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 #define sk_msg_iter_var(var)                    \
2286         do {                                    \
2287                 var++;                          \
2288                 if (var == MAX_SKB_FRAGS)       \
2289                         var = 0;                \
2290         } while (0)
2291
2292 BPF_CALL_4(bpf_msg_pull_data,
2293            struct sk_msg_buff *, msg, u32, start, u32, end, u64, flags)
2294 {
2295         unsigned int len = 0, offset = 0, copy = 0, poffset = 0;
2296         int bytes = end - start, bytes_sg_total;
2297         struct scatterlist *sg = msg->sg_data;
2298         int first_sg, last_sg, i, shift;
2299         unsigned char *p, *to, *from;
2300         struct page *page;
2301
2302         if (unlikely(flags || end <= start))
2303                 return -EINVAL;
2304
2305         /* First find the starting scatterlist element */
2306         i = msg->sg_start;
2307         do {
2308                 len = sg[i].length;
2309                 if (start < offset + len)
2310                         break;
2311                 offset += len;
2312                 sk_msg_iter_var(i);
2313         } while (i != msg->sg_end);
2314
2315         if (unlikely(start >= offset + len))
2316                 return -EINVAL;
2317
2318         first_sg = i;
2319         /* The start may point into the sg element so we need to also
2320          * account for the headroom.
2321          */
2322         bytes_sg_total = start - offset + bytes;
2323         if (!msg->sg_copy[i] && bytes_sg_total <= len)
2324                 goto out;
2325
2326         /* At this point we need to linearize multiple scatterlist
2327          * elements or a single shared page. Either way we need to
2328          * copy into a linear buffer exclusively owned by BPF. Then
2329          * place the buffer in the scatterlist and fixup the original
2330          * entries by removing the entries now in the linear buffer
2331          * and shifting the remaining entries. For now we do not try
2332          * to copy partial entries to avoid complexity of running out
2333          * of sg_entry slots. The downside is reading a single byte
2334          * will copy the entire sg entry.
2335          */
2336         do {
2337                 copy += sg[i].length;
2338                 sk_msg_iter_var(i);
2339                 if (bytes_sg_total <= copy)
2340                         break;
2341         } while (i != msg->sg_end);
2342         last_sg = i;
2343
2344         if (unlikely(bytes_sg_total > copy))
2345                 return -EINVAL;
2346
2347         page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP,
2348                            get_order(copy));
2349         if (unlikely(!page))
2350                 return -ENOMEM;
2351         p = page_address(page);
2352
2353         i = first_sg;
2354         do {
2355                 from = sg_virt(&sg[i]);
2356                 len = sg[i].length;
2357                 to = p + poffset;
2358
2359                 memcpy(to, from, len);
2360                 poffset += len;
2361                 sg[i].length = 0;
2362                 put_page(sg_page(&sg[i]));
2363
2364                 sk_msg_iter_var(i);
2365         } while (i != last_sg);
2366
2367         sg[first_sg].length = copy;
2368         sg_set_page(&sg[first_sg], page, copy, 0);
2369
2370         /* To repair sg ring we need to shift entries. If we only
2371          * had a single entry though we can just replace it and
2372          * be done. Otherwise walk the ring and shift the entries.
2373          */
2374         WARN_ON_ONCE(last_sg == first_sg);
2375         shift = last_sg > first_sg ?
2376                 last_sg - first_sg - 1 :
2377                 MAX_SKB_FRAGS - first_sg + last_sg - 1;
2378         if (!shift)
2379                 goto out;
2380
2381         i = first_sg;
2382         sk_msg_iter_var(i);
2383         do {
2384                 int move_from;
2385
2386                 if (i + shift >= MAX_SKB_FRAGS)
2387                         move_from = i + shift - MAX_SKB_FRAGS;
2388                 else
2389                         move_from = i + shift;
2390
2391                 if (move_from == msg->sg_end)
2392                         break;
2393
2394                 sg[i] = sg[move_from];
2395                 sg[move_from].length = 0;
2396                 sg[move_from].page_link = 0;
2397                 sg[move_from].offset = 0;
2398
2399                 sk_msg_iter_var(i);
2400         } while (1);
2401         msg->sg_end -= shift;
2402         if (msg->sg_end < 0)
2403                 msg->sg_end += MAX_SKB_FRAGS;
2404 out:
2405         msg->data = sg_virt(&sg[first_sg]) + start - offset;
2406         msg->data_end = msg->data + bytes;
2407
2408         return 0;
2409 }
2410
2411 static const struct bpf_func_proto bpf_msg_pull_data_proto = {
2412         .func           = bpf_msg_pull_data,
2413         .gpl_only       = false,
2414         .ret_type       = RET_INTEGER,
2415         .arg1_type      = ARG_PTR_TO_CTX,
2416         .arg2_type      = ARG_ANYTHING,
2417         .arg3_type      = ARG_ANYTHING,
2418         .arg4_type      = ARG_ANYTHING,
2419 };
2420
2421 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb)
2422 {
2423         return task_get_classid(skb);
2424 }
2425
2426 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = {
2427         .func           = bpf_get_cgroup_classid,
2428         .gpl_only       = false,
2429         .ret_type       = RET_INTEGER,
2430         .arg1_type      = ARG_PTR_TO_CTX,
2431 };
2432
2433 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb)
2434 {
2435         return dst_tclassid(skb);
2436 }
2437
2438 static const struct bpf_func_proto bpf_get_route_realm_proto = {
2439         .func           = bpf_get_route_realm,
2440         .gpl_only       = false,
2441         .ret_type       = RET_INTEGER,
2442         .arg1_type      = ARG_PTR_TO_CTX,
2443 };
2444
2445 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb)
2446 {
2447         /* If skb_clear_hash() was called due to mangling, we can
2448          * trigger SW recalculation here. Later access to hash
2449          * can then use the inline skb->hash via context directly
2450          * instead of calling this helper again.
2451          */
2452         return skb_get_hash(skb);
2453 }
2454
2455 static const struct bpf_func_proto bpf_get_hash_recalc_proto = {
2456         .func           = bpf_get_hash_recalc,
2457         .gpl_only       = false,
2458         .ret_type       = RET_INTEGER,
2459         .arg1_type      = ARG_PTR_TO_CTX,
2460 };
2461
2462 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb)
2463 {
2464         /* After all direct packet write, this can be used once for
2465          * triggering a lazy recalc on next skb_get_hash() invocation.
2466          */
2467         skb_clear_hash(skb);
2468         return 0;
2469 }
2470
2471 static const struct bpf_func_proto bpf_set_hash_invalid_proto = {
2472         .func           = bpf_set_hash_invalid,
2473         .gpl_only       = false,
2474         .ret_type       = RET_INTEGER,
2475         .arg1_type      = ARG_PTR_TO_CTX,
2476 };
2477
2478 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash)
2479 {
2480         /* Set user specified hash as L4(+), so that it gets returned
2481          * on skb_get_hash() call unless BPF prog later on triggers a
2482          * skb_clear_hash().
2483          */
2484         __skb_set_sw_hash(skb, hash, true);
2485         return 0;
2486 }
2487
2488 static const struct bpf_func_proto bpf_set_hash_proto = {
2489         .func           = bpf_set_hash,
2490         .gpl_only       = false,
2491         .ret_type       = RET_INTEGER,
2492         .arg1_type      = ARG_PTR_TO_CTX,
2493         .arg2_type      = ARG_ANYTHING,
2494 };
2495
2496 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto,
2497            u16, vlan_tci)
2498 {
2499         int ret;
2500
2501         if (unlikely(vlan_proto != htons(ETH_P_8021Q) &&
2502                      vlan_proto != htons(ETH_P_8021AD)))
2503                 vlan_proto = htons(ETH_P_8021Q);
2504
2505         bpf_push_mac_rcsum(skb);
2506         ret = skb_vlan_push(skb, vlan_proto, vlan_tci);
2507         bpf_pull_mac_rcsum(skb);
2508
2509         bpf_compute_data_pointers(skb);
2510         return ret;
2511 }
2512
2513 static const struct bpf_func_proto bpf_skb_vlan_push_proto = {
2514         .func           = bpf_skb_vlan_push,
2515         .gpl_only       = false,
2516         .ret_type       = RET_INTEGER,
2517         .arg1_type      = ARG_PTR_TO_CTX,
2518         .arg2_type      = ARG_ANYTHING,
2519         .arg3_type      = ARG_ANYTHING,
2520 };
2521
2522 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb)
2523 {
2524         int ret;
2525
2526         bpf_push_mac_rcsum(skb);
2527         ret = skb_vlan_pop(skb);
2528         bpf_pull_mac_rcsum(skb);
2529
2530         bpf_compute_data_pointers(skb);
2531         return ret;
2532 }
2533
2534 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = {
2535         .func           = bpf_skb_vlan_pop,
2536         .gpl_only       = false,
2537         .ret_type       = RET_INTEGER,
2538         .arg1_type      = ARG_PTR_TO_CTX,
2539 };
2540
2541 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len)
2542 {
2543         /* Caller already did skb_cow() with len as headroom,
2544          * so no need to do it here.
2545          */
2546         skb_push(skb, len);
2547         memmove(skb->data, skb->data + len, off);
2548         memset(skb->data + off, 0, len);
2549
2550         /* No skb_postpush_rcsum(skb, skb->data + off, len)
2551          * needed here as it does not change the skb->csum
2552          * result for checksum complete when summing over
2553          * zeroed blocks.
2554          */
2555         return 0;
2556 }
2557
2558 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len)
2559 {
2560         /* skb_ensure_writable() is not needed here, as we're
2561          * already working on an uncloned skb.
2562          */
2563         if (unlikely(!pskb_may_pull(skb, off + len)))
2564                 return -ENOMEM;
2565
2566         skb_postpull_rcsum(skb, skb->data + off, len);
2567         memmove(skb->data + len, skb->data, off);
2568         __skb_pull(skb, len);
2569
2570         return 0;
2571 }
2572
2573 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len)
2574 {
2575         bool trans_same = skb->transport_header == skb->network_header;
2576         int ret;
2577
2578         /* There's no need for __skb_push()/__skb_pull() pair to
2579          * get to the start of the mac header as we're guaranteed
2580          * to always start from here under eBPF.
2581          */
2582         ret = bpf_skb_generic_push(skb, off, len);
2583         if (likely(!ret)) {
2584                 skb->mac_header -= len;
2585                 skb->network_header -= len;
2586                 if (trans_same)
2587                         skb->transport_header = skb->network_header;
2588         }
2589
2590         return ret;
2591 }
2592
2593 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len)
2594 {
2595         bool trans_same = skb->transport_header == skb->network_header;
2596         int ret;
2597
2598         /* Same here, __skb_push()/__skb_pull() pair not needed. */
2599         ret = bpf_skb_generic_pop(skb, off, len);
2600         if (likely(!ret)) {
2601                 skb->mac_header += len;
2602                 skb->network_header += len;
2603                 if (trans_same)
2604                         skb->transport_header = skb->network_header;
2605         }
2606
2607         return ret;
2608 }
2609
2610 static int bpf_skb_proto_4_to_6(struct sk_buff *skb)
2611 {
2612         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2613         u32 off = skb_mac_header_len(skb);
2614         int ret;
2615
2616         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2617         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2618                 return -ENOTSUPP;
2619
2620         ret = skb_cow(skb, len_diff);
2621         if (unlikely(ret < 0))
2622                 return ret;
2623
2624         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2625         if (unlikely(ret < 0))
2626                 return ret;
2627
2628         if (skb_is_gso(skb)) {
2629                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2630
2631                 /* SKB_GSO_TCPV4 needs to be changed into
2632                  * SKB_GSO_TCPV6.
2633                  */
2634                 if (shinfo->gso_type & SKB_GSO_TCPV4) {
2635                         shinfo->gso_type &= ~SKB_GSO_TCPV4;
2636                         shinfo->gso_type |=  SKB_GSO_TCPV6;
2637                 }
2638
2639                 /* Due to IPv6 header, MSS needs to be downgraded. */
2640                 skb_decrease_gso_size(shinfo, len_diff);
2641                 /* Header must be checked, and gso_segs recomputed. */
2642                 shinfo->gso_type |= SKB_GSO_DODGY;
2643                 shinfo->gso_segs = 0;
2644         }
2645
2646         skb->protocol = htons(ETH_P_IPV6);
2647         skb_clear_hash(skb);
2648
2649         return 0;
2650 }
2651
2652 static int bpf_skb_proto_6_to_4(struct sk_buff *skb)
2653 {
2654         const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr);
2655         u32 off = skb_mac_header_len(skb);
2656         int ret;
2657
2658         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2659         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2660                 return -ENOTSUPP;
2661
2662         ret = skb_unclone(skb, GFP_ATOMIC);
2663         if (unlikely(ret < 0))
2664                 return ret;
2665
2666         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2667         if (unlikely(ret < 0))
2668                 return ret;
2669
2670         if (skb_is_gso(skb)) {
2671                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2672
2673                 /* SKB_GSO_TCPV6 needs to be changed into
2674                  * SKB_GSO_TCPV4.
2675                  */
2676                 if (shinfo->gso_type & SKB_GSO_TCPV6) {
2677                         shinfo->gso_type &= ~SKB_GSO_TCPV6;
2678                         shinfo->gso_type |=  SKB_GSO_TCPV4;
2679                 }
2680
2681                 /* Due to IPv4 header, MSS can be upgraded. */
2682                 skb_increase_gso_size(shinfo, len_diff);
2683                 /* Header must be checked, and gso_segs recomputed. */
2684                 shinfo->gso_type |= SKB_GSO_DODGY;
2685                 shinfo->gso_segs = 0;
2686         }
2687
2688         skb->protocol = htons(ETH_P_IP);
2689         skb_clear_hash(skb);
2690
2691         return 0;
2692 }
2693
2694 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto)
2695 {
2696         __be16 from_proto = skb->protocol;
2697
2698         if (from_proto == htons(ETH_P_IP) &&
2699               to_proto == htons(ETH_P_IPV6))
2700                 return bpf_skb_proto_4_to_6(skb);
2701
2702         if (from_proto == htons(ETH_P_IPV6) &&
2703               to_proto == htons(ETH_P_IP))
2704                 return bpf_skb_proto_6_to_4(skb);
2705
2706         return -ENOTSUPP;
2707 }
2708
2709 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto,
2710            u64, flags)
2711 {
2712         int ret;
2713
2714         if (unlikely(flags))
2715                 return -EINVAL;
2716
2717         /* General idea is that this helper does the basic groundwork
2718          * needed for changing the protocol, and eBPF program fills the
2719          * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace()
2720          * and other helpers, rather than passing a raw buffer here.
