!pskb_pull(skb, hdroff))
4689                         return -EBADMSG;
4690
4691                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4692                 skb_reset_network_header(skb);
4693                 skb_reset_transport_header(skb);
4694                 skb->encapsulation = 0;
4695
4696                 bpf_compute_data_pointers(skb);
4697                 bpf_update_srh_state(skb);
4698                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4699         case SEG6_LOCAL_ACTION_END_B6:
4700                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4701                         return -EBADMSG;
4702                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4703                                           param, param_len);
4704                 if (!err)
4705                         bpf_update_srh_state(skb);
4706
4707                 return err;
4708         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4709                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4710                         return -EBADMSG;
4711                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4712                                           param, param_len);
4713                 if (!err)
4714                         bpf_update_srh_state(skb);
4715
4716                 return err;
4717         default:
4718                 return -EINVAL;
4719         }
4720 }
4721
4722 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4723         .func           = bpf_lwt_seg6_action,
4724         .gpl_only       = false,
4725         .ret_type       = RET_INTEGER,
4726         .arg1_type      = ARG_PTR_TO_CTX,
4727         .arg2_type      = ARG_ANYTHING,
4728         .arg3_type      = ARG_PTR_TO_MEM,
4729         .arg4_type      = ARG_CONST_SIZE
4730 };
4731
4732 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4733            s32, len)
4734 {
4735         struct seg6_bpf_srh_state *srh_state =
4736                 this_cpu_ptr(&seg6_bpf_srh_states);
4737         struct ipv6_sr_hdr *srh = srh_state->srh;
4738         void *srh_end, *srh_tlvs, *ptr;
4739         struct ipv6hdr *hdr;
4740         int srhoff = 0;
4741         int ret;
4742
4743         if (unlikely(srh == NULL))
4744                 return -EINVAL;
4745
4746         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4747                         ((srh->first_segment + 1) << 4));
4748         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4749                         srh_state->hdrlen);
4750         ptr = skb->data + offset;
4751
4752         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4753                 return -EFAULT;
4754         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4755                 return -EFAULT;
4756
4757         if (len > 0) {
4758                 ret = skb_cow_head(skb, len);
4759                 if (unlikely(ret < 0))
4760                         return ret;
4761
4762                 ret = bpf_skb_net_hdr_push(skb, offset, len);
4763         } else {
4764                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4765         }
4766
4767         bpf_compute_data_pointers(skb);
4768         if (unlikely(ret < 0))
4769                 return ret;
4770
4771         hdr = (struct ipv6hdr *)skb->data;
4772         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4773
4774         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4775                 return -EINVAL;
4776         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4777         srh_state->hdrlen += len;
4778         srh_state->valid = false;
4779         return 0;
4780 }
4781
4782 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4783         .func           = bpf_lwt_seg6_adjust_srh,
4784         .gpl_only       = false,
4785         .ret_type       = RET_INTEGER,
4786         .arg1_type      = ARG_PTR_TO_CTX,
4787         .arg2_type      = ARG_ANYTHING,
4788         .arg3_type      = ARG_ANYTHING,
4789 };
4790 #endif /* CONFIG_IPV6_SEG6_BPF */
4791
4792 bool bpf_helper_changes_pkt_data(void *func)
4793 {
4794         if (func == bpf_skb_vlan_push ||
4795             func == bpf_skb_vlan_pop ||
4796             func == bpf_skb_store_bytes ||
4797             func == bpf_skb_change_proto ||
4798             func == bpf_skb_change_head ||
4799             func == sk_skb_change_head ||
4800             func == bpf_skb_change_tail ||
4801             func == sk_skb_change_tail ||
4802             func == bpf_skb_adjust_room ||
4803             func == bpf_skb_pull_data ||
4804             func == sk_skb_pull_data ||
4805             func == bpf_clone_redirect ||
4806             func == bpf_l3_csum_replace ||
4807             func == bpf_l4_csum_replace ||
4808             func == bpf_xdp_adjust_head ||
4809             func == bpf_xdp_adjust_meta ||
4810             func == bpf_msg_pull_data ||
4811             func == bpf_xdp_adjust_tail ||
4812 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4813             func == bpf_lwt_seg6_store_bytes ||
4814             func == bpf_lwt_seg6_adjust_srh ||
4815             func == bpf_lwt_seg6_action ||
4816 #endif
4817             func == bpf_lwt_push_encap)
4818                 return true;
4819
4820         return false;
4821 }
4822
4823 static const struct bpf_func_proto *
4824 bpf_base_func_proto(enum bpf_func_id func_id)
4825 {
4826         switch (func_id) {
4827         case BPF_FUNC_map_lookup_elem:
4828                 return &bpf_map_lookup_elem_proto;
4829         case BPF_FUNC_map_update_elem:
4830                 return &bpf_map_update_elem_proto;
4831         case BPF_FUNC_map_delete_elem:
4832                 return &bpf_map_delete_elem_proto;
4833         case BPF_FUNC_get_prandom_u32:
4834                 return &bpf_get_prandom_u32_proto;
4835         case BPF_FUNC_get_smp_processor_id:
4836                 return &bpf_get_raw_smp_processor_id_proto;
4837         case BPF_FUNC_get_numa_node_id:
4838                 return &bpf_get_numa_node_id_proto;
4839         case BPF_FUNC_tail_call:
4840                 return &bpf_tail_call_proto;
4841         case BPF_FUNC_ktime_get_ns:
4842                 return &bpf_ktime_get_ns_proto;
4843         case BPF_FUNC_trace_printk:
4844                 if (capable(CAP_SYS_ADMIN))
4845                         return bpf_get_trace_printk_proto();
4846                 /* else: fall through */
4847         default:
4848                 return NULL;
4849         }
4850 }
4851
4852 static const struct bpf_func_proto *
4853 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4854 {
4855         switch (func_id) {
4856         /* inet and inet6 sockets are created in a process
4857          * context so there is always a valid uid/gid
4858          */
4859         case BPF_FUNC_get_current_uid_gid:
4860                 return &bpf_get_current_uid_gid_proto;
4861         case BPF_FUNC_get_local_storage:
4862                 return &bpf_get_local_storage_proto;
4863         default:
4864                 return bpf_base_func_proto(func_id);
4865         }
4866 }
4867
4868 static const struct bpf_func_proto *
4869 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4870 {
4871         switch (func_id) {
4872         /* inet and inet6 sockets are created in a process
4873          * context so there is always a valid uid/gid
4874          */
4875         case BPF_FUNC_get_current_uid_gid:
4876                 return &bpf_get_current_uid_gid_proto;
4877         case BPF_FUNC_bind:
4878                 switch (prog->expected_attach_type) {
4879                 case BPF_CGROUP_INET4_CONNECT:
4880                 case BPF_CGROUP_INET6_CONNECT:
4881                         return &bpf_bind_proto;
4882                 default:
4883                         return NULL;
4884                 }
4885         case BPF_FUNC_get_socket_cookie:
4886                 return &bpf_get_socket_cookie_sock_addr_proto;
4887         case BPF_FUNC_get_local_storage:
4888                 return &bpf_get_local_storage_proto;
4889         default:
4890                 return bpf_base_func_proto(func_id);
4891         }
4892 }
4893
4894 static const struct bpf_func_proto *
4895 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4896 {
4897         switch (func_id) {
4898         case BPF_FUNC_skb_load_bytes:
4899                 return &bpf_skb_load_bytes_proto;
4900         case BPF_FUNC_skb_load_bytes_relative:
4901                 return &bpf_skb_load_bytes_relative_proto;
4902         case BPF_FUNC_get_socket_cookie:
4903                 return &bpf_get_socket_cookie_proto;
4904         case BPF_FUNC_get_socket_uid:
4905                 return &bpf_get_socket_uid_proto;
4906         default:
4907                 return bpf_base_func_proto(func_id);
4908         }
4909 }
4910
4911 static const struct bpf_func_proto *
4912 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4913 {
4914         switch (func_id) {
4915         case BPF_FUNC_get_local_storage:
4916                 return &bpf_get_local_storage_proto;
4917         default:
4918                 return sk_filter_func_proto(func_id, prog);
4919         }
4920 }
4921
4922 static const struct bpf_func_proto *
4923 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4924 {
4925         switch (func_id) {
4926         case BPF_FUNC_skb_store_bytes:
4927                 return &bpf_skb_store_bytes_proto;
4928         case BPF_FUNC_skb_load_bytes:
4929                 return &bpf_skb_load_bytes_proto;
4930         case BPF_FUNC_skb_load_bytes_relative:
4931                 return &bpf_skb_load_bytes_relative_proto;
4932         case BPF_FUNC_skb_pull_data:
4933                 return &bpf_skb_pull_data_proto;
4934         case BPF_FUNC_csum_diff:
4935                 return &bpf_csum_diff_proto;
4936         case BPF_FUNC_csum_update:
4937                 return &bpf_csum_update_proto;
4938         case BPF_FUNC_l3_csum_replace:
4939                 return &bpf_l3_csum_replace_proto;
4940         case BPF_FUNC_l4_csum_replace:
4941                 return &bpf_l4_csum_replace_proto;
4942         case BPF_FUNC_clone_redirect:
4943                 return &bpf_clone_redirect_proto;
4944         case BPF_FUNC_get_cgroup_classid:
4945                 return &bpf_get_cgroup_classid_proto;
4946         case BPF_FUNC_skb_vlan_push:
4947                 return &bpf_skb_vlan_push_proto;
4948         case BPF_FUNC_skb_vlan_pop:
4949                 return &bpf_skb_vlan_pop_proto;
4950         case BPF_FUNC_skb_change_proto:
4951                 return &bpf_skb_change_proto_proto;
4952         case BPF_FUNC_skb_change_type:
4953                 return &bpf_skb_change_type_proto;
4954         case BPF_FUNC_skb_adjust_room:
4955                 return &bpf_skb_adjust_room_proto;
4956         case BPF_FUNC_skb_change_tail:
4957                 return &bpf_skb_change_tail_proto;
4958         case BPF_FUNC_skb_get_tunnel_key:
4959                 return &bpf_skb_get_tunnel_key_proto;
4960         case BPF_FUNC_skb_set_tunnel_key:
4961                 return bpf_get_skb_set_tunnel_proto(func_id);
4962         case BPF_FUNC_skb_get_tunnel_opt:
4963                 return &bpf_skb_get_tunnel_opt_proto;
4964         case BPF_FUNC_skb_set_tunnel_opt:
4965                 return bpf_get_skb_set_tunnel_proto(func_id);
4966         case BPF_FUNC_redirect:
4967                 return &bpf_redirect_proto;
4968         case BPF_FUNC_get_route_realm:
4969                 return &bpf_get_route_realm_proto;
4970         case BPF_FUNC_get_hash_recalc:
4971                 return &bpf_get_hash_recalc_proto;
4972         case BPF_FUNC_set_hash_invalid:
4973                 return &bpf_set_hash_invalid_proto;
4974         case BPF_FUNC_set_hash:
4975                 return &bpf_set_hash_proto;
4976         case BPF_FUNC_perf_event_output:
4977                 return &bpf_skb_event_output_proto;
4978         case BPF_FUNC_get_smp_processor_id:
4979                 return &bpf_get_smp_processor_id_proto;
4980         case BPF_FUNC_skb_under_cgroup:
4981                 return &bpf_skb_under_cgroup_proto;
4982         case BPF_FUNC_get_socket_cookie:
4983                 return &bpf_get_socket_cookie_proto;
4984         case BPF_FUNC_get_socket_uid:
4985                 return &bpf_get_socket_uid_proto;
4986         case BPF_FUNC_fib_lookup:
4987                 return &bpf_skb_fib_lookup_proto;
4988 #ifdef CONFIG_XFRM
4989         case BPF_FUNC_skb_get_xfrm_state:
4990                 return &bpf_skb_get_xfrm_state_proto;
4991 #endif
4992 #ifdef CONFIG_SOCK_CGROUP_DATA
4993         case BPF_FUNC_skb_cgroup_id:
4994                 return &bpf_skb_cgroup_id_proto;
4995         case BPF_FUNC_skb_ancestor_cgroup_id:
4996                 return &bpf_skb_ancestor_cgroup_id_proto;
4997 #endif
4998         default:
4999                 return bpf_base_func_proto(func_id);
5000         }
5001 }
5002
5003 static const struct bpf_func_proto *
5004 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5005 {
5006         switch (func_id) {
5007         case BPF_FUNC_perf_event_output:
5008                 return &bpf_xdp_event_output_proto;
5009         case BPF_FUNC_get_smp_processor_id:
5010                 return &bpf_get_smp_processor_id_proto;
5011         case BPF_FUNC_csum_diff:
5012                 return &bpf_csum_diff_proto;
5013         case BPF_FUNC_xdp_adjust_head:
5014                 return &bpf_xdp_adjust_head_proto;
5015         case BPF_FUNC_xdp_adjust_meta:
5016                 return &bpf_xdp_adjust_meta_proto;
5017         case BPF_FUNC_redirect:
5018                 return &bpf_xdp_redirect_proto;
5019         case BPF_FUNC_redirect_map:
5020                 return &bpf_xdp_redirect_map_proto;
5021         case BPF_FUNC_xdp_adjust_tail:
5022                 return &bpf_xdp_adjust_tail_proto;
5023         case BPF_FUNC_fib_lookup:
5024                 return &bpf_xdp_fib_lookup_proto;
5025         default:
5026                 return bpf_base_func_proto(func_id);
5027         }
5028 }
5029
5030 static const struct bpf_func_proto *
5031 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5032 {
5033         switch (func_id) {
5034         case BPF_FUNC_setsockopt:
5035                 return &bpf_setsockopt_proto;
5036         case BPF_FUNC_getsockopt:
5037                 return &bpf_getsockopt_proto;
5038         case BPF_FUNC_sock_ops_cb_flags_set:
5039                 return &bpf_sock_ops_cb_flags_set_proto;
5040         case BPF_FUNC_sock_map_update:
5041                 return &bpf_sock_map_update_proto;
5042         case BPF_FUNC_sock_hash_update:
5043                 return &bpf_sock_hash_update_proto;
5044         case BPF_FUNC_get_socket_cookie:
5045                 return &bpf_get_socket_cookie_sock_ops_proto;
5046         case BPF_FUNC_get_local_storage:
5047                 return &bpf_get_local_storage_proto;
5048         default:
5049                 return bpf_base_func_proto(func_id);
5050         }
5051 }
5052
5053 static const struct bpf_func_proto *
5054 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5055 {
5056         switch (func_id) {
5057         case BPF_FUNC_msg_redirect_map:
5058                 return &bpf_msg_redirect_map_proto;
5059         case BPF_FUNC_msg_redirect_hash:
5060                 return &bpf_msg_redirect_hash_proto;
5061         case BPF_FUNC_msg_apply_bytes:
5062                 return &bpf_msg_apply_bytes_proto;
5063         case BPF_FUNC_msg_cork_bytes:
5064                 return &bpf_msg_cork_bytes_proto;
5065         case BPF_FUNC_msg_pull_data:
5066                 return &bpf_msg_pull_data_proto;
5067         case BPF_FUNC_get_local_storage:
5068                 return &bpf_get_local_storage_proto;
5069         default:
5070                 return bpf_base_func_proto(func_id);
5071         }
5072 }
5073
5074 static const struct bpf_func_proto *
5075 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5076 {
5077         switch (func_id) {
5078         case BPF_FUNC_skb_store_bytes:
5079                 return &bpf_skb_store_bytes_proto;
5080         case BPF_FUNC_skb_load_bytes:
5081                 return &bpf_skb_load_bytes_proto;
5082         case BPF_FUNC_skb_pull_data:
5083                 return &sk_skb_pull_data_proto;
5084         case BPF_FUNC_skb_change_tail:
5085                 return &sk_skb_change_tail_proto;
5086         case BPF_FUNC_skb_change_head:
5087                 return &sk_skb_change_head_proto;
5088         case BPF_FUNC_get_socket_cookie:
5089                 return &bpf_get_socket_cookie_proto;
5090         case BPF_FUNC_get_socket_uid:
5091                 return &bpf_get_socket_uid_proto;
5092         case BPF_FUNC_sk_redirect_map:
5093                 return &bpf_sk_redirect_map_proto;
5094         case BPF_FUNC_sk_redirect_hash:
5095                 return &bpf_sk_redirect_hash_proto;
5096         case BPF_FUNC_get_local_storage:
5097                 return &bpf_get_local_storage_proto;
5098         default:
5099                 return bpf_base_func_proto(func_id);
5100         }
5101 }
5102
5103 static const struct bpf_func_proto *
5104 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5105 {
5106         switch (func_id) {
5107         case BPF_FUNC_skb_load_bytes:
5108                 return &bpf_skb_load_bytes_proto;
5109         case BPF_FUNC_skb_pull_data:
5110                 return &bpf_skb_pull_data_proto;
5111         case BPF_FUNC_csum_diff:
5112                 return &bpf_csum_diff_proto;
5113         case BPF_FUNC_get_cgroup_classid:
5114                 return &bpf_get_cgroup_classid_proto;
5115         case BPF_FUNC_get_route_realm:
5116                 return &bpf_get_route_realm_proto;
5117         case BPF_FUNC_get_hash_recalc:
5118                 return &bpf_get_hash_recalc_proto;
5119         case BPF_FUNC_perf_event_output:
5120                 return &bpf_skb_event_output_proto;
5121         case BPF_FUNC_get_smp_processor_id:
5122                 return &bpf_get_smp_processor_id_proto;
5123         case BPF_FUNC_skb_under_cgroup:
5124                 return &bpf_skb_under_cgroup_proto;
5125         default:
5126                 return bpf_base_func_proto(func_id);
5127         }
5128 }
5129
5130 static const struct bpf_func_proto *
5131 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5132 {
5133         switch (func_id) {
5134         case BPF_FUNC_lwt_push_encap:
5135                 return &bpf_lwt_push_encap_proto;
5136         default:
5137                 return lwt_out_func_proto(func_id, prog);
5138         }
5139 }
5140
5141 static const struct bpf_func_proto *
5142 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5143 {
5144         switch (func_id) {
5145         case BPF_FUNC_skb_get_tunnel_key:
5146                 return &bpf_skb_get_tunnel_key_proto;
5147         case BPF_FUNC_skb_set_tunnel_key:
5148                 return bpf_get_skb_set_tunnel_proto(func_id);
5149         case BPF_FUNC_skb_get_tunnel_opt:
5150                 return &bpf_skb_get_tunnel_opt_proto;
5151         case BPF_FUNC_skb_set_tunnel_opt:
5152                 return bpf_get_skb_set_tunnel_proto(func_id);
5153         case BPF_FUNC_redirect:
5154                 return &bpf_redirect_proto;
5155         case BPF_FUNC_clone_redirect:
5156                 return &bpf_clone_redirect_proto;
5157         case BPF_FUNC_skb_change_tail:
5158                 return &bpf_skb_change_tail_proto;
5159         case BPF_FUNC_skb_change_head:
5160                 return &bpf_skb_change_head_proto;
5161         case BPF_FUNC_skb_store_bytes:
5162                 return &bpf_skb_store_bytes_proto;
5163         case BPF_FUNC_csum_update:
5164                 return &bpf_csum_update_proto;
5165         case BPF_FUNC_l3_csum_replace:
5166                 return &bpf_l3_csum_replace_proto;
5167         case BPF_FUNC_l4_csum_replace:
5168                 return &bpf_l4_csum_replace_proto;
5169         case BPF_FUNC_set_hash_invalid:
5170                 return &bpf_set_hash_invalid_proto;
5171         default:
5172                 return lwt_out_func_proto(func_id, prog);
5173         }
5174 }
5175
5176 static const struct bpf_func_proto *
5177 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5178 {
5179         switch (func_id) {
5180 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5181         case BPF_FUNC_lwt_seg6_store_bytes:
5182                 return &bpf_lwt_seg6_store_bytes_proto;
5183         case BPF_FUNC_lwt_seg6_action:
5184                 return &bpf_lwt_seg6_action_proto;
5185         case BPF_FUNC_lwt_seg6_adjust_srh:
5186                 return &bpf_lwt_seg6_adjust_srh_proto;
5187 #endif
5188         default:
5189                 return lwt_out_func_proto(func_id, prog);
5190         }
5191 }
5192
5193 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5194                                     const struct bpf_prog *prog,
5195                                     struct bpf_insn_access_aux *info)
5196 {
5197         const int size_default = sizeof(__u32);
5198
5199         if (off < 0 || off >= sizeof(struct __sk_buff))
5200                 return false;
5201
5202         /* The verifier guarantees that size > 0. */
5203         if (off % size != 0)
5204                 return false;
5205
5206         switch (off) {
5207         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5208                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5209                         return false;
5210                 break;
5211         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5212         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5213         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5214         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5215         case bpf_ctx_range(struct __sk_buff, data):
5216         case bpf_ctx_range(struct __sk_buff, data_meta):
5217         case bpf_ctx_range(struct __sk_buff, data_end):
5218                 if (size != size_default)
5219                         return false;
5220                 break;
5221         default:
5222                 /* Only narrow read access allowed for now. */
5223                 if (type == BPF_WRITE) {
5224                         if (size != size_default)
5225                                 return false;
5226                 } else {
5227                         bpf_ctx_record_field_size(info, size_default);
5228                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5229                                 return false;
5230                 }
5231         }
5232
5233         return true;
5234 }
5235
5236 static bool sk_filter_is_valid_access(int off, int size,
5237                                       enum bpf_access_type type,
5238                                       const struct bpf_prog *prog,
5239                                       struct bpf_insn_access_aux *info)
5240 {
5241         switch (off) {
5242         case bpf_ctx_range(struct __sk_buff, tc_classid):
5243         case bpf_ctx_range(struct __sk_buff, data):
5244         case bpf_ctx_range(struct __sk_buff, data_meta):
5245         case bpf_ctx_range(struct __sk_buff, data_end):
5246         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5247                 return false;
5248         }
5249
5250         if (type == BPF_WRITE) {
5251                 switch (off) {
5252                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5253                         break;
5254                 default:
5255                         return false;
5256                 }
5257         }
5258
5259         return bpf_skb_is_valid_access(off, size, type, prog, info);
5260 }
5261
5262 static bool lwt_is_valid_access(int off, int size,
5263                                 enum bpf_access_type type,
5264                                 const struct bpf_prog *prog,
5265                                 struct bpf_insn_access_aux *info)
5266 {
5267         switch (off) {
5268         case bpf_ctx_range(struct __sk_buff, tc_classid):
5269         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5270         case bpf_ctx_range(struct __sk_buff, data_meta):
5271                 return false;
5272         }
5273
5274         if (type == BPF_WRITE) {
5275                 switch (off) {
5276                 case bpf_ctx_range(struct __sk_buff, mark):
5277                 case bpf_ctx_range(struct __sk_buff, priority):
5278                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5279                         break;
5280                 default:
5281                         return false;
5282                 }
5283         }
5284
5285         switch (off) {
5286         case bpf_ctx_range(struct __sk_buff, data):
5287                 info->reg_type = PTR_TO_PACKET;
5288                 break;
5289         case bpf_ctx_range(struct __sk_buff, data_end):
5290                 info->reg_type = PTR_TO_PACKET_END;
5291                 break;
5292         }
5293
5294         return bpf_skb_is_valid_access(off, size, type, prog, info);
5295 }
5296
5297 /* Attach type specific accesses */
5298 static bool __sock_filter_check_attach_type(int off,
5299                                             enum bpf_access_type access_type,
5300                                             enum bpf_attach_type attach_type)
5301 {
5302         switch (off) {
5303         case offsetof(struct bpf_sock, bound_dev_if):
5304         case offsetof(struct bpf_sock, mark):
5305         case offsetof(struct bpf_sock, priority):
5306                 switch (attach_type) {
5307                 case BPF_CGROUP_INET_SOCK_CREATE:
5308                         goto full_access;
5309                 default:
5310                         return false;
5311                 }
5312         case bpf_ctx_range(struct bpf_sock, src_ip4):
5313                 switch (attach_type) {
5314                 case BPF_CGROUP_INET4_POST_BIND:
5315                         goto read_only;
5316                 default:
5317                         return false;
5318                 }
5319         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5320                 switch (attach_type) {
5321                 case BPF_CGROUP_INET6_POST_BIND:
5322                         goto read_only;
5323                 default:
5324                         return false;
5325                 }
5326         case bpf_ctx_range(struct bpf_sock, src_port):
5327                 switch (attach_type) {
5328                 case BPF_CGROUP_INET4_POST_BIND:
5329                 case BPF_CGROUP_INET6_POST_BIND:
5330                         goto read_only;
5331                 default:
5332                         return false;
5333                 }
5334         }
5335 read_only:
5336         return access_type == BPF_READ;
5337 full_access:
5338         return true;
5339 }
5340
5341 static bool __sock_filter_check_size(int off, int size,
5342                                      struct bpf_insn_access_aux *info)
5343 {
5344         const int size_default = sizeof(__u32);
5345
5346         switch (off) {
5347         case bpf_ctx_range(struct bpf_sock, src_ip4):
5348         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5349                 bpf_ctx_record_field_size(info, size_default);
5350                 return bpf_ctx_narrow_access_ok(off, size, size_default);
5351         }
5352
5353         return size == size_default;
5354 }
5355
5356 static bool sock_filter_is_valid_access(int off, int size,
5357                                         enum bpf_access_type type,
5358                                         const struct bpf_prog *prog,
5359                                         struct bpf_insn_access_aux *info)
5360 {
5361         if (off < 0 || off >= sizeof(struct bpf_sock))
5362                 return false;
5363         if (off % size != 0)
5364                 return false;
5365         if (!__sock_filter_check_attach_type(off, type,
5366                                              prog->expected_attach_type))
5367                 return false;
5368         if (!__sock_filter_check_size(off, size, info))
5369                 return false;
5370         return true;
5371 }
5372
5373 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5374                                 const struct bpf_prog *prog, int drop_verdict)
5375 {
5376         struct bpf_insn *insn = insn_buf;
5377
5378         if (!direct_write)
5379                 return 0;
5380
5381         /* if (!skb->cloned)
5382          *       goto start;
5383          *
5384          * (Fast-path, otherwise approximation that we might be
5385          *  a clone, do the rest in helper.)
5386          */
5387         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5388         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5389         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5390
5391         /* ret = bpf_skb_pull_data(skb, 0); */
5392         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5393         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5394         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5395                                BPF_FUNC_skb_pull_data);
5396         /* if (!ret)
5397          *      goto restore;
5398          * return TC_ACT_SHOT;
5399          */
5400         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5401         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5402         *insn++ = BPF_EXIT_INSN();
5403
5404         /* restore: */
5405         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5406         /* start: */
5407         *insn++ = prog->insnsi[0];
5408
5409         return insn - insn_buf;
5410 }
5411
5412 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5413                           struct bpf_insn *insn_buf)
5414 {
5415         bool indirect = BPF_MODE(orig->code) == BPF_IND;
5416         struct bpf_insn *insn = insn_buf;
5417
5418         /* We're guaranteed here that CTX is in R6. */
5419         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5420         if (!indirect) {
5421                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5422         } else {
5423                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5424                 if (orig->imm)
5425                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5426         }
5427
5428         switch (BPF_SIZE(orig->code)) {
5429         case BPF_B:
5430                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5431                 break;
5432         case BPF_H:
5433                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5434                 break;
5435         case BPF_W:
5436                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5437                 break;
5438         }
5439
5440         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5441         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5442         *insn++ = BPF_EXIT_INSN();
5443
5444         return insn - insn_buf;
5445 }
5446
5447 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5448                                const struct bpf_prog *prog)
5449 {
5450         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5451 }
5452
5453 static bool tc_cls_act_is_valid_access(int off, int size,
5454                                        enum bpf_access_type type,
5455                                        const struct bpf_prog *prog,
5456                                        struct bpf_insn_access_aux *info)
5457 {
5458         if (type == BPF_WRITE) {
5459                 switch (off) {
5460                 case bpf_ctx_range(struct __sk_buff, mark):
5461                 case bpf_ctx_range(struct __sk_buff, tc_index):
5462                 case bpf_ctx_range(struct __sk_buff, priority):
5463                 case bpf_ctx_range(struct __sk_buff, tc_classid):
5464                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5465                         break;
5466                 default:
5467                         return false;
5468                 }
5469         }
5470
5471         switch (off) {
5472         case bpf_ctx_range(struct __sk_buff, data):
5473                 info->reg_type = PTR_TO_PACKET;
5474                 break;
5475         case bpf_ctx_range(struct __sk_buff, data_meta):
5476                 info->reg_type = PTR_TO_PACKET_META;
5477                 break;
5478         case bpf_ctx_range(struct __sk_buff, data_end):
5479                 info->reg_type = PTR_TO_PACKET_END;
5480                 break;
5481         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5482                 return false;
5483         }
5484
5485         return bpf_skb_is_valid_access(off, size, type, prog, info);
5486 }
5487
5488 static bool __is_valid_xdp_access(int off, int size)
5489 {
5490         if (off < 0 || off >= sizeof(struct xdp_md))
5491                 return false;
5492         if (off % size != 0)
5493                 return false;
5494         if (size != sizeof(__u32))
5495                 return false;
5496
5497         return true;
5498 }
5499
5500 static bool xdp_is_valid_access(int off, int size,
5501                                 enum bpf_access_type type,
5502                                 const struct bpf_prog *prog,
5503                                 struct bpf_insn_access_aux *info)
5504 {
5505         if (type == BPF_WRITE) {
5506                 if (bpf_prog_is_dev_bound(prog->aux)) {
5507                         switch (off) {
5508                         case offsetof(struct xdp_md, rx_queue_index):
5509                                 return __is_valid_xdp_access(off, size);
5510                         }
5511                 }
5512                 return false;
5513         }
5514
5515         switch (off) {
5516         case offsetof(struct xdp_md, data):
5517                 info->reg_type = PTR_TO_PACKET;
5518                 break;
5519         case offsetof(struct xdp_md, data_meta):
5520                 info->reg_type = PTR_TO_PACKET_META;
5521                 break;
5522         case offsetof(struct xdp_md, data_end):
5523                 info->reg_type = PTR_TO_PACKET_END;
5524                 break;
5525         }
5526
5527         return __is_valid_xdp_access(off, size);
5528 }
5529
5530 void bpf_warn_invalid_xdp_action(u32 act)
5531 {
5532         const u32 act_max = XDP_REDIRECT;
5533
5534         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5535                   act > act_max ? "Illegal" : "Driver unsupported",
5536                   act);
5537 }
5538 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5539
5540 static bool sock_addr_is_valid_access(int off, int size,
5541                                       enum bpf_access_type type,
5542                                       const struct bpf_prog *prog,
5543                                       struct bpf_insn_access_aux *info)
5544 {
5545         const int size_default = sizeof(__u32);
5546
5547         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5548                 return false;
5549         if (off % size != 0)
5550                 return false;
5551
5552         /* Disallow access to IPv6 fields from IPv4 contex and vise
5553          * versa.