2721          *
2722          * The rationale is to keep this minimal and without a need to
2723          * deal with raw packet data. F.e. even if we would pass buffers
2724          * here, the program still needs to call the bpf_lX_csum_replace()
2725          * helpers anyway. Plus, this way we keep also separation of
2726          * concerns, since f.e. bpf_skb_store_bytes() should only take
2727          * care of stores.
2728          *
2729          * Currently, additional options and extension header space are
2730          * not supported, but flags register is reserved so we can adapt
2731          * that. For offloads, we mark packet as dodgy, so that headers
2732          * need to be verified first.
2733          */
2734         ret = bpf_skb_proto_xlat(skb, proto);
2735         bpf_compute_data_pointers(skb);
2736         return ret;
2737 }
2738
2739 static const struct bpf_func_proto bpf_skb_change_proto_proto = {
2740         .func           = bpf_skb_change_proto,
2741         .gpl_only       = false,
2742         .ret_type       = RET_INTEGER,
2743         .arg1_type      = ARG_PTR_TO_CTX,
2744         .arg2_type      = ARG_ANYTHING,
2745         .arg3_type      = ARG_ANYTHING,
2746 };
2747
2748 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type)
2749 {
2750         /* We only allow a restricted subset to be changed for now. */
2751         if (unlikely(!skb_pkt_type_ok(skb->pkt_type) ||
2752                      !skb_pkt_type_ok(pkt_type)))
2753                 return -EINVAL;
2754
2755         skb->pkt_type = pkt_type;
2756         return 0;
2757 }
2758
2759 static const struct bpf_func_proto bpf_skb_change_type_proto = {
2760         .func           = bpf_skb_change_type,
2761         .gpl_only       = false,
2762         .ret_type       = RET_INTEGER,
2763         .arg1_type      = ARG_PTR_TO_CTX,
2764         .arg2_type      = ARG_ANYTHING,
2765 };
2766
2767 static u32 bpf_skb_net_base_len(const struct sk_buff *skb)
2768 {
2769         switch (skb->protocol) {
2770         case htons(ETH_P_IP):
2771                 return sizeof(struct iphdr);
2772         case htons(ETH_P_IPV6):
2773                 return sizeof(struct ipv6hdr);
2774         default:
2775                 return ~0U;
2776         }
2777 }
2778
2779 static int bpf_skb_net_grow(struct sk_buff *skb, u32 len_diff)
2780 {
2781         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2782         int ret;
2783
2784         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2785         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2786                 return -ENOTSUPP;
2787
2788         ret = skb_cow(skb, len_diff);
2789         if (unlikely(ret < 0))
2790                 return ret;
2791
2792         ret = bpf_skb_net_hdr_push(skb, off, len_diff);
2793         if (unlikely(ret < 0))
2794                 return ret;
2795
2796         if (skb_is_gso(skb)) {
2797                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2798
2799                 /* Due to header grow, MSS needs to be downgraded. */
2800                 skb_decrease_gso_size(shinfo, len_diff);
2801                 /* Header must be checked, and gso_segs recomputed. */
2802                 shinfo->gso_type |= SKB_GSO_DODGY;
2803                 shinfo->gso_segs = 0;
2804         }
2805
2806         return 0;
2807 }
2808
2809 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 len_diff)
2810 {
2811         u32 off = skb_mac_header_len(skb) + bpf_skb_net_base_len(skb);
2812         int ret;
2813
2814         /* SCTP uses GSO_BY_FRAGS, thus cannot adjust it. */
2815         if (skb_is_gso(skb) && unlikely(skb_is_gso_sctp(skb)))
2816                 return -ENOTSUPP;
2817
2818         ret = skb_unclone(skb, GFP_ATOMIC);
2819         if (unlikely(ret < 0))
2820                 return ret;
2821
2822         ret = bpf_skb_net_hdr_pop(skb, off, len_diff);
2823         if (unlikely(ret < 0))
2824                 return ret;
2825
2826         if (skb_is_gso(skb)) {
2827                 struct skb_shared_info *shinfo = skb_shinfo(skb);
2828
2829                 /* Due to header shrink, MSS can be upgraded. */
2830                 skb_increase_gso_size(shinfo, len_diff);
2831                 /* Header must be checked, and gso_segs recomputed. */
2832                 shinfo->gso_type |= SKB_GSO_DODGY;
2833                 shinfo->gso_segs = 0;
2834         }
2835
2836         return 0;
2837 }
2838
2839 static u32 __bpf_skb_max_len(const struct sk_buff *skb)
2840 {
2841         return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len :
2842                           SKB_MAX_ALLOC;
2843 }
2844
2845 static int bpf_skb_adjust_net(struct sk_buff *skb, s32 len_diff)
2846 {
2847         bool trans_same = skb->transport_header == skb->network_header;
2848         u32 len_cur, len_diff_abs = abs(len_diff);
2849         u32 len_min = bpf_skb_net_base_len(skb);
2850         u32 len_max = __bpf_skb_max_len(skb);
2851         __be16 proto = skb->protocol;
2852         bool shrink = len_diff < 0;
2853         int ret;
2854
2855         if (unlikely(len_diff_abs > 0xfffU))
2856                 return -EFAULT;
2857         if (unlikely(proto != htons(ETH_P_IP) &&
2858                      proto != htons(ETH_P_IPV6)))
2859                 return -ENOTSUPP;
2860
2861         len_cur = skb->len - skb_network_offset(skb);
2862         if (skb_transport_header_was_set(skb) && !trans_same)
2863                 len_cur = skb_network_header_len(skb);
2864         if ((shrink && (len_diff_abs >= len_cur ||
2865                         len_cur - len_diff_abs < len_min)) ||
2866             (!shrink && (skb->len + len_diff_abs > len_max &&
2867                          !skb_is_gso(skb))))
2868                 return -ENOTSUPP;
2869
2870         ret = shrink ? bpf_skb_net_shrink(skb, len_diff_abs) :
2871                        bpf_skb_net_grow(skb, len_diff_abs);
2872
2873         bpf_compute_data_pointers(skb);
2874         return ret;
2875 }
2876
2877 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff,
2878            u32, mode, u64, flags)
2879 {
2880         if (unlikely(flags))
2881                 return -EINVAL;
2882         if (likely(mode == BPF_ADJ_ROOM_NET))
2883                 return bpf_skb_adjust_net(skb, len_diff);
2884
2885         return -ENOTSUPP;
2886 }
2887
2888 static const struct bpf_func_proto bpf_skb_adjust_room_proto = {
2889         .func           = bpf_skb_adjust_room,
2890         .gpl_only       = false,
2891         .ret_type       = RET_INTEGER,
2892         .arg1_type      = ARG_PTR_TO_CTX,
2893         .arg2_type      = ARG_ANYTHING,
2894         .arg3_type      = ARG_ANYTHING,
2895         .arg4_type      = ARG_ANYTHING,
2896 };
2897
2898 static u32 __bpf_skb_min_len(const struct sk_buff *skb)
2899 {
2900         u32 min_len = skb_network_offset(skb);
2901
2902         if (skb_transport_header_was_set(skb))
2903                 min_len = skb_transport_offset(skb);
2904         if (skb->ip_summed == CHECKSUM_PARTIAL)
2905                 min_len = skb_checksum_start_offset(skb) +
2906                           skb->csum_offset + sizeof(__sum16);
2907         return min_len;
2908 }
2909
2910 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len)
2911 {
2912         unsigned int old_len = skb->len;
2913         int ret;
2914
2915         ret = __skb_grow_rcsum(skb, new_len);
2916         if (!ret)
2917                 memset(skb->data + old_len, 0, new_len - old_len);
2918         return ret;
2919 }
2920
2921 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len)
2922 {
2923         return __skb_trim_rcsum(skb, new_len);
2924 }
2925
2926 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len,
2927                                         u64 flags)
2928 {
2929         u32 max_len = __bpf_skb_max_len(skb);
2930         u32 min_len = __bpf_skb_min_len(skb);
2931         int ret;
2932
2933         if (unlikely(flags || new_len > max_len || new_len < min_len))
2934                 return -EINVAL;
2935         if (skb->encapsulation)
2936                 return -ENOTSUPP;
2937
2938         /* The basic idea of this helper is that it's performing the
2939          * needed work to either grow or trim an skb, and eBPF program
2940          * rewrites the rest via helpers like bpf_skb_store_bytes(),
2941          * bpf_lX_csum_replace() and others rather than passing a raw
2942          * buffer here. This one is a slow path helper and intended
2943          * for replies with control messages.
2944          *
2945          * Like in bpf_skb_change_proto(), we want to keep this rather
2946          * minimal and without protocol specifics so that we are able
2947          * to separate concerns as in bpf_skb_store_bytes() should only
2948          * be the one responsible for writing buffers.
2949          *
2950          * It's really expected to be a slow path operation here for
2951          * control message replies, so we're implicitly linearizing,
2952          * uncloning and drop offloads from the skb by this.
2953          */
2954         ret = __bpf_try_make_writable(skb, skb->len);
2955         if (!ret) {
2956                 if (new_len > skb->len)
2957                         ret = bpf_skb_grow_rcsum(skb, new_len);
2958                 else if (new_len < skb->len)
2959                         ret = bpf_skb_trim_rcsum(skb, new_len);
2960                 if (!ret && skb_is_gso(skb))
2961                         skb_gso_reset(skb);
2962         }
2963         return ret;
2964 }
2965
2966 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2967            u64, flags)
2968 {
2969         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2970
2971         bpf_compute_data_pointers(skb);
2972         return ret;
2973 }
2974
2975 static const struct bpf_func_proto bpf_skb_change_tail_proto = {
2976         .func           = bpf_skb_change_tail,
2977         .gpl_only       = false,
2978         .ret_type       = RET_INTEGER,
2979         .arg1_type      = ARG_PTR_TO_CTX,
2980         .arg2_type      = ARG_ANYTHING,
2981         .arg3_type      = ARG_ANYTHING,
2982 };
2983
2984 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len,
2985            u64, flags)
2986 {
2987         int ret = __bpf_skb_change_tail(skb, new_len, flags);
2988
2989         bpf_compute_data_end_sk_skb(skb);
2990         return ret;
2991 }
2992
2993 static const struct bpf_func_proto sk_skb_change_tail_proto = {
2994         .func           = sk_skb_change_tail,
2995         .gpl_only       = false,
2996         .ret_type       = RET_INTEGER,
2997         .arg1_type      = ARG_PTR_TO_CTX,
2998         .arg2_type      = ARG_ANYTHING,
2999         .arg3_type      = ARG_ANYTHING,
3000 };
3001
3002 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room,
3003                                         u64 flags)
3004 {
3005         u32 max_len = __bpf_skb_max_len(skb);
3006         u32 new_len = skb->len + head_room;
3007         int ret;
3008
3009         if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) ||
3010                      new_len < skb->len))
3011                 return -EINVAL;
3012
3013         ret = skb_cow(skb, head_room);
3014         if (likely(!ret)) {
3015                 /* Idea for this helper is that we currently only
3016                  * allow to expand on mac header. This means that
3017                  * skb->protocol network header, etc, stay as is.
3018                  * Compared to bpf_skb_change_tail(), we're more
3019                  * flexible due to not needing to linearize or
3020                  * reset GSO. Intention for this helper is to be
3021                  * used by an L3 skb that needs to push mac header
3022                  * for redirection into L2 device.
3023                  */
3024                 __skb_push(skb, head_room);
3025                 memset(skb->data, 0, head_room);
3026                 skb_reset_mac_header(skb);
3027         }
3028
3029         return ret;
3030 }
3031
3032 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room,
3033            u64, flags)
3034 {
3035         int ret = __bpf_skb_change_head(skb, head_room, flags);
3036
3037         bpf_compute_data_pointers(skb);
3038         return ret;
3039 }
3040
3041 static const struct bpf_func_proto bpf_skb_change_head_proto = {
3042         .func           = bpf_skb_change_head,
3043         .gpl_only       = false,
3044         .ret_type       = RET_INTEGER,
3045         .arg1_type      = ARG_PTR_TO_CTX,
3046         .arg2_type      = ARG_ANYTHING,
3047         .arg3_type      = ARG_ANYTHING,
3048 };
3049
3050 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room,
3051            u64, flags)
3052 {
3053         int ret = __bpf_skb_change_head(skb, head_room, flags);
3054
3055         bpf_compute_data_end_sk_skb(skb);
3056         return ret;
3057 }
3058
3059 static const struct bpf_func_proto sk_skb_change_head_proto = {
3060         .func           = sk_skb_change_head,
3061         .gpl_only       = false,
3062         .ret_type       = RET_INTEGER,
3063         .arg1_type      = ARG_PTR_TO_CTX,
3064         .arg2_type      = ARG_ANYTHING,
3065         .arg3_type      = ARG_ANYTHING,
3066 };
3067 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp)
3068 {
3069         return xdp_data_meta_unsupported(xdp) ? 0 :
3070                xdp->data - xdp->data_meta;
3071 }
3072
3073 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset)
3074 {
3075         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3076         unsigned long metalen = xdp_get_metalen(xdp);
3077         void *data_start = xdp_frame_end + metalen;
3078         void *data = xdp->data + offset;
3079
3080         if (unlikely(data < data_start ||
3081                      data > xdp->data_end - ETH_HLEN))
3082                 return -EINVAL;
3083
3084         if (metalen)
3085                 memmove(xdp->data_meta + offset,
3086                         xdp->data_meta, metalen);
3087         xdp->data_meta += offset;
3088         xdp->data = data;
3089
3090         return 0;
3091 }
3092
3093 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = {
3094         .func           = bpf_xdp_adjust_head,
3095         .gpl_only       = false,
3096         .ret_type       = RET_INTEGER,
3097         .arg1_type      = ARG_PTR_TO_CTX,
3098         .arg2_type      = ARG_ANYTHING,
3099 };
3100
3101 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset)
3102 {
3103         void *data_end = xdp->data_end + offset;
3104
3105         /* only shrinking is allowed for now. */
3106         if (unlikely(offset >= 0))
3107                 return -EINVAL;
3108
3109         if (unlikely(data_end < xdp->data + ETH_HLEN))
3110                 return -EINVAL;
3111
3112         xdp->data_end = data_end;
3113
3114         return 0;
3115 }
3116
3117 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = {
3118         .func           = bpf_xdp_adjust_tail,
3119         .gpl_only       = false,
3120         .ret_type       = RET_INTEGER,
3121         .arg1_type      = ARG_PTR_TO_CTX,
3122         .arg2_type      = ARG_ANYTHING,
3123 };
3124
3125 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset)
3126 {
3127         void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame);
3128         void *meta = xdp->data_meta + offset;
3129         unsigned long metalen = xdp->data - meta;
3130
3131         if (xdp_data_meta_unsupported(xdp))
3132                 return -ENOTSUPP;
3133         if (unlikely(meta < xdp_frame_end ||
3134                      meta > xdp->data))
3135                 return -EINVAL;
3136         if (unlikely((metalen & (sizeof(__u32) - 1)) ||
3137                      (metalen > 32)))
3138                 return -EACCES;
3139
3140         xdp->data_meta = meta;
3141
3142         return 0;
3143 }
3144
3145 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = {
3146         .func           = bpf_xdp_adjust_meta,
3147         .gpl_only       = false,
3148         .ret_type       = RET_INTEGER,
3149         .arg1_type      = ARG_PTR_TO_CTX,
3150         .arg2_type      = ARG_ANYTHING,
3151 };
3152
3153 static int __bpf_tx_xdp(struct net_device *dev,
3154                         struct bpf_map *map,
3155                         struct xdp_buff *xdp,
3156                         u32 index)
3157 {
3158         struct xdp_frame *xdpf;
3159         int err, sent;
3160
3161         if (!dev->netdev_ops->ndo_xdp_xmit) {
3162                 return -EOPNOTSUPP;
3163         }
3164
3165         err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data);
3166         if (unlikely(err))
3167                 return err;
3168
3169         xdpf = convert_to_xdp_frame(xdp);
3170         if (unlikely(!xdpf))
3171                 return -EOVERFLOW;
3172
3173         sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH);
3174         if (sent <= 0)
3175                 return sent;
3176         return 0;
3177 }
3178
3179 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd,
3180                             struct bpf_map *map,
3181                             struct xdp_buff *xdp,
3182                             u32 index)
3183 {
3184         int err;
3185
3186         switch (map->map_type) {
3187         case BPF_MAP_TYPE_DEVMAP: {
3188                 struct bpf_dtab_netdev *dst = fwd;
3189
3190                 err = dev_map_enqueue(dst, xdp, dev_rx);
3191                 if (err)
3192                         return err;
3193                 __dev_map_insert_ctx(map, index);
3194                 break;
3195         }
3196         case BPF_MAP_TYPE_CPUMAP: {
3197                 struct bpf_cpu_map_entry *rcpu = fwd;
3198
3199                 err = cpu_map_enqueue(rcpu, xdp, dev_rx);
3200                 if (err)
3201                         return err;
3202                 __cpu_map_insert_ctx(map, index);
3203                 break;
3204         }
3205         case BPF_MAP_TYPE_XSKMAP: {
3206                 struct xdp_sock *xs = fwd;
3207
3208                 err = __xsk_map_redirect(map, xdp, xs);
3209                 return err;
3210         }
3211         default:
3212                 break;
3213         }
3214         return 0;
3215 }
3216
3217 void xdp_do_flush_map(void)
3218 {
3219         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3220         struct bpf_map *map = ri->map_to_flush;
3221
3222         ri->map_to_flush = NULL;
3223         if (map) {
3224                 switch (map->map_type) {
3225                 case BPF_MAP_TYPE_DEVMAP:
3226                         __dev_map_flush(map);
3227                         break;
3228                 case BPF_MAP_TYPE_CPUMAP:
3229                         __cpu_map_flush(map);
3230                         break;
3231                 case BPF_MAP_TYPE_XSKMAP:
3232                         __xsk_map_flush(map);
3233                         break;
3234                 default:
3235                         break;
3236                 }
3237         }
3238 }
3239 EXPORT_SYMBOL_GPL(xdp_do_flush_map);
3240
3241 static void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index)
3242 {
3243         switch (map->map_type) {
3244         case BPF_MAP_TYPE_DEVMAP:
3245                 return __dev_map_lookup_elem(map, index);
3246         case BPF_MAP_TYPE_CPUMAP:
3247                 return __cpu_map_lookup_elem(map, index);
3248         case BPF_MAP_TYPE_XSKMAP:
3249                 return __xsk_map_lookup_elem(map, index);
3250         default:
3251                 return NULL;
3252         }
3253 }
3254
3255 void bpf_clear_redirect_map(struct bpf_map *map)
3256 {
3257         struct bpf_redirect_info *ri;
3258         int cpu;
3259
3260         for_each_possible_cpu(cpu) {
3261                 ri = per_cpu_ptr(&bpf_redirect_info, cpu);
3262                 /* Avoid polluting remote cacheline due to writes if
3263                  * not needed. Once we pass this test, we need the
3264                  * cmpxchg() to make sure it hasn't been changed in
3265                  * the meantime by remote CPU.
3266                  */
3267                 if (unlikely(READ_ONCE(ri->map) == map))
3268                         cmpxchg(&ri->map, map, NULL);
3269         }
3270 }
3271
3272 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp,
3273                                struct bpf_prog *xdp_prog, struct bpf_map *map)
3274 {
3275         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3276         u32 index = ri->ifindex;
3277         void *fwd = NULL;
3278         int err;
3279
3280         ri->ifindex = 0;
3281         WRITE_ONCE(ri->map, NULL);
3282
3283         fwd = __xdp_map_lookup_elem(map, index);
3284         if (!fwd) {
3285                 err = -EINVAL;
3286                 goto err;
3287         }
3288         if (ri->map_to_flush && ri->map_to_flush != map)
3289                 xdp_do_flush_map();
3290
3291         err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index);
3292         if (unlikely(err))
3293                 goto err;
3294
3295         ri->map_to_flush = map;
3296         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3297         return 0;
3298 err:
3299         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3300         return err;
3301 }
3302
3303 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
3304                     struct bpf_prog *xdp_prog)
3305 {
3306         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3307         struct bpf_map *map = READ_ONCE(ri->map);
3308         struct net_device *fwd;
3309         u32 index = ri->ifindex;
3310         int err;
3311
3312         if (map)
3313                 return xdp_do_redirect_map(dev, xdp, xdp_prog, map);
3314
3315         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3316         ri->ifindex = 0;
3317         if (unlikely(!fwd)) {
3318                 err = -EINVAL;
3319                 goto err;
3320         }
3321
3322         err = __bpf_tx_xdp(fwd, NULL, xdp, 0);
3323         if (unlikely(err))
3324                 goto err;
3325
3326         _trace_xdp_redirect(dev, xdp_prog, index);
3327         return 0;
3328 err:
3329         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3330         return err;
3331 }
3332 EXPORT_SYMBOL_GPL(xdp_do_redirect);
3333
3334 static int xdp_do_generic_redirect_map(struct net_device *dev,
3335                                        struct sk_buff *skb,
3336                                        struct xdp_buff *xdp,
3337                                        struct bpf_prog *xdp_prog,
3338                                        struct bpf_map *map)
3339 {
3340         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3341         u32 index = ri->ifindex;
3342         void *fwd = NULL;
3343         int err = 0;
3344
3345         ri->ifindex = 0;
3346         WRITE_ONCE(ri->map, NULL);
3347
3348         fwd = __xdp_map_lookup_elem(map, index);
3349         if (unlikely(!fwd)) {
3350                 err = -EINVAL;
3351                 goto err;
3352         }
3353
3354         if (map->map_type == BPF_MAP_TYPE_DEVMAP) {
3355                 struct bpf_dtab_netdev *dst = fwd;
3356
3357                 err = dev_map_generic_redirect(dst, skb, xdp_prog);
3358                 if (unlikely(err))
3359                         goto err;
3360         } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) {
3361                 struct xdp_sock *xs = fwd;
3362
3363                 err = xsk_generic_rcv(xs, xdp);
3364                 if (err)
3365                         goto err;
3366                 consume_skb(skb);
3367         } else {
3368                 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */
3369                 err = -EBADRQC;
3370                 goto err;
3371         }
3372
3373         _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index);
3374         return 0;
3375 err:
3376         _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err);
3377         return err;
3378 }
3379
3380 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
3381                             struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
3382 {
3383         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3384         struct bpf_map *map = READ_ONCE(ri->map);
3385         u32 index = ri->ifindex;
3386         struct net_device *fwd;
3387         int err = 0;
3388
3389         if (map)
3390                 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog,
3391                                                    map);
3392         ri->ifindex = 0;
3393         fwd = dev_get_by_index_rcu(dev_net(dev), index);
3394         if (unlikely(!fwd)) {
3395                 err = -EINVAL;
3396                 goto err;
3397         }
3398
3399         err = xdp_ok_fwd_dev(fwd, skb->len);
3400         if (unlikely(err))
3401                 goto err;
3402
3403         skb->dev = fwd;
3404         _trace_xdp_redirect(dev, xdp_prog, index);
3405         generic_xdp_tx(skb, xdp_prog);
3406         return 0;
3407 err:
3408         _trace_xdp_redirect_err(dev, xdp_prog, index, err);
3409         return err;
3410 }
3411 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect);
3412
3413 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
3414 {
3415         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3416
3417         if (unlikely(flags))
3418                 return XDP_ABORTED;
3419
3420         ri->ifindex = ifindex;
3421         ri->flags = flags;
3422         WRITE_ONCE(ri->map, NULL);
3423
3424         return XDP_REDIRECT;
3425 }
3426
3427 static const struct bpf_func_proto bpf_xdp_redirect_proto = {
3428         .func           = bpf_xdp_redirect,
3429         .gpl_only       = false,
3430         .ret_type       = RET_INTEGER,
3431         .arg1_type      = ARG_ANYTHING,
3432         .arg2_type      = ARG_ANYTHING,
3433 };
3434
3435 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex,
3436            u64, flags)
3437 {
3438         struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
3439
3440         if (unlikely(flags))
3441                 return XDP_ABORTED;
3442
3443         ri->ifindex = ifindex;
3444         ri->flags = flags;
3445         WRITE_ONCE(ri->map, map);
3446
3447         return XDP_REDIRECT;
3448 }
3449
3450 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = {
3451         .func           = bpf_xdp_redirect_map,
3452         .gpl_only       = false,
3453         .ret_type       = RET_INTEGER,
3454         .arg1_type      = ARG_CONST_MAP_PTR,
3455         .arg2_type      = ARG_ANYTHING,
3456         .arg3_type      = ARG_ANYTHING,
3457 };
3458
3459 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb,
3460                                   unsigned long off, unsigned long len)
3461 {
3462         void *ptr = skb_header_pointer(skb, off, len, dst_buff);
3463
3464         if (unlikely(!ptr))
3465                 return len;
3466         if (ptr != dst_buff)
3467                 memcpy(dst_buff, ptr, len);
3468
3469         return 0;
3470 }
3471
3472 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map,
3473            u64, flags, void *, meta, u64, meta_size)
3474 {
3475         u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3476
3477         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3478                 return -EINVAL;
3479         if (unlikely(skb_size > skb->len))
3480                 return -EFAULT;
3481
3482         return bpf_event_output(map, flags, meta, meta_size, skb, skb_size,
3483                                 bpf_skb_copy);
3484 }
3485
3486 static const struct bpf_func_proto bpf_skb_event_output_proto = {
3487         .func           = bpf_skb_event_output,
3488         .gpl_only       = true,
3489         .ret_type       = RET_INTEGER,
3490         .arg1_type      = ARG_PTR_TO_CTX,
3491         .arg2_type      = ARG_CONST_MAP_PTR,
3492         .arg3_type      = ARG_ANYTHING,
3493         .arg4_type      = ARG_PTR_TO_MEM,
3494         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3495 };
3496
3497 static unsigned short bpf_tunnel_key_af(u64 flags)
3498 {
3499         return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET;
3500 }
3501
3502 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to,
3503            u32, size, u64, flags)
3504 {
3505         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3506         u8 compat[sizeof(struct bpf_tunnel_key)];
3507         void *to_orig = to;
3508         int err;
3509
3510         if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) {
3511                 err = -EINVAL;
3512                 goto err_clear;
3513         }
3514         if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) {
3515                 err = -EPROTO;
3516                 goto err_clear;
3517         }
3518         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3519                 err = -EINVAL;
3520                 switch (size) {
3521                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3522                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3523                         goto set_compat;
3524                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3525                         /* Fixup deprecated structure layouts here, so we have
3526                          * a common path later on.
3527                          */
3528                         if (ip_tunnel_info_af(info) != AF_INET)
3529                                 goto err_clear;
3530 set_compat:
3531                         to = (struct bpf_tunnel_key *)compat;
3532                         break;
3533                 default:
3534                         goto err_clear;
3535                 }
3536         }
3537
3538         to->tunnel_id = be64_to_cpu(info->key.tun_id);
3539         to->tunnel_tos = info->key.tos;
3540         to->tunnel_ttl = info->key.ttl;
3541         to->tunnel_ext = 0;
3542
3543         if (flags & BPF_F_TUNINFO_IPV6) {
3544                 memcpy(to->remote_ipv6, &info->key.u.ipv6.src,
3545                        sizeof(to->remote_ipv6));
3546                 to->tunnel_label = be32_to_cpu(info->key.label);
3547         } else {
3548                 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src);
3549                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
3550                 to->tunnel_label = 0;
3551         }
3552
3553         if (unlikely(size != sizeof(struct bpf_tunnel_key)))
3554                 memcpy(to_orig, to, size);
3555
3556         return 0;
3557 err_clear:
3558         memset(to_orig, 0, size);
3559         return err;
3560 }
3561
3562 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = {
3563         .func           = bpf_skb_get_tunnel_key,
3564         .gpl_only       = false,
3565         .ret_type       = RET_INTEGER,
3566         .arg1_type      = ARG_PTR_TO_CTX,
3567         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3568         .arg3_type      = ARG_CONST_SIZE,
3569         .arg4_type      = ARG_ANYTHING,
3570 };
3571
3572 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size)
3573 {
3574         const struct ip_tunnel_info *info = skb_tunnel_info(skb);
3575         int err;
3576
3577         if (unlikely(!info ||
3578                      !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) {
3579                 err = -ENOENT;
3580                 goto err_clear;
3581         }
3582         if (unlikely(size < info->options_len)) {
3583                 err = -ENOMEM;
3584                 goto err_clear;
3585         }
3586
3587         ip_tunnel_info_opts_get(to, info);
3588         if (size > info->options_len)
3589                 memset(to + info->options_len, 0, size - info->options_len);
3590
3591         return info->options_len;
3592 err_clear:
3593         memset(to, 0, size);
3594         return err;
3595 }
3596
3597 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = {
3598         .func           = bpf_skb_get_tunnel_opt,
3599         .gpl_only       = false,
3600         .ret_type       = RET_INTEGER,
3601         .arg1_type      = ARG_PTR_TO_CTX,
3602         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
3603         .arg3_type      = ARG_CONST_SIZE,
3604 };
3605
3606 static struct metadata_dst __percpu *md_dst;
3607
3608 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb,
3609            const struct bpf_tunnel_key *, from, u32, size, u64, flags)
3610 {
3611         struct metadata_dst *md = this_cpu_ptr(md_dst);
3612         u8 compat[sizeof(struct bpf_tunnel_key)];
3613         struct ip_tunnel_info *info;
3614
3615         if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX |
3616                                BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER)))
3617                 return -EINVAL;
3618         if (unlikely(size != sizeof(struct bpf_tunnel_key))) {
3619                 switch (size) {
3620                 case offsetof(struct bpf_tunnel_key, tunnel_label):
3621                 case offsetof(struct bpf_tunnel_key, tunnel_ext):
3622                 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]):
3623                         /* Fixup deprecated structure layouts here, so we have
3624                          * a common path later on.