5554          */
5555         switch (off) {
5556         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5557                 switch (prog->expected_attach_type) {
5558                 case BPF_CGROUP_INET4_BIND:
5559                 case BPF_CGROUP_INET4_CONNECT:
5560                 case BPF_CGROUP_UDP4_SENDMSG:
5561                         break;
5562                 default:
5563                         return false;
5564                 }
5565                 break;
5566         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5567                 switch (prog->expected_attach_type) {
5568                 case BPF_CGROUP_INET6_BIND:
5569                 case BPF_CGROUP_INET6_CONNECT:
5570                 case BPF_CGROUP_UDP6_SENDMSG:
5571                         break;
5572                 default:
5573                         return false;
5574                 }
5575                 break;
5576         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5577                 switch (prog->expected_attach_type) {
5578                 case BPF_CGROUP_UDP4_SENDMSG:
5579                         break;
5580                 default:
5581                         return false;
5582                 }
5583                 break;
5584         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5585                                 msg_src_ip6[3]):
5586                 switch (prog->expected_attach_type) {
5587                 case BPF_CGROUP_UDP6_SENDMSG:
5588                         break;
5589                 default:
5590                         return false;
5591                 }
5592                 break;
5593         }
5594
5595         switch (off) {
5596         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5597         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5598         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5599         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5600                                 msg_src_ip6[3]):
5601                 /* Only narrow read access allowed for now. */
5602                 if (type == BPF_READ) {
5603                         bpf_ctx_record_field_size(info, size_default);
5604                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5605                                 return false;
5606                 } else {
5607                         if (size != size_default)
5608                                 return false;
5609                 }
5610                 break;
5611         case bpf_ctx_range(struct bpf_sock_addr, user_port):
5612                 if (size != size_default)
5613                         return false;
5614                 break;
5615         default:
5616                 if (type == BPF_READ) {
5617                         if (size != size_default)
5618                                 return false;
5619                 } else {
5620                         return false;
5621                 }
5622         }
5623
5624         return true;
5625 }
5626
5627 static bool sock_ops_is_valid_access(int off, int size,
5628                                      enum bpf_access_type type,
5629                                      const struct bpf_prog *prog,
5630                                      struct bpf_insn_access_aux *info)
5631 {
5632         const int size_default = sizeof(__u32);
5633
5634         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5635                 return false;
5636
5637         /* The verifier guarantees that size > 0. */
5638         if (off % size != 0)
5639                 return false;
5640
5641         if (type == BPF_WRITE) {
5642                 switch (off) {
5643                 case offsetof(struct bpf_sock_ops, reply):
5644                 case offsetof(struct bpf_sock_ops, sk_txhash):
5645                         if (size != size_default)
5646                                 return false;
5647                         break;
5648                 default:
5649                         return false;
5650                 }
5651         } else {
5652                 switch (off) {
5653                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5654                                         bytes_acked):
5655                         if (size != sizeof(__u64))
5656                                 return false;
5657                         break;
5658                 default:
5659                         if (size != size_default)
5660                                 return false;
5661                         break;
5662                 }
5663         }
5664
5665         return true;
5666 }
5667
5668 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5669                            const struct bpf_prog *prog)
5670 {
5671         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5672 }
5673
5674 static bool sk_skb_is_valid_access(int off, int size,
5675                                    enum bpf_access_type type,
5676                                    const struct bpf_prog *prog,
5677                                    struct bpf_insn_access_aux *info)
5678 {
5679         switch (off) {
5680         case bpf_ctx_range(struct __sk_buff, tc_classid):
5681         case bpf_ctx_range(struct __sk_buff, data_meta):
5682                 return false;
5683         }
5684
5685         if (type == BPF_WRITE) {
5686                 switch (off) {
5687                 case bpf_ctx_range(struct __sk_buff, tc_index):
5688                 case bpf_ctx_range(struct __sk_buff, priority):
5689                         break;
5690                 default:
5691                         return false;
5692                 }
5693         }
5694
5695         switch (off) {
5696         case bpf_ctx_range(struct __sk_buff, mark):
5697                 return false;
5698         case bpf_ctx_range(struct __sk_buff, data):
5699                 info->reg_type = PTR_TO_PACKET;
5700                 break;
5701         case bpf_ctx_range(struct __sk_buff, data_end):
5702                 info->reg_type = PTR_TO_PACKET_END;
5703                 break;
5704         }
5705
5706         return bpf_skb_is_valid_access(off, size, type, prog, info);
5707 }
5708
5709 static bool sk_msg_is_valid_access(int off, int size,
5710                                    enum bpf_access_type type,
5711                                    const struct bpf_prog *prog,
5712                                    struct bpf_insn_access_aux *info)
5713 {
5714         if (type == BPF_WRITE)
5715                 return false;
5716
5717         switch (off) {
5718         case offsetof(struct sk_msg_md, data):
5719                 info->reg_type = PTR_TO_PACKET;
5720                 if (size != sizeof(__u64))
5721                         return false;
5722                 break;
5723         case offsetof(struct sk_msg_md, data_end):
5724                 info->reg_type = PTR_TO_PACKET_END;
5725                 if (size != sizeof(__u64))
5726                         return false;
5727                 break;
5728         default:
5729                 if (size != sizeof(__u32))
5730                         return false;
5731         }
5732
5733         if (off < 0 || off >= sizeof(struct sk_msg_md))
5734                 return false;
5735         if (off % size != 0)
5736                 return false;
5737
5738         return true;
5739 }
5740
5741 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5742                                   const struct bpf_insn *si,
5743                                   struct bpf_insn *insn_buf,
5744                                   struct bpf_prog *prog, u32 *target_size)
5745 {
5746         struct bpf_insn *insn = insn_buf;
5747         int off;
5748
5749         switch (si->off) {
5750         case offsetof(struct __sk_buff, len):
5751                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5752                                       bpf_target_off(struct sk_buff, len, 4,
5753                                                      target_size));
5754                 break;
5755
5756         case offsetof(struct __sk_buff, protocol):
5757                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5758                                       bpf_target_off(struct sk_buff, protocol, 2,
5759                                                      target_size));
5760                 break;
5761
5762         case offsetof(struct __sk_buff, vlan_proto):
5763                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5764                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
5765                                                      target_size));
5766                 break;
5767
5768         case offsetof(struct __sk_buff, priority):
5769                 if (type == BPF_WRITE)
5770                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5771                                               bpf_target_off(struct sk_buff, priority, 4,
5772                                                              target_size));
5773                 else
5774                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5775                                               bpf_target_off(struct sk_buff, priority, 4,
5776                                                              target_size));
5777                 break;
5778
5779         case offsetof(struct __sk_buff, ingress_ifindex):
5780                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5781                                       bpf_target_off(struct sk_buff, skb_iif, 4,
5782                                                      target_size));
5783                 break;
5784
5785         case offsetof(struct __sk_buff, ifindex):
5786                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5787                                       si->dst_reg, si->src_reg,
5788                                       offsetof(struct sk_buff, dev));
5789                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5790                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5791                                       bpf_target_off(struct net_device, ifindex, 4,
5792                                                      target_size));
5793                 break;
5794
5795         case offsetof(struct __sk_buff, hash):
5796                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5797                                       bpf_target_off(struct sk_buff, hash, 4,
5798                                                      target_size));
5799                 break;
5800
5801         case offsetof(struct __sk_buff, mark):
5802                 if (type == BPF_WRITE)
5803                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5804                                               bpf_target_off(struct sk_buff, mark, 4,
5805                                                              target_size));
5806                 else
5807                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5808                                               bpf_target_off(struct sk_buff, mark, 4,
5809                                                              target_size));
5810                 break;
5811
5812         case offsetof(struct __sk_buff, pkt_type):
5813                 *target_size = 1;
5814                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5815                                       PKT_TYPE_OFFSET());
5816                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5817 #ifdef __BIG_ENDIAN_BITFIELD
5818                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5819 #endif
5820                 break;
5821
5822         case offsetof(struct __sk_buff, queue_mapping):
5823                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5824                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
5825                                                      target_size));
5826                 break;
5827
5828         case offsetof(struct __sk_buff, vlan_present):
5829         case offsetof(struct __sk_buff, vlan_tci):
5830                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5831
5832                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5833                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
5834                                                      target_size));
5835                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5836                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5837                                                 ~VLAN_TAG_PRESENT);
5838                 } else {
5839                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5840                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5841                 }
5842                 break;
5843
5844         case offsetof(struct __sk_buff, cb[0]) ...
5845              offsetofend(struct __sk_buff, cb[4]) - 1:
5846                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5847                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5848                               offsetof(struct qdisc_skb_cb, data)) %
5849                              sizeof(__u64));
5850
5851                 prog->cb_access = 1;
5852                 off  = si->off;
5853                 off -= offsetof(struct __sk_buff, cb[0]);
5854                 off += offsetof(struct sk_buff, cb);
5855                 off += offsetof(struct qdisc_skb_cb, data);
5856                 if (type == BPF_WRITE)
5857                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5858                                               si->src_reg, off);
5859                 else
5860                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5861                                               si->src_reg, off);
5862                 break;
5863
5864         case offsetof(struct __sk_buff, tc_classid):
5865                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5866
5867                 off  = si->off;
5868                 off -= offsetof(struct __sk_buff, tc_classid);
5869                 off += offsetof(struct sk_buff, cb);
5870                 off += offsetof(struct qdisc_skb_cb, tc_classid);
5871                 *target_size = 2;
5872                 if (type == BPF_WRITE)
5873                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
5874                                               si->src_reg, off);
5875                 else
5876                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
5877                                               si->src_reg, off);
5878                 break;
5879
5880         case offsetof(struct __sk_buff, data):
5881                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
5882                                       si->dst_reg, si->src_reg,
5883                                       offsetof(struct sk_buff, data));
5884                 break;
5885
5886         case offsetof(struct __sk_buff, data_meta):
5887                 off  = si->off;
5888                 off -= offsetof(struct __sk_buff, data_meta);
5889                 off += offsetof(struct sk_buff, cb);
5890                 off += offsetof(struct bpf_skb_data_end, data_meta);
5891                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5892                                       si->src_reg, off);
5893                 break;
5894
5895         case offsetof(struct __sk_buff, data_end):
5896                 off  = si->off;
5897                 off -= offsetof(struct __sk_buff, data_end);
5898                 off += offsetof(struct sk_buff, cb);
5899                 off += offsetof(struct bpf_skb_data_end, data_end);
5900                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5901                                       si->src_reg, off);
5902                 break;
5903
5904         case offsetof(struct __sk_buff, tc_index):
5905 #ifdef CONFIG_NET_SCHED
5906                 if (type == BPF_WRITE)
5907                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
5908                                               bpf_target_off(struct sk_buff, tc_index, 2,
5909                                                              target_size));
5910                 else
5911                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5912                                               bpf_target_off(struct sk_buff, tc_index, 2,
5913                                                              target_size));
5914 #else
5915                 *target_size = 2;
5916                 if (type == BPF_WRITE)
5917                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
5918                 else
5919                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5920 #endif
5921                 break;
5922
5923         case offsetof(struct __sk_buff, napi_id):
5924 #if defined(CONFIG_NET_RX_BUSY_POLL)
5925                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5926                                       bpf_target_off(struct sk_buff, napi_id, 4,
5927                                                      target_size));
5928                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
5929                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5930 #else
5931                 *target_size = 4;
5932                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5933 #endif
5934                 break;
5935         case offsetof(struct __sk_buff, family):
5936                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
5937
5938                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5939                                       si->dst_reg, si->src_reg,
5940                                       offsetof(struct sk_buff, sk));
5941                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5942                                       bpf_target_off(struct sock_common,
5943                                                      skc_family,
5944                                                      2, target_size));
5945                 break;
5946         case offsetof(struct __sk_buff, remote_ip4):
5947                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
5948
5949                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5950                                       si->dst_reg, si->src_reg,
5951                                       offsetof(struct sk_buff, sk));
5952                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5953                                       bpf_target_off(struct sock_common,
5954                                                      skc_daddr,
5955                                                      4, target_size));
5956                 break;
5957         case offsetof(struct __sk_buff, local_ip4):
5958                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5959                                           skc_rcv_saddr) != 4);
5960
5961                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5962                                       si->dst_reg, si->src_reg,
5963                                       offsetof(struct sk_buff, sk));
5964                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5965                                       bpf_target_off(struct sock_common,
5966                                                      skc_rcv_saddr,
5967                                                      4, target_size));
5968                 break;
5969         case offsetof(struct __sk_buff, remote_ip6[0]) ...