3625                          */
3626                         memcpy(compat, from, size);
3627                         memset(compat + size, 0, sizeof(compat) - size);
3628                         from = (const struct bpf_tunnel_key *) compat;
3629                         break;
3630                 default:
3631                         return -EINVAL;
3632                 }
3633         }
3634         if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) ||
3635                      from->tunnel_ext))
3636                 return -EINVAL;
3637
3638         skb_dst_drop(skb);
3639         dst_hold((struct dst_entry *) md);
3640         skb_dst_set(skb, (struct dst_entry *) md);
3641
3642         info = &md->u.tun_info;
3643         memset(info, 0, sizeof(*info));
3644         info->mode = IP_TUNNEL_INFO_TX;
3645
3646         info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE;
3647         if (flags & BPF_F_DONT_FRAGMENT)
3648                 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT;
3649         if (flags & BPF_F_ZERO_CSUM_TX)
3650                 info->key.tun_flags &= ~TUNNEL_CSUM;
3651         if (flags & BPF_F_SEQ_NUMBER)
3652                 info->key.tun_flags |= TUNNEL_SEQ;
3653
3654         info->key.tun_id = cpu_to_be64(from->tunnel_id);
3655         info->key.tos = from->tunnel_tos;
3656         info->key.ttl = from->tunnel_ttl;
3657
3658         if (flags & BPF_F_TUNINFO_IPV6) {
3659                 info->mode |= IP_TUNNEL_INFO_IPV6;
3660                 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6,
3661                        sizeof(from->remote_ipv6));
3662                 info->key.label = cpu_to_be32(from->tunnel_label) &
3663                                   IPV6_FLOWLABEL_MASK;
3664         } else {
3665                 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4);
3666         }
3667
3668         return 0;
3669 }
3670
3671 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = {
3672         .func           = bpf_skb_set_tunnel_key,
3673         .gpl_only       = false,
3674         .ret_type       = RET_INTEGER,
3675         .arg1_type      = ARG_PTR_TO_CTX,
3676         .arg2_type      = ARG_PTR_TO_MEM,
3677         .arg3_type      = ARG_CONST_SIZE,
3678         .arg4_type      = ARG_ANYTHING,
3679 };
3680
3681 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb,
3682            const u8 *, from, u32, size)
3683 {
3684         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3685         const struct metadata_dst *md = this_cpu_ptr(md_dst);
3686
3687         if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1))))
3688                 return -EINVAL;
3689         if (unlikely(size > IP_TUNNEL_OPTS_MAX))
3690                 return -ENOMEM;
3691
3692         ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT);
3693
3694         return 0;
3695 }
3696
3697 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = {
3698         .func           = bpf_skb_set_tunnel_opt,
3699         .gpl_only       = false,
3700         .ret_type       = RET_INTEGER,
3701         .arg1_type      = ARG_PTR_TO_CTX,
3702         .arg2_type      = ARG_PTR_TO_MEM,
3703         .arg3_type      = ARG_CONST_SIZE,
3704 };
3705
3706 static const struct bpf_func_proto *
3707 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which)
3708 {
3709         if (!md_dst) {
3710                 struct metadata_dst __percpu *tmp;
3711
3712                 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX,
3713                                                 METADATA_IP_TUNNEL,
3714                                                 GFP_KERNEL);
3715                 if (!tmp)
3716                         return NULL;
3717                 if (cmpxchg(&md_dst, NULL, tmp))
3718                         metadata_dst_free_percpu(tmp);
3719         }
3720
3721         switch (which) {
3722         case BPF_FUNC_skb_set_tunnel_key:
3723                 return &bpf_skb_set_tunnel_key_proto;
3724         case BPF_FUNC_skb_set_tunnel_opt:
3725                 return &bpf_skb_set_tunnel_opt_proto;
3726         default:
3727                 return NULL;
3728         }
3729 }
3730
3731 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map,
3732            u32, idx)
3733 {
3734         struct bpf_array *array = container_of(map, struct bpf_array, map);
3735         struct cgroup *cgrp;
3736         struct sock *sk;
3737
3738         sk = skb_to_full_sk(skb);
3739         if (!sk || !sk_fullsock(sk))
3740                 return -ENOENT;
3741         if (unlikely(idx >= array->map.max_entries))
3742                 return -E2BIG;
3743
3744         cgrp = READ_ONCE(array->ptrs[idx]);
3745         if (unlikely(!cgrp))
3746                 return -EAGAIN;
3747
3748         return sk_under_cgroup_hierarchy(sk, cgrp);
3749 }
3750
3751 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = {
3752         .func           = bpf_skb_under_cgroup,
3753         .gpl_only       = false,
3754         .ret_type       = RET_INTEGER,
3755         .arg1_type      = ARG_PTR_TO_CTX,
3756         .arg2_type      = ARG_CONST_MAP_PTR,
3757         .arg3_type      = ARG_ANYTHING,
3758 };
3759
3760 #ifdef CONFIG_SOCK_CGROUP_DATA
3761 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb)
3762 {
3763         struct sock *sk = skb_to_full_sk(skb);
3764         struct cgroup *cgrp;
3765
3766         if (!sk || !sk_fullsock(sk))
3767                 return 0;
3768
3769         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3770         return cgrp->kn->id.id;
3771 }
3772
3773 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = {
3774         .func           = bpf_skb_cgroup_id,
3775         .gpl_only       = false,
3776         .ret_type       = RET_INTEGER,
3777         .arg1_type      = ARG_PTR_TO_CTX,
3778 };
3779
3780 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int,
3781            ancestor_level)
3782 {
3783         struct sock *sk = skb_to_full_sk(skb);
3784         struct cgroup *ancestor;
3785         struct cgroup *cgrp;
3786
3787         if (!sk || !sk_fullsock(sk))
3788                 return 0;
3789
3790         cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
3791         ancestor = cgroup_ancestor(cgrp, ancestor_level);
3792         if (!ancestor)
3793                 return 0;
3794
3795         return ancestor->kn->id.id;
3796 }
3797
3798 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = {
3799         .func           = bpf_skb_ancestor_cgroup_id,
3800         .gpl_only       = false,
3801         .ret_type       = RET_INTEGER,
3802         .arg1_type      = ARG_PTR_TO_CTX,
3803         .arg2_type      = ARG_ANYTHING,
3804 };
3805 #endif
3806
3807 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff,
3808                                   unsigned long off, unsigned long len)
3809 {
3810         memcpy(dst_buff, src_buff + off, len);
3811         return 0;
3812 }
3813
3814 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map,
3815            u64, flags, void *, meta, u64, meta_size)
3816 {
3817         u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32;
3818
3819         if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK)))
3820                 return -EINVAL;
3821         if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data)))
3822                 return -EFAULT;
3823
3824         return bpf_event_output(map, flags, meta, meta_size, xdp->data,
3825                                 xdp_size, bpf_xdp_copy);
3826 }
3827
3828 static const struct bpf_func_proto bpf_xdp_event_output_proto = {
3829         .func           = bpf_xdp_event_output,
3830         .gpl_only       = true,
3831         .ret_type       = RET_INTEGER,
3832         .arg1_type      = ARG_PTR_TO_CTX,
3833         .arg2_type      = ARG_CONST_MAP_PTR,
3834         .arg3_type      = ARG_ANYTHING,
3835         .arg4_type      = ARG_PTR_TO_MEM,
3836         .arg5_type      = ARG_CONST_SIZE_OR_ZERO,
3837 };
3838
3839 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb)
3840 {
3841         return skb->sk ? sock_gen_cookie(skb->sk) : 0;
3842 }
3843
3844 static const struct bpf_func_proto bpf_get_socket_cookie_proto = {
3845         .func           = bpf_get_socket_cookie,
3846         .gpl_only       = false,
3847         .ret_type       = RET_INTEGER,
3848         .arg1_type      = ARG_PTR_TO_CTX,
3849 };
3850
3851 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx)
3852 {
3853         return sock_gen_cookie(ctx->sk);
3854 }
3855
3856 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = {
3857         .func           = bpf_get_socket_cookie_sock_addr,
3858         .gpl_only       = false,
3859         .ret_type       = RET_INTEGER,
3860         .arg1_type      = ARG_PTR_TO_CTX,
3861 };
3862
3863 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx)
3864 {
3865         return sock_gen_cookie(ctx->sk);
3866 }
3867
3868 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = {
3869         .func           = bpf_get_socket_cookie_sock_ops,
3870         .gpl_only       = false,
3871         .ret_type       = RET_INTEGER,
3872         .arg1_type      = ARG_PTR_TO_CTX,
3873 };
3874
3875 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb)
3876 {
3877         struct sock *sk = sk_to_full_sk(skb->sk);
3878         kuid_t kuid;
3879
3880         if (!sk || !sk_fullsock(sk))
3881                 return overflowuid;
3882         kuid = sock_net_uid(sock_net(sk), sk);
3883         return from_kuid_munged(sock_net(sk)->user_ns, kuid);
3884 }
3885
3886 static const struct bpf_func_proto bpf_get_socket_uid_proto = {
3887         .func           = bpf_get_socket_uid,
3888         .gpl_only       = false,
3889         .ret_type       = RET_INTEGER,
3890         .arg1_type      = ARG_PTR_TO_CTX,
3891 };
3892
3893 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock,
3894            int, level, int, optname, char *, optval, int, optlen)
3895 {
3896         struct sock *sk = bpf_sock->sk;
3897         int ret = 0;
3898         int val;
3899
3900         if (!sk_fullsock(sk))
3901                 return -EINVAL;
3902
3903         if (level == SOL_SOCKET) {
3904                 if (optlen != sizeof(int))
3905                         return -EINVAL;
3906                 val = *((int *)optval);
3907
3908                 /* Only some socketops are supported */
3909                 switch (optname) {
3910                 case SO_RCVBUF:
3911                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
3912                         sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
3913                         break;
3914                 case SO_SNDBUF:
3915                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
3916                         sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
3917                         break;
3918                 case SO_MAX_PACING_RATE:
3919                         sk->sk_max_pacing_rate = val;
3920                         sk->sk_pacing_rate = min(sk->sk_pacing_rate,
3921                                                  sk->sk_max_pacing_rate);
3922                         break;
3923                 case SO_PRIORITY:
3924                         sk->sk_priority = val;
3925                         break;
3926                 case SO_RCVLOWAT:
3927                         if (val < 0)
3928                                 val = INT_MAX;
3929                         sk->sk_rcvlowat = val ? : 1;
3930                         break;
3931                 case SO_MARK:
3932                         sk->sk_mark = val;
3933                         break;
3934                 default:
3935                         ret = -EINVAL;
3936                 }
3937 #ifdef CONFIG_INET
3938         } else if (level == SOL_IP) {
3939                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
3940                         return -EINVAL;
3941
3942                 val = *((int *)optval);
3943                 /* Only some options are supported */
3944                 switch (optname) {
3945                 case IP_TOS:
3946                         if (val < -1 || val > 0xff) {
3947                                 ret = -EINVAL;
3948                         } else {
3949                                 struct inet_sock *inet = inet_sk(sk);
3950
3951                                 if (val == -1)
3952                                         val = 0;
3953                                 inet->tos = val;
3954                         }
3955                         break;
3956                 default:
3957                         ret = -EINVAL;
3958                 }
3959 #if IS_ENABLED(CONFIG_IPV6)
3960         } else if (level == SOL_IPV6) {
3961                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
3962                         return -EINVAL;
3963
3964                 val = *((int *)optval);
3965                 /* Only some options are supported */
3966                 switch (optname) {
3967                 case IPV6_TCLASS:
3968                         if (val < -1 || val > 0xff) {
3969                                 ret = -EINVAL;
3970                         } else {
3971                                 struct ipv6_pinfo *np = inet6_sk(sk);
3972
3973                                 if (val == -1)
3974                                         val = 0;
3975                                 np->tclass = val;
3976                         }
3977                         break;
3978                 default:
3979                         ret = -EINVAL;
3980                 }
3981 #endif
3982         } else if (level == SOL_TCP &&
3983                    sk->sk_prot->setsockopt == tcp_setsockopt) {
3984                 if (optname == TCP_CONGESTION) {
3985                         char name[TCP_CA_NAME_MAX];
3986                         bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN;
3987
3988                         strncpy(name, optval, min_t(long, optlen,
3989                                                     TCP_CA_NAME_MAX-1));
3990                         name[TCP_CA_NAME_MAX-1] = 0;
3991                         ret = tcp_set_congestion_control(sk, name, false,
3992                                                          reinit);
3993                 } else {
3994                         struct tcp_sock *tp = tcp_sk(sk);
3995
3996                         if (optlen != sizeof(int))
3997                                 return -EINVAL;
3998
3999                         val = *((int *)optval);
4000                         /* Only some options are supported */
4001                         switch (optname) {
4002                         case TCP_BPF_IW:
4003                                 if (val <= 0 || tp->data_segs_out > 0)
4004                                         ret = -EINVAL;
4005                                 else
4006                                         tp->snd_cwnd = val;
4007                                 break;
4008                         case TCP_BPF_SNDCWND_CLAMP:
4009                                 if (val <= 0) {
4010                                         ret = -EINVAL;
4011                                 } else {
4012                                         tp->snd_cwnd_clamp = val;
4013                                         tp->snd_ssthresh = val;
4014                                 }
4015                                 break;
4016                         default:
4017                                 ret = -EINVAL;
4018                         }
4019                 }
4020 #endif
4021         } else {
4022                 ret = -EINVAL;
4023         }
4024         return ret;
4025 }
4026
4027 static const struct bpf_func_proto bpf_setsockopt_proto = {
4028         .func           = bpf_setsockopt,
4029         .gpl_only       = false,
4030         .ret_type       = RET_INTEGER,
4031         .arg1_type      = ARG_PTR_TO_CTX,
4032         .arg2_type      = ARG_ANYTHING,
4033         .arg3_type      = ARG_ANYTHING,
4034         .arg4_type      = ARG_PTR_TO_MEM,
4035         .arg5_type      = ARG_CONST_SIZE,
4036 };
4037
4038 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock,
4039            int, level, int, optname, char *, optval, int, optlen)
4040 {
4041         struct sock *sk = bpf_sock->sk;
4042
4043         if (!sk_fullsock(sk))
4044                 goto err_clear;
4045
4046 #ifdef CONFIG_INET
4047         if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) {
4048                 if (optname == TCP_CONGESTION) {
4049                         struct inet_connection_sock *icsk = inet_csk(sk);
4050
4051                         if (!icsk->icsk_ca_ops || optlen <= 1)
4052                                 goto err_clear;
4053                         strncpy(optval, icsk->icsk_ca_ops->name, optlen);
4054                         optval[optlen - 1] = 0;
4055                 } else {
4056                         goto err_clear;
4057                 }
4058         } else if (level == SOL_IP) {
4059                 struct inet_sock *inet = inet_sk(sk);
4060
4061                 if (optlen != sizeof(int) || sk->sk_family != AF_INET)
4062                         goto err_clear;
4063
4064                 /* Only some options are supported */
4065                 switch (optname) {
4066                 case IP_TOS:
4067                         *((int *)optval) = (int)inet->tos;
4068                         break;
4069                 default:
4070                         goto err_clear;
4071                 }
4072 #if IS_ENABLED(CONFIG_IPV6)
4073         } else if (level == SOL_IPV6) {
4074                 struct ipv6_pinfo *np = inet6_sk(sk);
4075
4076                 if (optlen != sizeof(int) || sk->sk_family != AF_INET6)
4077                         goto err_clear;
4078
4079                 /* Only some options are supported */
4080                 switch (optname) {
4081                 case IPV6_TCLASS:
4082                         *((int *)optval) = (int)np->tclass;
4083                         break;
4084                 default:
4085                         goto err_clear;
4086                 }
4087 #endif
4088         } else {
4089                 goto err_clear;
4090         }
4091         return 0;
4092 #endif
4093 err_clear:
4094         memset(optval, 0, optlen);
4095         return -EINVAL;
4096 }
4097
4098 static const struct bpf_func_proto bpf_getsockopt_proto = {
4099         .func           = bpf_getsockopt,
4100         .gpl_only       = false,
4101         .ret_type       = RET_INTEGER,
4102         .arg1_type      = ARG_PTR_TO_CTX,
4103         .arg2_type      = ARG_ANYTHING,
4104         .arg3_type      = ARG_ANYTHING,
4105         .arg4_type      = ARG_PTR_TO_UNINIT_MEM,
4106         .arg5_type      = ARG_CONST_SIZE,
4107 };
4108
4109 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock,
4110            int, argval)
4111 {
4112         struct sock *sk = bpf_sock->sk;
4113         int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS;
4114
4115         if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk))
4116                 return -EINVAL;
4117
4118         if (val)
4119                 tcp_sk(sk)->bpf_sock_ops_cb_flags = val;
4120
4121         return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS);
4122 }
4123
4124 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = {
4125         .func           = bpf_sock_ops_cb_flags_set,
4126         .gpl_only       = false,
4127         .ret_type       = RET_INTEGER,
4128         .arg1_type      = ARG_PTR_TO_CTX,
4129         .arg2_type      = ARG_ANYTHING,
4130 };
4131
4132 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly;
4133 EXPORT_SYMBOL_GPL(ipv6_bpf_stub);
4134
4135 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr,
4136            int, addr_len)
4137 {
4138 #ifdef CONFIG_INET
4139         struct sock *sk = ctx->sk;
4140         int err;
4141
4142         /* Binding to port can be expensive so it's prohibited in the helper.