5970              offsetof(struct __sk_buff, remote_ip6[3]):
5971 #if IS_ENABLED(CONFIG_IPV6)
5972                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5973                                           skc_v6_daddr.s6_addr32[0]) != 4);
5974
5975                 off = si->off;
5976                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
5977
5978                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5979                                       si->dst_reg, si->src_reg,
5980                                       offsetof(struct sk_buff, sk));
5981                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5982                                       offsetof(struct sock_common,
5983                                                skc_v6_daddr.s6_addr32[0]) +
5984                                       off);
5985 #else
5986                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5987 #endif
5988                 break;
5989         case offsetof(struct __sk_buff, local_ip6[0]) ...
5990              offsetof(struct __sk_buff, local_ip6[3]):
5991 #if IS_ENABLED(CONFIG_IPV6)
5992                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5993                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
5994
5995                 off = si->off;
5996                 off -= offsetof(struct __sk_buff, local_ip6[0]);
5997
5998                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5999                                       si->dst_reg, si->src_reg,
6000                                       offsetof(struct sk_buff, sk));
6001                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6002                                       offsetof(struct sock_common,
6003                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6004                                       off);
6005 #else
6006                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6007 #endif
6008                 break;
6009
6010         case offsetof(struct __sk_buff, remote_port):
6011                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6012
6013                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6014                                       si->dst_reg, si->src_reg,
6015                                       offsetof(struct sk_buff, sk));
6016                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6017                                       bpf_target_off(struct sock_common,
6018                                                      skc_dport,
6019                                                      2, target_size));
6020 #ifndef __BIG_ENDIAN_BITFIELD
6021                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6022 #endif
6023                 break;
6024
6025         case offsetof(struct __sk_buff, local_port):
6026                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6027
6028                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6029                                       si->dst_reg, si->src_reg,
6030                                       offsetof(struct sk_buff, sk));
6031                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6032                                       bpf_target_off(struct sock_common,
6033                                                      skc_num, 2, target_size));
6034                 break;
6035         }
6036
6037         return insn - insn_buf;
6038 }
6039
6040 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
6041                                           const struct bpf_insn *si,
6042                                           struct bpf_insn *insn_buf,
6043                                           struct bpf_prog *prog, u32 *target_size)
6044 {
6045         struct bpf_insn *insn = insn_buf;
6046         int off;
6047
6048         switch (si->off) {
6049         case offsetof(struct bpf_sock, bound_dev_if):
6050                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6051
6052                 if (type == BPF_WRITE)
6053                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6054                                         offsetof(struct sock, sk_bound_dev_if));
6055                 else
6056                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6057                                       offsetof(struct sock, sk_bound_dev_if));
6058                 break;
6059
6060         case offsetof(struct bpf_sock, mark):
6061                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6062
6063                 if (type == BPF_WRITE)
6064                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6065                                         offsetof(struct sock, sk_mark));
6066                 else
6067                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6068                                       offsetof(struct sock, sk_mark));
6069                 break;
6070
6071         case offsetof(struct bpf_sock, priority):
6072                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6073
6074                 if (type == BPF_WRITE)
6075                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6076                                         offsetof(struct sock, sk_priority));
6077                 else
6078                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6079                                       offsetof(struct sock, sk_priority));
6080                 break;
6081
6082         case offsetof(struct bpf_sock, family):
6083                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6084
6085                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6086                                       offsetof(struct sock, sk_family));
6087                 break;
6088
6089         case offsetof(struct bpf_sock, type):
6090                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6091                                       offsetof(struct sock, __sk_flags_offset));
6092                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6093                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6094                 break;
6095
6096         case offsetof(struct bpf_sock, protocol):
6097                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6098                                       offsetof(struct sock, __sk_flags_offset));
6099                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6100                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6101                 break;
6102
6103         case offsetof(struct bpf_sock, src_ip4):
6104                 *insn++ = BPF_LDX_MEM(
6105                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6106                         bpf_target_off(struct sock_common, skc_rcv_saddr,
6107                                        FIELD_SIZEOF(struct sock_common,
6108                                                     skc_rcv_saddr),
6109                                        target_size));
6110                 break;
6111
6112         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6113 #if IS_ENABLED(CONFIG_IPV6)
6114                 off = si->off;
6115                 off -= offsetof(struct bpf_sock, src_ip6[0]);
6116                 *insn++ = BPF_LDX_MEM(
6117                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6118                         bpf_target_off(
6119                                 struct sock_common,
6120                                 skc_v6_rcv_saddr.s6_addr32[0],
6121                                 FIELD_SIZEOF(struct sock_common,
6122                                              skc_v6_rcv_saddr.s6_addr32[0]),
6123                                 target_size) + off);
6124 #else
6125                 (void)off;
6126                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6127 #endif
6128                 break;
6129
6130         case offsetof(struct bpf_sock, src_port):
6131                 *insn++ = BPF_LDX_MEM(
6132                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6133                         si->dst_reg, si->src_reg,
6134                         bpf_target_off(struct sock_common, skc_num,
6135                                        FIELD_SIZEOF(struct sock_common,
6136                                                     skc_num),
6137                                        target_size));
6138                 break;
6139         }
6140
6141         return insn - insn_buf;
6142 }
6143
6144 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6145                                          const struct bpf_insn *si,
6146                                          struct bpf_insn *insn_buf,
6147                                          struct bpf_prog *prog, u32 *target_size)
6148 {
6149         struct bpf_insn *insn = insn_buf;
6150
6151         switch (si->off) {
6152         case offsetof(struct __sk_buff, ifindex):
6153                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6154                                       si->dst_reg, si->src_reg,
6155                                       offsetof(struct sk_buff, dev));
6156                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6157                                       bpf_target_off(struct net_device, ifindex, 4,
6158                                                      target_size));
6159                 break;
6160         default:
6161                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6162                                               target_size);
6163         }
6164
6165         return insn - insn_buf;
6166 }
6167
6168 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6169                                   const struct bpf_insn *si,
6170                                   struct bpf_insn *insn_buf,
6171                                   struct bpf_prog *prog, u32 *target_size)
6172 {
6173         struct bpf_insn *insn = insn_buf;
6174
6175         switch (si->off) {
6176         case offsetof(struct xdp_md, data):
6177                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6178                                       si->dst_reg, si->src_reg,
6179                                       offsetof(struct xdp_buff, data));
6180                 break;
6181         case offsetof(struct xdp_md, data_meta):
6182                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6183                                       si->dst_reg, si->src_reg,
6184                                       offsetof(struct xdp_buff, data_meta));
6185                 break;
6186         case offsetof(struct xdp_md, data_end):
6187                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6188                                       si->dst_reg, si->src_reg,
6189                                       offsetof(struct xdp_buff, data_end));
6190                 break;
6191         case offsetof(struct xdp_md, ingress_ifindex):
6192                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6193                                       si->dst_reg, si->src_reg,
6194                                       offsetof(struct xdp_buff, rxq));
6195                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6196                                       si->dst_reg, si->dst_reg,
6197                                       offsetof(struct xdp_rxq_info, dev));
6198                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6199                                       offsetof(struct net_device, ifindex));
6200                 break;
6201         case offsetof(struct xdp_md, rx_queue_index):
6202                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6203                                       si->dst_reg, si->src_reg,
6204                                       offsetof(struct xdp_buff, rxq));
6205                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6206                                       offsetof(struct xdp_rxq_info,
6207                                                queue_index));
6208                 break;
6209         }
6210
6211         return insn - insn_buf;
6212 }
6213
6214 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6215  * context Structure, F is Field in context structure that contains a pointer
6216  * to Nested Structure of type NS that has the field NF.
6217  *
6218  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6219  * sure that SIZE is not greater than actual size of S.F.NF.
6220  *
6221  * If offset OFF is provided, the load happens from that offset relative to
6222  * offset of NF.
6223  */
6224 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
6225         do {                                                                   \
6226                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
6227                                       si->src_reg, offsetof(S, F));            \
6228                 *insn++ = BPF_LDX_MEM(                                         \
6229                         SIZE, si->dst_reg, si->dst_reg,                        \
6230                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6231                                        target_size)                            \
6232                                 + OFF);                                        \
6233         } while (0)
6234
6235 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
6236         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
6237                                              BPF_FIELD_SIZEOF(NS, NF), 0)
6238
6239 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6240  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6241  *
6242  * It doesn't support SIZE argument though since narrow stores are not
6243  * supported for now.
6244  *
6245  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6246  * "register" since two registers available in convert_ctx_access are not
6247  * enough: we can't override neither SRC, since it contains value to store, nor
6248  * DST since it contains pointer to context that may be used by later
6249  * instructions. But we need a temporary place to save pointer to nested
6250  * structure whose field we want to store to.
6251  */
6252 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
6253         do {                                                                   \
6254                 int tmp_reg = BPF_REG_9;                                       \
6255                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6256                         --tmp_reg;                                             \
6257                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6258                         --tmp_reg;                                             \
6259                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
6260                                       offsetof(S, TF));                        \
6261                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
6262                                       si->dst_reg, offsetof(S, F));            \
6263                 *insn++ = BPF_STX_MEM(                                         \
6264                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
6265                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6266                                        target_size)                            \
6267                                 + OFF);                                        \
6268                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
6269                                       offsetof(S, TF));                        \
6270         } while (0)
6271
6272 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6273                                                       TF)                      \
6274         do {                                                                   \
6275                 if (type == BPF_WRITE) {                                       \
6276                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
6277                                                          TF);                  \
6278                 } else {                                                       \
6279                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
6280                                 S, NS, F, NF, SIZE, OFF);  \
6281                 }                                                              \
6282         } while (0)
6283
6284 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
6285         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
6286                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6287
6288 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6289                                         const struct bpf_insn *si,
6290                                         struct bpf_insn *insn_buf,
6291                                         struct bpf_prog *prog, u32 *target_size)
6292 {
6293         struct bpf_insn *insn = insn_buf;
6294         int off;
6295
6296         switch (si->off) {
6297         case offsetof(struct bpf_sock_addr, user_family):
6298                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6299                                             struct sockaddr, uaddr, sa_family);
6300                 break;
6301
6302         case offsetof(struct bpf_sock_addr, user_ip4):
6303                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6304                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6305                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6306                 break;
6307
6308         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6309                 off = si->off;
6310                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6311                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6312                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6313                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6314                         tmp_reg);
6315                 break;
6316
6317         case offsetof(struct bpf_sock_addr, user_port):
6318                 /* To get port we need to know sa_family first and then treat
6319                  * sockaddr as either sockaddr_in or sockaddr_in6.
6320                  * Though we can simplify since port field has same offset and
6321                  * size in both structures.
6322                  * Here we check this invariant and use just one of the
6323                  * structures if it's true.
6324                  */
6325                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6326                              offsetof(struct sockaddr_in6, sin6_port));
6327                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6328                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6329                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6330                                                      struct sockaddr_in6, uaddr,
6331                                                      sin6_port, tmp_reg);
6332                 break;
6333
6334         case offsetof(struct bpf_sock_addr, family):
6335                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6336                                             struct sock, sk, sk_family);
6337                 break;
6338
6339         case offsetof(struct bpf_sock_addr, type):
6340                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6341                         struct bpf_sock_addr_kern, struct sock, sk,
6342                         __sk_flags_offset, BPF_W, 0);
6343                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6344                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6345                 break;
6346
6347         case offsetof(struct bpf_sock_addr, protocol):
6348                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6349                         struct bpf_sock_addr_kern, struct sock, sk,
6350                         __sk_flags_offset, BPF_W, 0);
6351                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6352                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6353                                         SK_FL_PROTO_SHIFT);
6354                 break;
6355
6356         case offsetof(struct bpf_sock_addr, msg_src_ip4):
6357                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6358                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6359                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6360                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6361                 break;
6362
6363         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6364                                 msg_src_ip6[3]):
6365                 off = si->off;
6366                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6367                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6368                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6369                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6370                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6371                 break;
6372         }
6373
6374         return insn - insn_buf;
6375 }
6376
6377 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6378                                        const struct bpf_insn *si,
6379                                        struct bpf_insn *insn_buf,
6380                                        struct bpf_prog *prog,
6381                                        u32 *target_size)
6382 {
6383         struct bpf_insn *insn = insn_buf;
6384         int off;
6385
6386         switch (si->off) {
6387         case offsetof(struct bpf_sock_ops, op) ...
6388              offsetof(struct bpf_sock_ops, replylong[3]):
6389                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6390                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6391                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6392                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6393                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6394                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6395                 off = si->off;
6396                 off -= offsetof(struct bpf_sock_ops, op);
6397                 off += offsetof(struct bpf_sock_ops_kern, op);
6398                 if (type == BPF_WRITE)
6399                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6400                                               off);
6401                 else
6402                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6403                                               off);
6404                 break;
6405
6406         case offsetof(struct bpf_sock_ops, family):
6407                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6408
6409                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6410                                               struct bpf_sock_ops_kern, sk),
6411                                       si->dst_reg, si->src_reg,
6412                                       offsetof(struct bpf_sock_ops_kern, sk));
6413                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6414                                       offsetof(struct sock_common, skc_family));
6415                 break;
6416
6417         case offsetof(struct bpf_sock_ops, remote_ip4):
6418                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6419
6420                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6421                                                 struct bpf_sock_ops_kern, sk),
6422                                       si->dst_reg, si->src_reg,
6423                                       offsetof(struct bpf_sock_ops_kern, sk));
6424                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6425                                       offsetof(struct sock_common, skc_daddr));
6426                 break;
6427
6428         case offsetof(struct bpf_sock_ops, local_ip4):
6429                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6430                                           skc_rcv_saddr) != 4);
6431
6432                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6433                                               struct bpf_sock_ops_kern, sk),
6434                                       si->dst_reg, si->src_reg,
6435                                       offsetof(struct bpf_sock_ops_kern, sk));
6436                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6437                                       offsetof(struct sock_common,
6438                                                skc_rcv_saddr));
6439                 break;
6440
6441         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6442              offsetof(struct bpf_sock_ops, remote_ip6[3]):
6443 #if IS_ENABLED(CONFIG_IPV6)
6444                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6445                                           skc_v6_daddr.s6_addr32[0]) != 4);
6446
6447                 off = si->off;
6448                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6449                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6450                                                 struct bpf_sock_ops_kern, sk),
6451                                       si->dst_reg, si->src_reg,
6452                                       offsetof(struct bpf_sock_ops_kern, sk));
6453                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6454                                       offsetof(struct sock_common,
6455                                                skc_v6_daddr.s6_addr32[0]) +
6456                                       off);
6457 #else
6458                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6459 #endif
6460                 break;
6461
6462         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6463              offsetof(struct bpf_sock_ops, local_ip6[3]):
6464 #if IS_ENABLED(CONFIG_IPV6)
6465                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6466                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6467
6468                 off = si->off;
6469                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6470                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6471                                                 struct bpf_sock_ops_kern, sk),
6472                                       si->dst_reg, si->src_reg,
6473                                       offsetof(struct bpf_sock_ops_kern, sk));
6474                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6475                                       offsetof(struct sock_common,
6476                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6477                                       off);
6478 #else
6479                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6480 #endif
6481                 break;
6482
6483         case offsetof(struct bpf_sock_ops, remote_port):
6484                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6485
6486                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6487                                                 struct bpf_sock_ops_kern, sk),
6488                                       si->dst_reg, si->src_reg,
6489                                       offsetof(struct bpf_sock_ops_kern, sk));
6490                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6491                                       offsetof(struct sock_common, skc_dport));
6492 #ifndef __BIG_ENDIAN_BITFIELD
6493                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6494 #endif
6495                 break;
6496
6497         case offsetof(struct bpf_sock_ops, local_port):
6498                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6499
6500                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6501                                                 struct bpf_sock_ops_kern, sk),
6502                                       si->dst_reg, si->src_reg,
6503                                       offsetof(struct bpf_sock_ops_kern, sk));
6504                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6505                                       offsetof(struct sock_common, skc_num));
6506                 break;
6507
6508         case offsetof(struct bpf_sock_ops, is_fullsock):
6509                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6510                                                 struct bpf_sock_ops_kern,
6511                                                 is_fullsock),
6512                                       si->dst_reg, si->src_reg,
6513                                       offsetof(struct bpf_sock_ops_kern,
6514                                                is_fullsock));
6515                 break;
6516
6517         case offsetof(struct bpf_sock_ops, state):
6518                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6519
6520                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6521                                                 struct bpf_sock_ops_kern, sk),
6522                                       si->dst_reg, si->src_reg,
6523                                       offsetof(struct bpf_sock_ops_kern, sk));
6524                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6525                                       offsetof(struct sock_common, skc_state));
6526                 break;
6527
6528         case offsetof(struct bpf_sock_ops, rtt_min):
6529                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6530                              sizeof(struct minmax));
6531                 BUILD_BUG_ON(sizeof(struct minmax) <
6532                              sizeof(struct minmax_sample));
6533
6534                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6535                                                 struct bpf_sock_ops_kern, sk),
6536                                       si->dst_reg, si->src_reg,
6537                                       offsetof(struct bpf_sock_ops_kern, sk));
6538                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6539                                       offsetof(struct tcp_sock, rtt_min) +
6540                                       FIELD_SIZEOF(struct minmax_sample, t));
6541                 break;
6542
6543 /* Helper macro for adding read access to tcp_sock or sock fields. */
6544 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6545         do {                                                                  \
6546                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6547                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6548                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6549                                                 struct bpf_sock_ops_kern,     \
6550                                                 is_fullsock),                 \
6551                                       si->dst_reg, si->src_reg,               \
6552                                       offsetof(struct bpf_sock_ops_kern,      \
6553                                                is_fullsock));                 \
6554                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
6555                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6556                                                 struct bpf_sock_ops_kern, sk),\
6557                                       si->dst_reg, si->src_reg,               \
6558                                       offsetof(struct bpf_sock_ops_kern, sk));\
6559                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
6560                                                        OBJ_FIELD),            \
6561                                       si->dst_reg, si->dst_reg,               \
6562                                       offsetof(OBJ, OBJ_FIELD));              \
6563         } while (0)
6564
6565 /* Helper macro for adding write access to tcp_sock or sock fields.
6566  * The macro is called with two registers, dst_reg which contains a pointer
6567  * to ctx (context) and src_reg which contains the value that should be
6568  * stored. However, we need an additional register since we cannot overwrite
6569  * dst_reg because it may be used later in the program.
6570  * Instead we "borrow" one of the other register. We first save its value
6571  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6572  * it at the end of the macro.