4143          * Only binding to IP is supported.
4144          */
4145         err = -EINVAL;
4146         if (addr->sa_family == AF_INET) {
4147                 if (addr_len < sizeof(struct sockaddr_in))
4148                         return err;
4149                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
4150                         return err;
4151                 return __inet_bind(sk, addr, addr_len, true, false);
4152 #if IS_ENABLED(CONFIG_IPV6)
4153         } else if (addr->sa_family == AF_INET6) {
4154                 if (addr_len < SIN6_LEN_RFC2133)
4155                         return err;
4156                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
4157                         return err;
4158                 /* ipv6_bpf_stub cannot be NULL, since it's called from
4159                  * bpf_cgroup_inet6_connect hook and ipv6 is already loaded
4160                  */
4161                 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false);
4162 #endif /* CONFIG_IPV6 */
4163         }
4164 #endif /* CONFIG_INET */
4165
4166         return -EAFNOSUPPORT;
4167 }
4168
4169 static const struct bpf_func_proto bpf_bind_proto = {
4170         .func           = bpf_bind,
4171         .gpl_only       = false,
4172         .ret_type       = RET_INTEGER,
4173         .arg1_type      = ARG_PTR_TO_CTX,
4174         .arg2_type      = ARG_PTR_TO_MEM,
4175         .arg3_type      = ARG_CONST_SIZE,
4176 };
4177
4178 #ifdef CONFIG_XFRM
4179 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index,
4180            struct bpf_xfrm_state *, to, u32, size, u64, flags)
4181 {
4182         const struct sec_path *sp = skb_sec_path(skb);
4183         const struct xfrm_state *x;
4184
4185         if (!sp || unlikely(index >= sp->len || flags))
4186                 goto err_clear;
4187
4188         x = sp->xvec[index];
4189
4190         if (unlikely(size != sizeof(struct bpf_xfrm_state)))
4191                 goto err_clear;
4192
4193         to->reqid = x->props.reqid;
4194         to->spi = x->id.spi;
4195         to->family = x->props.family;
4196         to->ext = 0;
4197
4198         if (to->family == AF_INET6) {
4199                 memcpy(to->remote_ipv6, x->props.saddr.a6,
4200                        sizeof(to->remote_ipv6));
4201         } else {
4202                 to->remote_ipv4 = x->props.saddr.a4;
4203                 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3);
4204         }
4205
4206         return 0;
4207 err_clear:
4208         memset(to, 0, size);
4209         return -EINVAL;
4210 }
4211
4212 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = {
4213         .func           = bpf_skb_get_xfrm_state,
4214         .gpl_only       = false,
4215         .ret_type       = RET_INTEGER,
4216         .arg1_type      = ARG_PTR_TO_CTX,
4217         .arg2_type      = ARG_ANYTHING,
4218         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
4219         .arg4_type      = ARG_CONST_SIZE,
4220         .arg5_type      = ARG_ANYTHING,
4221 };
4222 #endif
4223
4224 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6)
4225 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params,
4226                                   const struct neighbour *neigh,
4227                                   const struct net_device *dev)
4228 {
4229         memcpy(params->dmac, neigh->ha, ETH_ALEN);
4230         memcpy(params->smac, dev->dev_addr, ETH_ALEN);
4231         params->h_vlan_TCI = 0;
4232         params->h_vlan_proto = 0;
4233         params->ifindex = dev->ifindex;
4234
4235         return 0;
4236 }
4237 #endif
4238
4239 #if IS_ENABLED(CONFIG_INET)
4240 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4241                                u32 flags, bool check_mtu)
4242 {
4243         struct in_device *in_dev;
4244         struct neighbour *neigh;
4245         struct net_device *dev;
4246         struct fib_result res;
4247         struct fib_nh *nh;
4248         struct flowi4 fl4;
4249         int err;
4250         u32 mtu;
4251
4252         dev = dev_get_by_index_rcu(net, params->ifindex);
4253         if (unlikely(!dev))
4254                 return -ENODEV;
4255
4256         /* verify forwarding is enabled on this interface */
4257         in_dev = __in_dev_get_rcu(dev);
4258         if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev)))
4259                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4260
4261         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4262                 fl4.flowi4_iif = 1;
4263                 fl4.flowi4_oif = params->ifindex;
4264         } else {
4265                 fl4.flowi4_iif = params->ifindex;
4266                 fl4.flowi4_oif = 0;
4267         }
4268         fl4.flowi4_tos = params->tos & IPTOS_RT_MASK;
4269         fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
4270         fl4.flowi4_flags = 0;
4271
4272         fl4.flowi4_proto = params->l4_protocol;
4273         fl4.daddr = params->ipv4_dst;
4274         fl4.saddr = params->ipv4_src;
4275         fl4.fl4_sport = params->sport;
4276         fl4.fl4_dport = params->dport;
4277
4278         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4279                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4280                 struct fib_table *tb;
4281
4282                 tb = fib_get_table(net, tbid);
4283                 if (unlikely(!tb))
4284                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4285
4286                 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF);
4287         } else {
4288                 fl4.flowi4_mark = 0;
4289                 fl4.flowi4_secid = 0;
4290                 fl4.flowi4_tun_key.tun_id = 0;
4291                 fl4.flowi4_uid = sock_net_uid(net, NULL);
4292
4293                 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF);
4294         }
4295
4296         if (err) {
4297                 /* map fib lookup errors to RTN_ type */
4298                 if (err == -EINVAL)
4299                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4300                 if (err == -EHOSTUNREACH)
4301                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4302                 if (err == -EACCES)
4303                         return BPF_FIB_LKUP_RET_PROHIBIT;
4304
4305                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4306         }
4307
4308         if (res.type != RTN_UNICAST)
4309                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4310
4311         if (res.fi->fib_nhs > 1)
4312                 fib_select_path(net, &res, &fl4, NULL);
4313
4314         if (check_mtu) {
4315                 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst);
4316                 if (params->tot_len > mtu)
4317                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4318         }
4319
4320         nh = &res.fi->fib_nh[res.nh_sel];
4321
4322         /* do not handle lwt encaps right now */
4323         if (nh->nh_lwtstate)
4324                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4325
4326         dev = nh->nh_dev;
4327         if (nh->nh_gw)
4328                 params->ipv4_dst = nh->nh_gw;
4329
4330         params->rt_metric = res.fi->fib_priority;
4331
4332         /* xdp and cls_bpf programs are run in RCU-bh so
4333          * rcu_read_lock_bh is not needed here
4334          */
4335         neigh = __ipv4_neigh_lookup_noref(dev, (__force u32)params->ipv4_dst);
4336         if (!neigh)
4337                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4338
4339         return bpf_fib_set_fwd_params(params, neigh, dev);
4340 }
4341 #endif
4342
4343 #if IS_ENABLED(CONFIG_IPV6)
4344 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params,
4345                                u32 flags, bool check_mtu)
4346 {
4347         struct in6_addr *src = (struct in6_addr *) params->ipv6_src;
4348         struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst;
4349         struct neighbour *neigh;
4350         struct net_device *dev;
4351         struct inet6_dev *idev;
4352         struct fib6_info *f6i;
4353         struct flowi6 fl6;
4354         int strict = 0;
4355         int oif;
4356         u32 mtu;
4357
4358         /* link local addresses are never forwarded */
4359         if (rt6_need_strict(dst) || rt6_need_strict(src))
4360                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4361
4362         dev = dev_get_by_index_rcu(net, params->ifindex);
4363         if (unlikely(!dev))
4364                 return -ENODEV;
4365
4366         idev = __in6_dev_get_safely(dev);
4367         if (unlikely(!idev || !net->ipv6.devconf_all->forwarding))
4368                 return BPF_FIB_LKUP_RET_FWD_DISABLED;
4369
4370         if (flags & BPF_FIB_LOOKUP_OUTPUT) {
4371                 fl6.flowi6_iif = 1;
4372                 oif = fl6.flowi6_oif = params->ifindex;
4373         } else {
4374                 oif = fl6.flowi6_iif = params->ifindex;
4375                 fl6.flowi6_oif = 0;
4376                 strict = RT6_LOOKUP_F_HAS_SADDR;
4377         }
4378         fl6.flowlabel = params->flowinfo;
4379         fl6.flowi6_scope = 0;
4380         fl6.flowi6_flags = 0;
4381         fl6.mp_hash = 0;
4382
4383         fl6.flowi6_proto = params->l4_protocol;
4384         fl6.daddr = *dst;
4385         fl6.saddr = *src;
4386         fl6.fl6_sport = params->sport;
4387         fl6.fl6_dport = params->dport;
4388
4389         if (flags & BPF_FIB_LOOKUP_DIRECT) {
4390                 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN;
4391                 struct fib6_table *tb;
4392
4393                 tb = ipv6_stub->fib6_get_table(net, tbid);
4394                 if (unlikely(!tb))
4395                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4396
4397                 f6i = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, strict);
4398         } else {
4399                 fl6.flowi6_mark = 0;
4400                 fl6.flowi6_secid = 0;
4401                 fl6.flowi6_tun_key.tun_id = 0;
4402                 fl6.flowi6_uid = sock_net_uid(net, NULL);
4403
4404                 f6i = ipv6_stub->fib6_lookup(net, oif, &fl6, strict);
4405         }
4406
4407         if (unlikely(IS_ERR_OR_NULL(f6i) || f6i == net->ipv6.fib6_null_entry))
4408                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4409
4410         if (unlikely(f6i->fib6_flags & RTF_REJECT)) {
4411                 switch (f6i->fib6_type) {
4412                 case RTN_BLACKHOLE:
4413                         return BPF_FIB_LKUP_RET_BLACKHOLE;
4414                 case RTN_UNREACHABLE:
4415                         return BPF_FIB_LKUP_RET_UNREACHABLE;
4416                 case RTN_PROHIBIT:
4417                         return BPF_FIB_LKUP_RET_PROHIBIT;
4418                 default:
4419                         return BPF_FIB_LKUP_RET_NOT_FWDED;
4420                 }
4421         }
4422
4423         if (f6i->fib6_type != RTN_UNICAST)
4424                 return BPF_FIB_LKUP_RET_NOT_FWDED;
4425
4426         if (f6i->fib6_nsiblings && fl6.flowi6_oif == 0)
4427                 f6i = ipv6_stub->fib6_multipath_select(net, f6i, &fl6,
4428                                                        fl6.flowi6_oif, NULL,
4429                                                        strict);
4430
4431         if (check_mtu) {
4432                 mtu = ipv6_stub->ip6_mtu_from_fib6(f6i, dst, src);
4433                 if (params->tot_len > mtu)
4434                         return BPF_FIB_LKUP_RET_FRAG_NEEDED;
4435         }
4436
4437         if (f6i->fib6_nh.nh_lwtstate)
4438                 return BPF_FIB_LKUP_RET_UNSUPP_LWT;
4439
4440         if (f6i->fib6_flags & RTF_GATEWAY)
4441                 *dst = f6i->fib6_nh.nh_gw;
4442
4443         dev = f6i->fib6_nh.nh_dev;
4444         params->rt_metric = f6i->fib6_metric;
4445
4446         /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is
4447          * not needed here. Can not use __ipv6_neigh_lookup_noref here
4448          * because we need to get nd_tbl via the stub
4449          */
4450         neigh = ___neigh_lookup_noref(ipv6_stub->nd_tbl, neigh_key_eq128,
4451                                       ndisc_hashfn, dst, dev);
4452         if (!neigh)
4453                 return BPF_FIB_LKUP_RET_NO_NEIGH;
4454
4455         return bpf_fib_set_fwd_params(params, neigh, dev);
4456 }
4457 #endif
4458
4459 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx,
4460            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4461 {
4462         if (plen < sizeof(*params))
4463                 return -EINVAL;
4464
4465         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4466                 return -EINVAL;
4467
4468         switch (params->family) {
4469 #if IS_ENABLED(CONFIG_INET)
4470         case AF_INET:
4471                 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params,
4472                                            flags, true);
4473 #endif
4474 #if IS_ENABLED(CONFIG_IPV6)
4475         case AF_INET6:
4476                 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params,
4477                                            flags, true);
4478 #endif
4479         }
4480         return -EAFNOSUPPORT;
4481 }
4482
4483 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = {
4484         .func           = bpf_xdp_fib_lookup,
4485         .gpl_only       = true,
4486         .ret_type       = RET_INTEGER,
4487         .arg1_type      = ARG_PTR_TO_CTX,
4488         .arg2_type      = ARG_PTR_TO_MEM,
4489         .arg3_type      = ARG_CONST_SIZE,
4490         .arg4_type      = ARG_ANYTHING,
4491 };
4492
4493 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb,
4494            struct bpf_fib_lookup *, params, int, plen, u32, flags)
4495 {
4496         struct net *net = dev_net(skb->dev);
4497         int rc = -EAFNOSUPPORT;
4498
4499         if (plen < sizeof(*params))
4500                 return -EINVAL;
4501
4502         if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT))
4503                 return -EINVAL;
4504
4505         switch (params->family) {
4506 #if IS_ENABLED(CONFIG_INET)
4507         case AF_INET:
4508                 rc = bpf_ipv4_fib_lookup(net, params, flags, false);
4509                 break;
4510 #endif
4511 #if IS_ENABLED(CONFIG_IPV6)
4512         case AF_INET6:
4513                 rc = bpf_ipv6_fib_lookup(net, params, flags, false);
4514                 break;
4515 #endif
4516         }
4517
4518         if (!rc) {
4519                 struct net_device *dev;
4520
4521                 dev = dev_get_by_index_rcu(net, params->ifindex);
4522                 if (!is_skb_forwardable(dev, skb))
4523                         rc = BPF_FIB_LKUP_RET_FRAG_NEEDED;
4524         }
4525
4526         return rc;
4527 }
4528
4529 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = {
4530         .func           = bpf_skb_fib_lookup,
4531         .gpl_only       = true,
4532         .ret_type       = RET_INTEGER,
4533         .arg1_type      = ARG_PTR_TO_CTX,
4534         .arg2_type      = ARG_PTR_TO_MEM,
4535         .arg3_type      = ARG_CONST_SIZE,
4536         .arg4_type      = ARG_ANYTHING,
4537 };
4538
4539 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4540 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len)
4541 {
4542         int err;
4543         struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr;
4544
4545         if (!seg6_validate_srh(srh, len))
4546                 return -EINVAL;
4547
4548         switch (type) {
4549         case BPF_LWT_ENCAP_SEG6_INLINE:
4550                 if (skb->protocol != htons(ETH_P_IPV6))
4551                         return -EBADMSG;
4552
4553                 err = seg6_do_srh_inline(skb, srh);
4554                 break;
4555         case BPF_LWT_ENCAP_SEG6:
4556                 skb_reset_inner_headers(skb);
4557                 skb->encapsulation = 1;
4558                 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6);
4559                 break;
4560         default:
4561                 return -EINVAL;
4562         }
4563
4564         bpf_compute_data_pointers(skb);
4565         if (err)
4566                 return err;
4567
4568         ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4569         skb_set_transport_header(skb, sizeof(struct ipv6hdr));
4570
4571         return seg6_lookup_nexthop(skb, NULL, 0);
4572 }
4573 #endif /* CONFIG_IPV6_SEG6_BPF */
4574
4575 BPF_CALL_4(bpf_lwt_push_encap, struct sk_buff *, skb, u32, type, void *, hdr,
4576            u32, len)
4577 {
4578         switch (type) {
4579 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4580         case BPF_LWT_ENCAP_SEG6:
4581         case BPF_LWT_ENCAP_SEG6_INLINE:
4582                 return bpf_push_seg6_encap(skb, type, hdr, len);
4583 #endif
4584         default:
4585                 return -EINVAL;
4586         }
4587 }
4588
4589 static const struct bpf_func_proto bpf_lwt_push_encap_proto = {
4590         .func           = bpf_lwt_push_encap,
4591         .gpl_only       = false,
4592         .ret_type       = RET_INTEGER,
4593         .arg1_type      = ARG_PTR_TO_CTX,
4594         .arg2_type      = ARG_ANYTHING,
4595         .arg3_type      = ARG_PTR_TO_MEM,
4596         .arg4_type      = ARG_CONST_SIZE
4597 };
4598
4599 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4600 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
4601            const void *, from, u32, len)
4602 {
4603         struct seg6_bpf_srh_state *srh_state =
4604                 this_cpu_ptr(&seg6_bpf_srh_states);
4605         struct ipv6_sr_hdr *srh = srh_state->srh;
4606         void *srh_tlvs, *srh_end, *ptr;
4607         int srhoff = 0;
4608
4609         if (srh == NULL)
4610                 return -EINVAL;
4611
4612         srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4));
4613         srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen);
4614
4615         ptr = skb->data + offset;
4616         if (ptr >= srh_tlvs && ptr + len <= srh_end)
4617                 srh_state->valid = false;
4618         else if (ptr < (void *)&srh->flags ||
4619                  ptr + len > (void *)&srh->segments)
4620                 return -EFAULT;
4621
4622         if (unlikely(bpf_try_make_writable(skb, offset + len)))
4623                 return -EFAULT;
4624         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4625                 return -EINVAL;
4626         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4627
4628         memcpy(skb->data + offset, from, len);
4629         return 0;
4630 }
4631
4632 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = {
4633         .func           = bpf_lwt_seg6_store_bytes,
4634         .gpl_only       = false,
4635         .ret_type       = RET_INTEGER,
4636         .arg1_type      = ARG_PTR_TO_CTX,
4637         .arg2_type      = ARG_ANYTHING,
4638         .arg3_type      = ARG_PTR_TO_MEM,
4639         .arg4_type      = ARG_CONST_SIZE
4640 };
4641
4642 static void bpf_update_srh_state(struct sk_buff *skb)
4643 {
4644         struct seg6_bpf_srh_state *srh_state =
4645                 this_cpu_ptr(&seg6_bpf_srh_states);
4646         int srhoff = 0;
4647
4648         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) {
4649                 srh_state->srh = NULL;
4650         } else {
4651                 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4652                 srh_state->hdrlen = srh_state->srh->hdrlen << 3;
4653                 srh_state->valid = true;
4654         }
4655 }
4656
4657 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
4658            u32, action, void *, param, u32, param_len)
4659 {
4660         struct seg6_bpf_srh_state *srh_state =
4661                 this_cpu_ptr(&seg6_bpf_srh_states);
4662         int hdroff = 0;
4663         int err;
4664
4665         switch (action) {
4666         case SEG6_LOCAL_ACTION_END_X:
4667                 if (!seg6_bpf_has_valid_srh(skb))
4668                         return -EBADMSG;
4669                 if (param_len != sizeof(struct in6_addr))
4670                         return -EINVAL;
4671                 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0);
4672         case SEG6_LOCAL_ACTION_END_T:
4673                 if (!seg6_bpf_has_valid_srh(skb))
4674                         return -EBADMSG;
4675                 if (param_len != sizeof(int))
4676                         return -EINVAL;
4677                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4678         case SEG6_LOCAL_ACTION_END_DT6:
4679                 if (!seg6_bpf_has_valid_srh(skb))
4680                         return -EBADMSG;
4681                 if (param_len != sizeof(int))
4682                         return -EINVAL;
4683
4684                 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0)
4685                         return -EBADMSG;
4686                 if (!pskb_pull(skb, hdroff))
4687                         return -EBADMSG;
4688
4689                 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff);
4690                 skb_reset_network_header(skb);
4691                 skb_reset_transport_header(skb);
4692                 skb->encapsulation = 0;
4693
4694                 bpf_compute_data_pointers(skb);
4695                 bpf_update_srh_state(skb);
4696                 return seg6_lookup_nexthop(skb, NULL, *(int *)param);
4697         case SEG6_LOCAL_ACTION_END_B6:
4698                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4699                         return -EBADMSG;
4700                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE,
4701                                           param, param_len);
4702                 if (!err)
4703                         bpf_update_srh_state(skb);
4704
4705                 return err;
4706         case SEG6_LOCAL_ACTION_END_B6_ENCAP:
4707                 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
4708                         return -EBADMSG;
4709                 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6,
4710                                           param, param_len);
4711                 if (!err)
4712                         bpf_update_srh_state(skb);
4713
4714                 return err;
4715         default:
4716                 return -EINVAL;
4717         }
4718 }
4719
4720 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = {
4721         .func           = bpf_lwt_seg6_action,
4722         .gpl_only       = false,
4723         .ret_type       = RET_INTEGER,
4724         .arg1_type      = ARG_PTR_TO_CTX,
4725         .arg2_type      = ARG_ANYTHING,
4726         .arg3_type      = ARG_PTR_TO_MEM,
4727         .arg4_type      = ARG_CONST_SIZE
4728 };
4729
4730 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
4731            s32, len)
4732 {
4733         struct seg6_bpf_srh_state *srh_state =
4734                 this_cpu_ptr(&seg6_bpf_srh_states);
4735         struct ipv6_sr_hdr *srh = srh_state->srh;
4736         void *srh_end, *srh_tlvs, *ptr;
4737         struct ipv6hdr *hdr;
4738         int srhoff = 0;
4739         int ret;
4740
4741         if (unlikely(srh == NULL))
4742                 return -EINVAL;
4743
4744         srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) +
4745                         ((srh->first_segment + 1) << 4));
4746         srh_end = (void *)((unsigned char *)srh + sizeof(*srh) +
4747                         srh_state->hdrlen);
4748         ptr = skb->data + offset;
4749
4750         if (unlikely(ptr < srh_tlvs || ptr > srh_end))
4751                 return -EFAULT;
4752         if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end))
4753                 return -EFAULT;
4754
4755         if (len > 0) {
4756                 ret = skb_cow_head(skb, len);
4757                 if (unlikely(ret < 0))
4758                         return ret;
4759
4760                 ret = bpf_skb_net_hdr_push(skb, offset, len);
4761         } else {
4762                 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len);
4763         }
4764
4765         bpf_compute_data_pointers(skb);
4766         if (unlikely(ret < 0))
4767                 return ret;
4768
4769         hdr = (struct ipv6hdr *)skb->data;
4770         hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr));
4771
4772         if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0)
4773                 return -EINVAL;
4774         srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff);
4775         srh_state->hdrlen += len;
4776         srh_state->valid = false;
4777         return 0;
4778 }
4779
4780 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = {
4781         .func           = bpf_lwt_seg6_adjust_srh,
4782         .gpl_only       = false,
4783         .ret_type       = RET_INTEGER,
4784         .arg1_type      = ARG_PTR_TO_CTX,
4785         .arg2_type      = ARG_ANYTHING,
4786         .arg3_type      = ARG_ANYTHING,
4787 };
4788 #endif /* CONFIG_IPV6_SEG6_BPF */
4789
4790 bool bpf_helper_changes_pkt_data(void *func)
4791 {
4792         if (func == bpf_skb_vlan_push ||
4793             func == bpf_skb_vlan_pop ||
4794             func == bpf_skb_store_bytes ||
4795             func == bpf_skb_change_proto ||
4796             func == bpf_skb_change_head ||
4797             func == sk_skb_change_head ||
4798             func == bpf_skb_change_tail ||
4799             func == sk_skb_change_tail ||
4800             func == bpf_skb_adjust_room ||
4801             func == bpf_skb_pull_data ||
4802             func == sk_skb_pull_data ||
4803             func == bpf_clone_redirect ||
4804             func == bpf_l3_csum_replace ||
4805             func == bpf_l4_csum_replace ||
4806             func == bpf_xdp_adjust_head ||
4807             func == bpf_xdp_adjust_meta ||
4808             func == bpf_msg_pull_data ||
4809             func == bpf_xdp_adjust_tail ||
4810 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
4811             func == bpf_lwt_seg6_store_bytes ||
4812             func == bpf_lwt_seg6_adjust_srh ||
4813             func == bpf_lwt_seg6_action ||
4814 #endif
4815             func == bpf_lwt_push_encap)
4816                 return true;
4817
4818         return false;
4819 }
4820
4821 static const struct bpf_func_proto *
4822 bpf_base_func_proto(enum bpf_func_id func_id)
4823 {
4824         switch (func_id) {
4825         case BPF_FUNC_map_lookup_elem:
4826                 return &bpf_map_lookup_elem_proto;
4827         case BPF_FUNC_map_update_elem:
4828                 return &bpf_map_update_elem_proto;
4829         case BPF_FUNC_map_delete_elem:
4830                 return &bpf_map_delete_elem_proto;
4831         case BPF_FUNC_get_prandom_u32:
4832                 return &bpf_get_prandom_u32_proto;
4833         case BPF_FUNC_get_smp_processor_id:
4834                 return &bpf_get_raw_smp_processor_id_proto;
4835         case BPF_FUNC_get_numa_node_id:
4836                 return &bpf_get_numa_node_id_proto;
4837         case BPF_FUNC_tail_call:
4838                 return &bpf_tail_call_proto;
4839         case BPF_FUNC_ktime_get_ns:
4840                 return &bpf_ktime_get_ns_proto;
4841         case BPF_FUNC_trace_printk:
4842                 if (capable(CAP_SYS_ADMIN))
4843                         return bpf_get_trace_printk_proto();
4844                 /* else: fall through */
4845         default:
4846                 return NULL;
4847         }
4848 }
4849
4850 static const struct bpf_func_proto *
4851 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4852 {
4853         switch (func_id) {
4854         /* inet and inet6 sockets are created in a process
4855          * context so there is always a valid uid/gid
4856          */
4857         case BPF_FUNC_get_current_uid_gid:
4858                 return &bpf_get_current_uid_gid_proto;
4859         case BPF_FUNC_get_local_storage:
4860                 return &bpf_get_local_storage_proto;
4861         default:
4862                 return bpf_base_func_proto(func_id);
4863         }
4864 }
4865
4866 static const struct bpf_func_proto *
4867 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4868 {
4869         switch (func_id) {
4870         /* inet and inet6 sockets are created in a process
4871          * context so there is always a valid uid/gid
4872          */
4873         case BPF_FUNC_get_current_uid_gid:
4874                 return &bpf_get_current_uid_gid_proto;
4875         case BPF_FUNC_bind:
4876                 switch (prog->expected_attach_type) {
4877                 case BPF_CGROUP_INET4_CONNECT:
4878                 case BPF_CGROUP_INET6_CONNECT:
4879                         return &bpf_bind_proto;
4880                 default:
4881                         return NULL;
4882                 }
4883         case BPF_FUNC_get_socket_cookie:
4884                 return &bpf_get_socket_cookie_sock_addr_proto;
4885         case BPF_FUNC_get_local_storage:
4886                 return &bpf_get_local_storage_proto;
4887         default:
4888                 return bpf_base_func_proto(func_id);
4889         }
4890 }
4891
4892 static const struct bpf_func_proto *
4893 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4894 {
4895         switch (func_id) {
4896         case BPF_FUNC_skb_load_bytes:
4897                 return &bpf_skb_load_bytes_proto;
4898         case BPF_FUNC_skb_load_bytes_relative:
4899                 return &bpf_skb_load_bytes_relative_proto;
4900         case BPF_FUNC_get_socket_cookie:
4901                 return &bpf_get_socket_cookie_proto;
4902         case BPF_FUNC_get_socket_uid:
4903                 return &bpf_get_socket_uid_proto;
4904         default:
4905                 return bpf_base_func_proto(func_id);
4906         }
4907 }
4908
4909 static const struct bpf_func_proto *
4910 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4911 {
4912         switch (func_id) {
4913         case BPF_FUNC_get_local_storage:
4914                 return &bpf_get_local_storage_proto;
4915         default:
4916                 return sk_filter_func_proto(func_id, prog);
4917         }
4918 }
4919
4920 static const struct bpf_func_proto *
4921 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
4922 {
4923         switch (func_id) {
4924         case BPF_FUNC_skb_store_bytes:
4925                 return &bpf_skb_store_bytes_proto;
4926         case BPF_FUNC_skb_load_bytes:
4927                 return &bpf_skb_load_bytes_proto;
4928         case BPF_FUNC_skb_load_bytes_relative:
4929                 return &bpf_skb_load_bytes_relative_proto;
4930         case BPF_FUNC_skb_pull_data:
4931                 return &bpf_skb_pull_data_proto;
4932         case BPF_FUNC_csum_diff:
4933                 return &bpf_csum_diff_proto;
4934         case BPF_FUNC_csum_update:
4935                 return &bpf_csum_update_proto;
4936         case BPF_FUNC_l3_csum_replace:
4937                 return &bpf_l3_csum_replace_proto;
4938         case BPF_FUNC_l4_csum_replace:
4939                 return &bpf_l4_csum_replace_proto;
4940         case BPF_FUNC_clone_redirect:
4941                 return &bpf_clone_redirect_proto;
4942         case BPF_FUNC_get_cgroup_classid:
4943                 return &bpf_get_cgroup_classid_proto;
4944         case BPF_FUNC_skb_vlan_push:
4945                 return &bpf_skb_vlan_push_proto;
4946         case BPF_FUNC_skb_vlan_pop:
4947                 return &bpf_skb_vlan_pop_proto;
4948         case BPF_FUNC_skb_change_proto:
4949                 return &bpf_skb_change_proto_proto;
4950         case BPF_FUNC_skb_change_type:
4951                 return &bpf_skb_change_type_proto;
4952         case BPF_FUNC_skb_adjust_room:
4953                 return &bpf_skb_adjust_room_proto;
4954         case BPF_FUNC_skb_change_tail:
4955                 return &bpf_skb_change_tail_proto;
4956         case BPF_FUNC_skb_get_tunnel_key:
4957                 return &bpf_skb_get_tunnel_key_proto;
4958         case BPF_FUNC_skb_set_tunnel_key:
4959                 return bpf_get_skb_set_tunnel_proto(func_id);
4960         case BPF_FUNC_skb_get_tunnel_opt:
4961                 return &bpf_skb_get_tunnel_opt_proto;
4962         case BPF_FUNC_skb_set_tunnel_opt:
4963                 return bpf_get_skb_set_tunnel_proto(func_id);
4964         case BPF_FUNC_redirect:
4965                 return &bpf_redirect_proto;
4966         case BPF_FUNC_get_route_realm:
4967                 return &bpf_get_route_realm_proto;
4968         case BPF_FUNC_get_hash_recalc:
4969                 return &bpf_get_hash_recalc_proto;
4970         case BPF_FUNC_set_hash_invalid:
4971                 return &bpf_set_hash_invalid_proto;
4972         case BPF_FUNC_set_hash:
4973                 return &bpf_set_hash_proto;
4974         case BPF_FUNC_perf_event_output:
4975                 return &bpf_skb_event_output_proto;
4976         case BPF_FUNC_get_smp_processor_id:
4977                 return &bpf_get_smp_processor_id_proto;
4978         case BPF_FUNC_skb_under_cgroup:
4979                 return &bpf_skb_under_cgroup_proto;
4980         case BPF_FUNC_get_socket_cookie:
4981                 return &bpf_get_socket_cookie_proto;
4982         case BPF_FUNC_get_socket_uid:
4983                 return &bpf_get_socket_uid_proto;
4984         case BPF_FUNC_fib_lookup:
4985                 return &bpf_skb_fib_lookup_proto;
4986 #ifdef CONFIG_XFRM
4987         case BPF_FUNC_skb_get_xfrm_state:
4988                 return &bpf_skb_get_xfrm_state_proto;
4989 #endif
4990 #ifdef CONFIG_SOCK_CGROUP_DATA
4991         case BPF_FUNC_skb_cgroup_id:
4992                 return &bpf_skb_cgroup_id_proto;
4993         case BPF_FUNC_skb_ancestor_cgroup_id:
4994                 return &bpf_skb_ancestor_cgroup_id_proto;
4995 #endif
4996         default:
4997                 return bpf_base_func_proto(func_id);
4998         }
4999 }
5000
5001 static const struct bpf_func_proto *
5002 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5003 {
5004         switch (func_id) {
5005         case BPF_FUNC_perf_event_output:
5006                 return &bpf_xdp_event_output_proto;
5007         case BPF_FUNC_get_smp_processor_id:
5008                 return &bpf_get_smp_processor_id_proto;
5009         case BPF_FUNC_csum_diff:
5010                 return &bpf_csum_diff_proto;
5011         case BPF_FUNC_xdp_adjust_head:
5012                 return &bpf_xdp_adjust_head_proto;
5013         case BPF_FUNC_xdp_adjust_meta:
5014                 return &bpf_xdp_adjust_meta_proto;
5015         case BPF_FUNC_redirect:
5016                 return &bpf_xdp_redirect_proto;
5017         case BPF_FUNC_redirect_map:
5018                 return &bpf_xdp_redirect_map_proto;
5019         case BPF_FUNC_xdp_adjust_tail:
5020                 return &bpf_xdp_adjust_tail_proto;
5021         case BPF_FUNC_fib_lookup:
5022                 return &bpf_xdp_fib_lookup_proto;
5023         default:
5024                 return bpf_base_func_proto(func_id);
5025         }
5026 }
5027
5028 static const struct bpf_func_proto *
5029 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5030 {
5031         switch (func_id) {
5032         case BPF_FUNC_setsockopt:
5033                 return &bpf_setsockopt_proto;
5034         case BPF_FUNC_getsockopt:
5035                 return &bpf_getsockopt_proto;
5036         case BPF_FUNC_sock_ops_cb_flags_set:
5037                 return &bpf_sock_ops_cb_flags_set_proto;
5038         case BPF_FUNC_sock_map_update:
5039                 return &bpf_sock_map_update_proto;
5040         case BPF_FUNC_sock_hash_update:
5041                 return &bpf_sock_hash_update_proto;
5042         case BPF_FUNC_get_socket_cookie:
5043                 return &bpf_get_socket_cookie_sock_ops_proto;
5044         case BPF_FUNC_get_local_storage:
5045                 return &bpf_get_local_storage_proto;
5046         default:
5047                 return bpf_base_func_proto(func_id);
5048         }
5049 }
5050
5051 static const struct bpf_func_proto *
5052 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5053 {
5054         switch (func_id) {
5055         case BPF_FUNC_msg_redirect_map:
5056                 return &bpf_msg_redirect_map_proto;
5057         case BPF_FUNC_msg_redirect_hash:
5058                 return &bpf_msg_redirect_hash_proto;
5059         case BPF_FUNC_msg_apply_bytes:
5060                 return &bpf_msg_apply_bytes_proto;
5061         case BPF_FUNC_msg_cork_bytes:
5062                 return &bpf_msg_cork_bytes_proto;
5063         case BPF_FUNC_msg_pull_data:
5064                 return &bpf_msg_pull_data_proto;
5065         case BPF_FUNC_get_local_storage:
5066                 return &bpf_get_local_storage_proto;
5067         default:
5068                 return bpf_base_func_proto(func_id);
5069         }
5070 }
5071
5072 static const struct bpf_func_proto *
5073 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5074 {
5075         switch (func_id) {
5076         case BPF_FUNC_skb_store_bytes:
5077                 return &bpf_skb_store_bytes_proto;
5078         case BPF_FUNC_skb_load_bytes:
5079                 return &bpf_skb_load_bytes_proto;
5080         case BPF_FUNC_skb_pull_data:
5081                 return &sk_skb_pull_data_proto;
5082         case BPF_FUNC_skb_change_tail:
5083                 return &sk_skb_change_tail_proto;
5084         case BPF_FUNC_skb_change_head:
5085                 return &sk_skb_change_head_proto;
5086         case BPF_FUNC_get_socket_cookie:
5087                 return &bpf_get_socket_cookie_proto;
5088         case BPF_FUNC_get_socket_uid:
5089                 return &bpf_get_socket_uid_proto;
5090         case BPF_FUNC_sk_redirect_map:
5091                 return &bpf_sk_redirect_map_proto;
5092         case BPF_FUNC_sk_redirect_hash:
5093                 return &bpf_sk_redirect_hash_proto;
5094         case BPF_FUNC_get_local_storage:
5095                 return &bpf_get_local_storage_proto;
5096         default:
5097                 return bpf_base_func_proto(func_id);
5098         }
5099 }
5100
5101 static const struct bpf_func_proto *
5102 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5103 {
5104         switch (func_id) {
5105         case BPF_FUNC_skb_load_bytes:
5106                 return &bpf_skb_load_bytes_proto;
5107         case BPF_FUNC_skb_pull_data:
5108                 return &bpf_skb_pull_data_proto;
5109         case BPF_FUNC_csum_diff:
5110                 return &bpf_csum_diff_proto;
5111         case BPF_FUNC_get_cgroup_classid:
5112                 return &bpf_get_cgroup_classid_proto;
5113         case BPF_FUNC_get_route_realm:
5114                 return &bpf_get_route_realm_proto;
5115         case BPF_FUNC_get_hash_recalc:
5116                 return &bpf_get_hash_recalc_proto;
5117         case BPF_FUNC_perf_event_output:
5118                 return &bpf_skb_event_output_proto;
5119         case BPF_FUNC_get_smp_processor_id:
5120                 return &bpf_get_smp_processor_id_proto;
5121         case BPF_FUNC_skb_under_cgroup:
5122                 return &bpf_skb_under_cgroup_proto;
5123         default:
5124                 return bpf_base_func_proto(func_id);
5125         }
5126 }
5127
5128 static const struct bpf_func_proto *
5129 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5130 {
5131         switch (func_id) {
5132         case BPF_FUNC_lwt_push_encap:
5133                 return &bpf_lwt_push_encap_proto;
5134         default:
5135                 return lwt_out_func_proto(func_id, prog);
5136         }
5137 }
5138
5139 static const struct bpf_func_proto *
5140 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5141 {
5142         switch (func_id) {
5143         case BPF_FUNC_skb_get_tunnel_key:
5144                 return &bpf_skb_get_tunnel_key_proto;
5145         case BPF_FUNC_skb_set_tunnel_key:
5146                 return bpf_get_skb_set_tunnel_proto(func_id);
5147         case BPF_FUNC_skb_get_tunnel_opt:
5148                 return &bpf_skb_get_tunnel_opt_proto;
5149         case BPF_FUNC_skb_set_tunnel_opt:
5150                 return bpf_get_skb_set_tunnel_proto(func_id);
5151         case BPF_FUNC_redirect:
5152                 return &bpf_redirect_proto;
5153         case BPF_FUNC_clone_redirect:
5154                 return &bpf_clone_redirect_proto;
5155         case BPF_FUNC_skb_change_tail:
5156                 return &bpf_skb_change_tail_proto;
5157         case BPF_FUNC_skb_change_head:
5158                 return &bpf_skb_change_head_proto;
5159         case BPF_FUNC_skb_store_bytes:
5160                 return &bpf_skb_store_bytes_proto;
5161         case BPF_FUNC_csum_update:
5162                 return &bpf_csum_update_proto;
5163         case BPF_FUNC_l3_csum_replace:
5164                 return &bpf_l3_csum_replace_proto;
5165         case BPF_FUNC_l4_csum_replace:
5166                 return &bpf_l4_csum_replace_proto;
5167         case BPF_FUNC_set_hash_invalid:
5168                 return &bpf_set_hash_invalid_proto;
5169         default:
5170                 return lwt_out_func_proto(func_id, prog);
5171         }
5172 }
5173
5174 static const struct bpf_func_proto *
5175 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
5176 {
5177         switch (func_id) {
5178 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF)
5179         case BPF_FUNC_lwt_seg6_store_bytes:
5180                 return &bpf_lwt_seg6_store_bytes_proto;
5181         case BPF_FUNC_lwt_seg6_action:
5182                 return &bpf_lwt_seg6_action_proto;
5183         case BPF_FUNC_lwt_seg6_adjust_srh:
5184                 return &bpf_lwt_seg6_adjust_srh_proto;
5185 #endif
5186         default:
5187                 return lwt_out_func_proto(func_id, prog);
5188         }
5189 }
5190
5191 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type,
5192                                     const struct bpf_prog *prog,
5193                                     struct bpf_insn_access_aux *info)
5194 {
5195         const int size_default = sizeof(__u32);
5196
5197         if (off < 0 || off >= sizeof(struct __sk_buff))
5198                 return false;
5199
5200         /* The verifier guarantees that size > 0. */
5201         if (off % size != 0)
5202                 return false;
5203
5204         switch (off) {
5205         case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5206                 if (off + size > offsetofend(struct __sk_buff, cb[4]))
5207                         return false;
5208                 break;
5209         case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]):
5210         case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]):
5211         case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4):
5212         case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4):
5213         case bpf_ctx_range(struct __sk_buff, data):
5214         case bpf_ctx_range(struct __sk_buff, data_meta):
5215         case bpf_ctx_range(struct __sk_buff, data_end):
5216                 if (size != size_default)
5217                         return false;
5218                 break;
5219         default:
5220                 /* Only narrow read access allowed for now. */
5221                 if (type == BPF_WRITE) {
5222                         if (size != size_default)
5223                                 return false;
5224                 } else {
5225                         bpf_ctx_record_field_size(info, size_default);
5226                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5227                                 return false;
5228                 }
5229         }
5230
5231         return true;
5232 }
5233
5234 static bool sk_filter_is_valid_access(int off, int size,
5235                                       enum bpf_access_type type,
5236                                       const struct bpf_prog *prog,
5237                                       struct bpf_insn_access_aux *info)
5238 {
5239         switch (off) {
5240         case bpf_ctx_range(struct __sk_buff, tc_classid):
5241         case bpf_ctx_range(struct __sk_buff, data):
5242         case bpf_ctx_range(struct __sk_buff, data_meta):
5243         case bpf_ctx_range(struct __sk_buff, data_end):
5244         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5245                 return false;
5246         }
5247
5248         if (type == BPF_WRITE) {
5249                 switch (off) {
5250                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5251                         break;
5252                 default:
5253                         return false;
5254                 }
5255         }
5256
5257         return bpf_skb_is_valid_access(off, size, type, prog, info);
5258 }
5259
5260 static bool lwt_is_valid_access(int off, int size,
5261                                 enum bpf_access_type type,
5262                                 const struct bpf_prog *prog,
5263                                 struct bpf_insn_access_aux *info)
5264 {
5265         switch (off) {
5266         case bpf_ctx_range(struct __sk_buff, tc_classid):
5267         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5268         case bpf_ctx_range(struct __sk_buff, data_meta):
5269                 return false;
5270         }
5271
5272         if (type == BPF_WRITE) {
5273                 switch (off) {
5274                 case bpf_ctx_range(struct __sk_buff, mark):
5275                 case bpf_ctx_range(struct __sk_buff, priority):
5276                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5277                         break;
5278                 default:
5279                         return false;
5280                 }
5281         }
5282
5283         switch (off) {
5284         case bpf_ctx_range(struct __sk_buff, data):
5285                 info->reg_type = PTR_TO_PACKET;
5286                 break;
5287         case bpf_ctx_range(struct __sk_buff, data_end):
5288                 info->reg_type = PTR_TO_PACKET_END;
5289                 break;
5290         }