6573  */
6574 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6575         do {                                                                  \
6576                 int reg = BPF_REG_9;                                          \
6577                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6578                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6579                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6580                         reg--;                                                \
6581                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6582                         reg--;                                                \
6583                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
6584                                       offsetof(struct bpf_sock_ops_kern,      \
6585                                                temp));                        \
6586                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6587                                                 struct bpf_sock_ops_kern,     \
6588                                                 is_fullsock),                 \
6589                                       reg, si->dst_reg,                       \
6590                                       offsetof(struct bpf_sock_ops_kern,      \
6591                                                is_fullsock));                 \
6592                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
6593                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6594                                                 struct bpf_sock_ops_kern, sk),\
6595                                       reg, si->dst_reg,                       \
6596                                       offsetof(struct bpf_sock_ops_kern, sk));\
6597                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
6598                                       reg, si->src_reg,                       \
6599                                       offsetof(OBJ, OBJ_FIELD));              \
6600                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
6601                                       offsetof(struct bpf_sock_ops_kern,      \
6602                                                temp));                        \
6603         } while (0)
6604
6605 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
6606         do {                                                                  \
6607                 if (TYPE == BPF_WRITE)                                        \
6608                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6609                 else                                                          \
6610                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6611         } while (0)
6612
6613         case offsetof(struct bpf_sock_ops, snd_cwnd):
6614                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6615                 break;
6616
6617         case offsetof(struct bpf_sock_ops, srtt_us):
6618                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6619                 break;
6620
6621         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6622                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6623                                    struct tcp_sock);
6624                 break;
6625
6626         case offsetof(struct bpf_sock_ops, snd_ssthresh):
6627                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6628                 break;
6629
6630         case offsetof(struct bpf_sock_ops, rcv_nxt):
6631                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6632                 break;
6633
6634         case offsetof(struct bpf_sock_ops, snd_nxt):
6635                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6636                 break;
6637
6638         case offsetof(struct bpf_sock_ops, snd_una):
6639                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6640                 break;
6641
6642         case offsetof(struct bpf_sock_ops, mss_cache):
6643                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6644                 break;
6645
6646         case offsetof(struct bpf_sock_ops, ecn_flags):
6647                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6648                 break;
6649
6650         case offsetof(struct bpf_sock_ops, rate_delivered):
6651                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6652                                    struct tcp_sock);
6653                 break;
6654
6655         case offsetof(struct bpf_sock_ops, rate_interval_us):
6656                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6657                                    struct tcp_sock);
6658                 break;
6659
6660         case offsetof(struct bpf_sock_ops, packets_out):
6661                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6662                 break;
6663
6664         case offsetof(struct bpf_sock_ops, retrans_out):
6665                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6666                 break;
6667
6668         case offsetof(struct bpf_sock_ops, total_retrans):
6669                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6670                                    struct tcp_sock);
6671                 break;
6672
6673         case offsetof(struct bpf_sock_ops, segs_in):
6674                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6675                 break;
6676
6677         case offsetof(struct bpf_sock_ops, data_segs_in):
6678                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6679                 break;
6680
6681         case offsetof(struct bpf_sock_ops, segs_out):
6682                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6683                 break;
6684
6685         case offsetof(struct bpf_sock_ops, data_segs_out):
6686                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6687                                    struct tcp_sock);
6688                 break;
6689
6690         case offsetof(struct bpf_sock_ops, lost_out):
6691                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6692                 break;
6693
6694         case offsetof(struct bpf_sock_ops, sacked_out):
6695                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6696                 break;
6697
6698         case offsetof(struct bpf_sock_ops, sk_txhash):
6699                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6700                                           struct sock, type);
6701                 break;
6702
6703         case offsetof(struct bpf_sock_ops, bytes_received):
6704                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6705                                    struct tcp_sock);
6706                 break;
6707
6708         case offsetof(struct bpf_sock_ops, bytes_acked):
6709                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6710                 break;
6711
6712         }
6713         return insn - insn_buf;
6714 }
6715
6716 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6717                                      const struct bpf_insn *si,
6718                                      struct bpf_insn *insn_buf,
6719                                      struct bpf_prog *prog, u32 *target_size)
6720 {
6721         struct bpf_insn *insn = insn_buf;
6722         int off;
6723
6724         switch (si->off) {
6725         case offsetof(struct __sk_buff, data_end):
6726                 off  = si->off;
6727                 off -= offsetof(struct __sk_buff, data_end);
6728                 off += offsetof(struct sk_buff, cb);
6729                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6730                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6731                                       si->src_reg, off);
6732                 break;
6733         default:
6734                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6735                                               target_size);
6736         }
6737
6738         return insn - insn_buf;
6739 }
6740
6741 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6742                                      const struct bpf_insn *si,
6743                                      struct bpf_insn *insn_buf,
6744                                      struct bpf_prog *prog, u32 *target_size)
6745 {
6746         struct bpf_insn *insn = insn_buf;
6747 #if IS_ENABLED(CONFIG_IPV6)
6748         int off;
6749 #endif
6750
6751         switch (si->off) {
6752         case offsetof(struct sk_msg_md, data):
6753                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data),
6754                                       si->dst_reg, si->src_reg,
6755                                       offsetof(struct sk_msg_buff, data));
6756                 break;
6757         case offsetof(struct sk_msg_md, data_end):
6758                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data_end),
6759                                       si->dst_reg, si->src_reg,
6760                                       offsetof(struct sk_msg_buff, data_end));
6761                 break;
6762         case offsetof(struct sk_msg_md, family):
6763                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6764
6765                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6766                                               struct sk_msg_buff, sk),
6767                                       si->dst_reg, si->src_reg,
6768                                       offsetof(struct sk_msg_buff, sk));
6769                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6770                                       offsetof(struct sock_common, skc_family));
6771                 break;
6772
6773         case offsetof(struct sk_msg_md, remote_ip4):
6774                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6775
6776                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6777                                                 struct sk_msg_buff, sk),
6778                                       si->dst_reg, si->src_reg,
6779                                       offsetof(struct sk_msg_buff, sk));
6780                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6781                                       offsetof(struct sock_common, skc_daddr));
6782                 break;
6783
6784         case offsetof(struct sk_msg_md, local_ip4):
6785                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6786                                           skc_rcv_saddr) != 4);
6787
6788                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6789                                               struct sk_msg_buff, sk),
6790                                       si->dst_reg, si->src_reg,
6791                                       offsetof(struct sk_msg_buff, sk));
6792                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6793                                       offsetof(struct sock_common,
6794                                                skc_rcv_saddr));
6795                 break;
6796
6797         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6798              offsetof(struct sk_msg_md, remote_ip6[3]):
6799 #if IS_ENABLED(CONFIG_IPV6)
6800                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6801                                           skc_v6_daddr.s6_addr32[0]) != 4);
6802
6803                 off = si->off;
6804                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6805                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6806                                                 struct sk_msg_buff, sk),
6807                                       si->dst_reg, si->src_reg,
6808                                       offsetof(struct sk_msg_buff, sk));
6809                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6810                                       offsetof(struct sock_common,
6811                                                skc_v6_daddr.s6_addr32[0]) +
6812                                       off);
6813 #else
6814                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6815 #endif
6816                 break;
6817
6818         case offsetof(struct sk_msg_md, local_ip6[0]) ...
6819              offsetof(struct sk_msg_md, local_ip6[3]):
6820 #if IS_ENABLED(CONFIG_IPV6)
6821                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6822                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6823
6824                 off = si->off;
6825                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6826                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6827                                                 struct sk_msg_buff, sk),
6828                                       si->dst_reg, si->src_reg,
6829                                       offsetof(struct sk_msg_buff, sk));
6830                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6831                                       offsetof(struct sock_common,
6832                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6833                                       off);
6834 #else
6835                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6836 #endif
6837                 break;
6838
6839         case offsetof(struct sk_msg_md, remote_port):
6840                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6841
6842                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6843                                                 struct sk_msg_buff, sk),
6844                                       si->dst_reg, si->src_reg,
6845                                       offsetof(struct sk_msg_buff, sk));
6846                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6847                                       offsetof(struct sock_common, skc_dport));
6848 #ifndef __BIG_ENDIAN_BITFIELD
6849                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6850 #endif
6851                 break;
6852
6853         case offsetof(struct sk_msg_md, local_port):
6854                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6855
6856                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6857                                                 struct sk_msg_buff, sk),
6858                                       si->dst_reg, si->src_reg,
6859                                       offsetof(struct sk_msg_buff, sk));
6860                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6861                                       offsetof(struct sock_common, skc_num));
6862                 break;
6863         }
6864
6865         return insn - insn_buf;
6866 }
6867
6868 const struct bpf_verifier_ops sk_filter_verifier_ops = {
6869         .get_func_proto         = sk_filter_func_proto,
6870         .is_valid_access        = sk_filter_is_valid_access,
6871         .convert_ctx_access     = bpf_convert_ctx_access,
6872         .gen_ld_abs             = bpf_gen_ld_abs,
6873 };
6874
6875 const struct bpf_prog_ops sk_filter_prog_ops = {
6876         .test_run               = bpf_prog_test_run_skb,
6877 };
6878
6879 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
6880         .get_func_proto         = tc_cls_act_func_proto,
6881         .is_valid_access        = tc_cls_act_is_valid_access,
6882         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
6883         .gen_prologue           = tc_cls_act_prologue,
6884         .gen_ld_abs             = bpf_gen_ld_abs,
6885 };
6886
6887 const struct bpf_prog_ops tc_cls_act_prog_ops = {
6888         .test_run               = bpf_prog_test_run_skb,
6889 };
6890
6891 const struct bpf_verifier_ops xdp_verifier_ops = {
6892         .get_func_proto         = xdp_func_proto,
6893         .is_valid_access        = xdp_is_valid_access,
6894         .convert_ctx_access     = xdp_convert_ctx_access,
6895 };
6896
6897 const struct bpf_prog_ops xdp_prog_ops = {
6898         .test_run               = bpf_prog_test_run_xdp,
6899 };
6900
6901 const struct bpf_verifier_ops cg_skb_verifier_ops = {
6902         .get_func_proto         = cg_skb_func_proto,
6903         .is_valid_access        = sk_filter_is_valid_access,
6904         .convert_ctx_access     = bpf_convert_ctx_access,
6905 };
6906
6907 const struct bpf_prog_ops cg_skb_prog_ops = {
6908         .test_run               = bpf_prog_test_run_skb,
6909 };
6910
6911 const struct bpf_verifier_ops lwt_in_verifier_ops = {
6912         .get_func_proto         = lwt_in_func_proto,
6913         .is_valid_access        = lwt_is_valid_access,
6914         .convert_ctx_access     = bpf_convert_ctx_access,
6915 };
6916
6917 const struct bpf_prog_ops lwt_in_prog_ops = {
6918         .test_run               = bpf_prog_test_run_skb,
6919 };
6920
6921 const struct bpf_verifier_ops lwt_out_verifier_ops = {
6922         .get_func_proto         = lwt_out_func_proto,
6923         .is_valid_access        = lwt_is_valid_access,
6924         .convert_ctx_access     = bpf_convert_ctx_access,
6925 };
6926
6927 const struct bpf_prog_ops lwt_out_prog_ops = {
6928         .test_run               = bpf_prog_test_run_skb,
6929 };
6930
6931 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
6932         .get_func_proto         = lwt_xmit_func_proto,
6933         .is_valid_access        = lwt_is_valid_access,
6934         .convert_ctx_access     = bpf_convert_ctx_access,
6935         .gen_prologue           = tc_cls_act_prologue,
6936 };
6937
6938 const struct bpf_prog_ops lwt_xmit_prog_ops = {
6939         .test_run               = bpf_prog_test_run_skb,
6940 };
6941
6942 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
6943         .get_func_proto         = lwt_seg6local_func_proto,
6944         .is_valid_access        = lwt_is_valid_access,
6945         .convert_ctx_access     = bpf_convert_ctx_access,
6946 };
6947
6948 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
6949         .test_run               = bpf_prog_test_run_skb,
6950 };
6951
6952 const struct bpf_verifier_ops cg_sock_verifier_ops = {
6953         .get_func_proto         = sock_filter_func_proto,
6954         .is_valid_access        = sock_filter_is_valid_access,
6955         .convert_ctx_access     = sock_filter_convert_ctx_access,
6956 };
6957
6958 const struct bpf_prog_ops cg_sock_prog_ops = {
6959 };
6960
6961 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
6962         .get_func_proto         = sock_addr_func_proto,
6963         .is_valid_access        = sock_addr_is_valid_access,
6964         .convert_ctx_access     = sock_addr_convert_ctx_access,
6965 };
6966
6967 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
6968 };
6969
6970 const struct bpf_verifier_ops sock_ops_verifier_ops = {
6971         .get_func_proto         = sock_ops_func_proto,
6972         .is_valid_access        = sock_ops_is_valid_access,
6973         .convert_ctx_access     = sock_ops_convert_ctx_access,
6974 };
6975
6976 const struct bpf_prog_ops sock_ops_prog_ops = {
6977 };
6978
6979 const struct bpf_verifier_ops sk_skb_verifier_ops = {
6980         .get_func_proto         = sk_skb_func_proto,
6981         .is_valid_access        = sk_skb_is_valid_access,
6982         .convert_ctx_access     = sk_skb_convert_ctx_access,
6983         .gen_prologue           = sk_skb_prologue,
6984 };
6985
6986 const struct bpf_prog_ops sk_skb_prog_ops = {
6987 };
6988
6989 const struct bpf_verifier_ops sk_msg_verifier_ops = {
6990         .get_func_proto         = sk_msg_func_proto,
6991         .is_valid_access        = sk_msg_is_valid_access,
6992         .convert_ctx_access     = sk_msg_convert_ctx_access,
6993 };
6994
6995 const struct bpf_prog_ops sk_msg_prog_ops = {
6996 };
6997
6998 int sk_detach_filter(struct sock *sk)
6999 {
7000         int ret = -ENOENT;
7001         struct sk_filter *filter;
7002
7003         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7004                 return -EPERM;
7005
7006         filter = rcu_dereference_protected(sk->sk_filter,
7007                                            lockdep_sock_is_held(sk));
7008         if (filter) {
7009                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7010                 sk_filter_uncharge(sk, filter);
7011                 ret = 0;
7012         }
7013
7014         return ret;
7015 }
7016 EXPORT_SYMBOL_GPL(sk_detach_filter);
7017
7018 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7019                   unsigned int len)
7020 {
7021         struct sock_fprog_kern *fprog;
7022         struct sk_filter *filter;
7023         int ret = 0;
7024
7025         lock_sock(sk);
7026         filter = rcu_dereference_protected(sk->sk_filter,
7027                                            lockdep_sock_is_held(sk));
7028         if (!filter)
7029                 goto out;
7030
7031         /* We're copying the filter that has been originally attached,
7032          * so no conversion/decode needed anymore. eBPF programs that
7033          * have no original program cannot be dumped through this.
7034          */
7035         ret = -EACCES;
7036         fprog = filter->prog->orig_prog;
7037         if (!fprog)
7038                 goto out;
7039
7040         ret = fprog->len;
7041         if (!len)
7042                 /* User space only enquires number of filter blocks. */
7043                 goto out;
7044
7045         ret = -EINVAL;
7046         if (len < fprog->len)
7047                 goto out;
7048
7049         ret = -EFAULT;
7050         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7051                 goto out;
7052
7053         /* Instead of bytes, the API requests to return the number
7054          * of filter blocks.
7055          */
7056         ret = fprog->len;
7057 out:
7058         release_sock(sk);
7059         return ret;
7060 }
7061
7062 #ifdef CONFIG_INET
7063 struct sk_reuseport_kern {
7064         struct sk_buff *skb;
7065         struct sock *sk;
7066         struct sock *selected_sk;
7067         void *data_end;
7068         u32 hash;
7069         u32 reuseport_id;
7070         bool bind_inany;
7071 };
7072
7073 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7074                                     struct sock_reuseport *reuse,
7075                                     struct sock *sk, struct sk_buff *skb,
7076                                     u32 hash)
7077 {
7078         reuse_kern->skb = skb;
7079         reuse_kern->sk = sk;
7080         reuse_kern->selected_sk = NULL;
7081         reuse_kern->data_end = skb->data + skb_headlen(skb);
7082         reuse_kern->hash = hash;
7083         reuse_kern->reuseport_id = reuse->reuseport_id;
7084         reuse_kern->bind_inany = reuse->bind_inany;
7085 }
7086
7087 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7088                                   struct bpf_prog *prog, struct sk_buff *skb,
7089                                   u32 hash)
7090 {
7091         struct sk_reuseport_kern reuse_kern;
7092         enum sk_action action;
7093
7094         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7095         action = BPF_PROG_RUN(prog, &reuse_kern);
7096
7097         if (action == SK_PASS)
7098                 return reuse_kern.selected_sk;
7099         else
7100                 return ERR_PTR(-ECONNREFUSED);
7101 }
7102
7103 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7104            struct bpf_map *, map, void *, key, u32, flags)
7105 {
7106         struct sock_reuseport *reuse;
7107         struct sock *selected_sk;
7108
7109         selected_sk = map->ops->map_lookup_elem(map, key);
7110         if (!selected_sk)
7111                 return -ENOENT;
7112
7113         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7114         if (!reuse)
7115                 /* selected_sk is unhashed (e.g. by close()) after the
7116                  * above map_lookup_elem().  Treat selected_sk has already
7117                  * been removed from the map.
7118                  */
7119                 return -ENOENT;
7120
7121         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7122                 struct sock *sk;
7123
7124                 if (unlikely(!reuse_kern->reuseport_id))
7125                         /* There is a small race between adding the
7126                          * sk to the map and setting the
7127                          * reuse_kern->reuseport_id.
7128                          * Treat it as the sk has not been added to
7129                          * the bpf map yet.
7130                          */
7131                         return -ENOENT;
7132
7133                 sk = reuse_kern->sk;
7134                 if (sk->sk_protocol != selected_sk->sk_protocol)
7135                         return -EPROTOTYPE;
7136                 else if (sk->sk_family != selected_sk->sk_family)
7137                         return -EAFNOSUPPORT;
7138
7139                 /* Catch all. Likely bound to a different sockaddr. */
7140                 return -EBADFD;
7141         }
7142
7143         reuse_kern->selected_sk = selected_sk;
7144
7145         return 0;
7146 }
7147
7148 static const struct bpf_func_proto sk_select_reuseport_proto = {
7149         .func           = sk_select_reuseport,
7150         .gpl_only       = false,
7151         .ret_type       = RET_INTEGER,
7152         .arg1_type      = ARG_PTR_TO_CTX,
7153         .arg2_type      = ARG_CONST_MAP_PTR,
7154         .arg3_type      = ARG_PTR_TO_MAP_KEY,
7155         .arg4_type      = ARG_ANYTHING,
7156 };
7157
7158 BPF_CALL_4(sk_reuseport_load_bytes,
7159            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7160            void *, to, u32, len)
7161 {
7162         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7163 }
7164
7165 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7166         .func           = sk_reuseport_load_bytes,
7167         .gpl_only       = false,
7168         .ret_type       = RET_INTEGER,
7169         .arg1_type      = ARG_PTR_TO_CTX,
7170         .arg2_type      = ARG_ANYTHING,
7171         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7172         .arg4_type      = ARG_CONST_SIZE,
7173 };
7174
7175 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7176            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7177            void *, to, u32, len, u32, start_header)
7178 {
7179         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7180                                                len, start_header);
7181 }
7182
7183 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7184         .func           = sk_reuseport_load_bytes_relative,
7185         .gpl_only       = false,
7186         .ret_type       = RET_INTEGER,
7187         .arg1_type      = ARG_PTR_TO_CTX,
7188         .arg2_type      = ARG_ANYTHING,
7189         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7190         .arg4_type      = ARG_CONST_SIZE,
7191         .arg5_type      = ARG_ANYTHING,
7192 };
7193
7194 static const struct bpf_func_proto *
7195 sk_reuseport_func_proto(enum bpf_func_id func_id,
7196                         const struct bpf_prog *prog)
7197 {
7198         switch (func_id) {
7199         case BPF_FUNC_sk_select_reuseport:
7200                 return &sk_select_reuseport_proto;
7201         case BPF_FUNC_skb_load_bytes:
7202                 return &sk_reuseport_load_bytes_proto;
7203         case BPF_FUNC_skb_load_bytes_relative:
7204                 return &sk_reuseport_load_bytes_relative_proto;
7205         default:
7206                 return bpf_base_func_proto(func_id);
7207         }
7208 }
7209
7210 static bool
7211 sk_reuseport_is_valid_access(int off, int size,
7212                              enum bpf_access_type type,
7213                              const struct bpf_prog *prog,
7214                              struct bpf_insn_access_aux *info)
7215 {
7216         const u32 size_default = sizeof(__u32);
7217
7218         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7219             off % size || type != BPF_READ)
7220                 return false;
7221
7222         switch (off) {
7223         case offsetof(struct sk_reuseport_md, data):
7224                 info->reg_type = PTR_TO_PACKET;
7225                 return size == sizeof(__u64);
7226
7227         case offsetof(struct sk_reuseport_md, data_end):
7228                 info->reg_type = PTR_TO_PACKET_END;
7229                 return size == sizeof(__u64);
7230
7231         case offsetof(struct sk_reuseport_md, hash):
7232                 return size == size_default;
7233
7234         /* Fields that allow narrowing */
7235         case offsetof(struct sk_reuseport_md, eth_protocol):
7236                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7237                         return false;
7238         case offsetof(struct sk_reuseport_md, ip_protocol):
7239         case offsetof(struct sk_reuseport_md, bind_inany):
7240         case offsetof(struct sk_reuseport_md, len):
7241                 bpf_ctx_record_field_size(info, size_default);
7242                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7243
7244         default:
7245                 return false;
7246         }
7247 }
7248
7249 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
7250         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7251                               si->dst_reg, si->src_reg,                 \
7252                               bpf_target_off(struct sk_reuseport_kern, F, \
7253                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7254                                              target_size));             \
7255         })
7256
7257 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
7258         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
7259                                     struct sk_buff,                     \
7260                                     skb,                                \
7261                                     SKB_FIELD)
7262
7263 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7264         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
7265                                              struct sock,               \
7266                                              sk,                        \
7267                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
7268
7269 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7270                                            const struct bpf_insn *si,
7271                                            struct bpf_insn *insn_buf,
7272                                            struct bpf_prog *prog,
7273                                            u32 *target_size)
7274 {
7275         struct bpf_insn *insn = insn_buf;
7276
7277         switch (si->off) {
7278         case offsetof(struct sk_reuseport_md, data):
7279                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7280                 break;
7281
7282         case offsetof(struct sk_reuseport_md, len):
7283                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7284                 break;
7285
7286         case offsetof(struct sk_reuseport_md, eth_protocol):
7287                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7288                 break;
7289
7290         case offsetof(struct sk_reuseport_md, ip_protocol):
7291                 BUILD_BUG_ON(hweight_long(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7292                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7293                                                     BPF_W, 0);
7294                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7295                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7296                                         SK_FL_PROTO_SHIFT);
7297                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
7298                  * aware.  No further narrowing or masking is needed.
7299                  */
7300                 *target_size = 1;
7301                 break;
7302
7303         case offsetof(struct sk_reuseport_md, data_end):
7304                 SK_REUSEPORT_LOAD_FIELD(data_end);
7305                 break;
7306
7307         case offsetof(struct sk_reuseport_md, hash):
7308                 SK_REUSEPORT_LOAD_FIELD(hash);
7309                 break;
7310
7311         case offsetof(struct sk_reuseport_md, bind_inany):
7312                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
7313                 break;
7314         }
7315
7316         return insn - insn_buf;
7317 }
7318
7319 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
7320         .get_func_proto         = sk_reuseport_func_proto,
7321         .is_valid_access        = sk_reuseport_is_valid_access,
7322         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
7323 };
7324
7325 const struct bpf_prog_ops sk_reuseport_prog_ops = {
7326 };
7327 #endif /* CONFIG_INET */