5291
5292         return bpf_skb_is_valid_access(off, size, type, prog, info);
5293 }
5294
5295 /* Attach type specific accesses */
5296 static bool __sock_filter_check_attach_type(int off,
5297                                             enum bpf_access_type access_type,
5298                                             enum bpf_attach_type attach_type)
5299 {
5300         switch (off) {
5301         case offsetof(struct bpf_sock, bound_dev_if):
5302         case offsetof(struct bpf_sock, mark):
5303         case offsetof(struct bpf_sock, priority):
5304                 switch (attach_type) {
5305                 case BPF_CGROUP_INET_SOCK_CREATE:
5306                         goto full_access;
5307                 default:
5308                         return false;
5309                 }
5310         case bpf_ctx_range(struct bpf_sock, src_ip4):
5311                 switch (attach_type) {
5312                 case BPF_CGROUP_INET4_POST_BIND:
5313                         goto read_only;
5314                 default:
5315                         return false;
5316                 }
5317         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5318                 switch (attach_type) {
5319                 case BPF_CGROUP_INET6_POST_BIND:
5320                         goto read_only;
5321                 default:
5322                         return false;
5323                 }
5324         case bpf_ctx_range(struct bpf_sock, src_port):
5325                 switch (attach_type) {
5326                 case BPF_CGROUP_INET4_POST_BIND:
5327                 case BPF_CGROUP_INET6_POST_BIND:
5328                         goto read_only;
5329                 default:
5330                         return false;
5331                 }
5332         }
5333 read_only:
5334         return access_type == BPF_READ;
5335 full_access:
5336         return true;
5337 }
5338
5339 static bool __sock_filter_check_size(int off, int size,
5340                                      struct bpf_insn_access_aux *info)
5341 {
5342         const int size_default = sizeof(__u32);
5343
5344         switch (off) {
5345         case bpf_ctx_range(struct bpf_sock, src_ip4):
5346         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
5347                 bpf_ctx_record_field_size(info, size_default);
5348                 return bpf_ctx_narrow_access_ok(off, size, size_default);
5349         }
5350
5351         return size == size_default;
5352 }
5353
5354 static bool sock_filter_is_valid_access(int off, int size,
5355                                         enum bpf_access_type type,
5356                                         const struct bpf_prog *prog,
5357                                         struct bpf_insn_access_aux *info)
5358 {
5359         if (off < 0 || off >= sizeof(struct bpf_sock))
5360                 return false;
5361         if (off % size != 0)
5362                 return false;
5363         if (!__sock_filter_check_attach_type(off, type,
5364                                              prog->expected_attach_type))
5365                 return false;
5366         if (!__sock_filter_check_size(off, size, info))
5367                 return false;
5368         return true;
5369 }
5370
5371 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write,
5372                                 const struct bpf_prog *prog, int drop_verdict)
5373 {
5374         struct bpf_insn *insn = insn_buf;
5375
5376         if (!direct_write)
5377                 return 0;
5378
5379         /* if (!skb->cloned)
5380          *       goto start;
5381          *
5382          * (Fast-path, otherwise approximation that we might be
5383          *  a clone, do the rest in helper.)
5384          */
5385         *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET());
5386         *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK);
5387         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7);
5388
5389         /* ret = bpf_skb_pull_data(skb, 0); */
5390         *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
5391         *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2);
5392         *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0,
5393                                BPF_FUNC_skb_pull_data);
5394         /* if (!ret)
5395          *      goto restore;
5396          * return TC_ACT_SHOT;
5397          */
5398         *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2);
5399         *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict);
5400         *insn++ = BPF_EXIT_INSN();
5401
5402         /* restore: */
5403         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
5404         /* start: */
5405         *insn++ = prog->insnsi[0];
5406
5407         return insn - insn_buf;
5408 }
5409
5410 static int bpf_gen_ld_abs(const struct bpf_insn *orig,
5411                           struct bpf_insn *insn_buf)
5412 {
5413         bool indirect = BPF_MODE(orig->code) == BPF_IND;
5414         struct bpf_insn *insn = insn_buf;
5415
5416         /* We're guaranteed here that CTX is in R6. */
5417         *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX);
5418         if (!indirect) {
5419                 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm);
5420         } else {
5421                 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg);
5422                 if (orig->imm)
5423                         *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm);
5424         }
5425
5426         switch (BPF_SIZE(orig->code)) {
5427         case BPF_B:
5428                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache);
5429                 break;
5430         case BPF_H:
5431                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache);
5432                 break;
5433         case BPF_W:
5434                 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache);
5435                 break;
5436         }
5437
5438         *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2);
5439         *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0);
5440         *insn++ = BPF_EXIT_INSN();
5441
5442         return insn - insn_buf;
5443 }
5444
5445 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write,
5446                                const struct bpf_prog *prog)
5447 {
5448         return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT);
5449 }
5450
5451 static bool tc_cls_act_is_valid_access(int off, int size,
5452                                        enum bpf_access_type type,
5453                                        const struct bpf_prog *prog,
5454                                        struct bpf_insn_access_aux *info)
5455 {
5456         if (type == BPF_WRITE) {
5457                 switch (off) {
5458                 case bpf_ctx_range(struct __sk_buff, mark):
5459                 case bpf_ctx_range(struct __sk_buff, tc_index):
5460                 case bpf_ctx_range(struct __sk_buff, priority):
5461                 case bpf_ctx_range(struct __sk_buff, tc_classid):
5462                 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]):
5463                         break;
5464                 default:
5465                         return false;
5466                 }
5467         }
5468
5469         switch (off) {
5470         case bpf_ctx_range(struct __sk_buff, data):
5471                 info->reg_type = PTR_TO_PACKET;
5472                 break;
5473         case bpf_ctx_range(struct __sk_buff, data_meta):
5474                 info->reg_type = PTR_TO_PACKET_META;
5475                 break;
5476         case bpf_ctx_range(struct __sk_buff, data_end):
5477                 info->reg_type = PTR_TO_PACKET_END;
5478                 break;
5479         case bpf_ctx_range_till(struct __sk_buff, family, local_port):
5480                 return false;
5481         }
5482
5483         return bpf_skb_is_valid_access(off, size, type, prog, info);
5484 }
5485
5486 static bool __is_valid_xdp_access(int off, int size)
5487 {
5488         if (off < 0 || off >= sizeof(struct xdp_md))
5489                 return false;
5490         if (off % size != 0)
5491                 return false;
5492         if (size != sizeof(__u32))
5493                 return false;
5494
5495         return true;
5496 }
5497
5498 static bool xdp_is_valid_access(int off, int size,
5499                                 enum bpf_access_type type,
5500                                 const struct bpf_prog *prog,
5501                                 struct bpf_insn_access_aux *info)
5502 {
5503         if (type == BPF_WRITE) {
5504                 if (bpf_prog_is_dev_bound(prog->aux)) {
5505                         switch (off) {
5506                         case offsetof(struct xdp_md, rx_queue_index):
5507                                 return __is_valid_xdp_access(off, size);
5508                         }
5509                 }
5510                 return false;
5511         }
5512
5513         switch (off) {
5514         case offsetof(struct xdp_md, data):
5515                 info->reg_type = PTR_TO_PACKET;
5516                 break;
5517         case offsetof(struct xdp_md, data_meta):
5518                 info->reg_type = PTR_TO_PACKET_META;
5519                 break;
5520         case offsetof(struct xdp_md, data_end):
5521                 info->reg_type = PTR_TO_PACKET_END;
5522                 break;
5523         }
5524
5525         return __is_valid_xdp_access(off, size);
5526 }
5527
5528 void bpf_warn_invalid_xdp_action(u32 act)
5529 {
5530         const u32 act_max = XDP_REDIRECT;
5531
5532         WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n",
5533                   act > act_max ? "Illegal" : "Driver unsupported",
5534                   act);
5535 }
5536 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action);
5537
5538 static bool sock_addr_is_valid_access(int off, int size,
5539                                       enum bpf_access_type type,
5540                                       const struct bpf_prog *prog,
5541                                       struct bpf_insn_access_aux *info)
5542 {
5543         const int size_default = sizeof(__u32);
5544
5545         if (off < 0 || off >= sizeof(struct bpf_sock_addr))
5546                 return false;
5547         if (off % size != 0)
5548                 return false;
5549
5550         /* Disallow access to IPv6 fields from IPv4 contex and vise
5551          * versa.
5552          */
5553         switch (off) {
5554         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5555                 switch (prog->expected_attach_type) {
5556                 case BPF_CGROUP_INET4_BIND:
5557                 case BPF_CGROUP_INET4_CONNECT:
5558                 case BPF_CGROUP_UDP4_SENDMSG:
5559                         break;
5560                 default:
5561                         return false;
5562                 }
5563                 break;
5564         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5565                 switch (prog->expected_attach_type) {
5566                 case BPF_CGROUP_INET6_BIND:
5567                 case BPF_CGROUP_INET6_CONNECT:
5568                 case BPF_CGROUP_UDP6_SENDMSG:
5569                         break;
5570                 default:
5571                         return false;
5572                 }
5573                 break;
5574         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5575                 switch (prog->expected_attach_type) {
5576                 case BPF_CGROUP_UDP4_SENDMSG:
5577                         break;
5578                 default:
5579                         return false;
5580                 }
5581                 break;
5582         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5583                                 msg_src_ip6[3]):
5584                 switch (prog->expected_attach_type) {
5585                 case BPF_CGROUP_UDP6_SENDMSG:
5586                         break;
5587                 default:
5588                         return false;
5589                 }
5590                 break;
5591         }
5592
5593         switch (off) {
5594         case bpf_ctx_range(struct bpf_sock_addr, user_ip4):
5595         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
5596         case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4):
5597         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
5598                                 msg_src_ip6[3]):
5599                 /* Only narrow read access allowed for now. */
5600                 if (type == BPF_READ) {
5601                         bpf_ctx_record_field_size(info, size_default);
5602                         if (!bpf_ctx_narrow_access_ok(off, size, size_default))
5603                                 return false;
5604                 } else {
5605                         if (size != size_default)
5606                                 return false;
5607                 }
5608                 break;
5609         case bpf_ctx_range(struct bpf_sock_addr, user_port):
5610                 if (size != size_default)
5611                         return false;
5612                 break;
5613         default:
5614                 if (type == BPF_READ) {
5615                         if (size != size_default)
5616                                 return false;
5617                 } else {
5618                         return false;
5619                 }
5620         }
5621
5622         return true;
5623 }
5624
5625 static bool sock_ops_is_valid_access(int off, int size,
5626                                      enum bpf_access_type type,
5627                                      const struct bpf_prog *prog,
5628                                      struct bpf_insn_access_aux *info)
5629 {
5630         const int size_default = sizeof(__u32);
5631
5632         if (off < 0 || off >= sizeof(struct bpf_sock_ops))
5633                 return false;
5634
5635         /* The verifier guarantees that size > 0. */
5636         if (off % size != 0)
5637                 return false;
5638
5639         if (type == BPF_WRITE) {
5640                 switch (off) {
5641                 case offsetof(struct bpf_sock_ops, reply):
5642                 case offsetof(struct bpf_sock_ops, sk_txhash):
5643                         if (size != size_default)
5644                                 return false;
5645                         break;
5646                 default:
5647                         return false;
5648                 }
5649         } else {
5650                 switch (off) {
5651                 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received,
5652                                         bytes_acked):
5653                         if (size != sizeof(__u64))
5654                                 return false;
5655                         break;
5656                 default:
5657                         if (size != size_default)
5658                                 return false;
5659                         break;
5660                 }
5661         }
5662
5663         return true;
5664 }
5665
5666 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write,
5667                            const struct bpf_prog *prog)
5668 {
5669         return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP);
5670 }
5671
5672 static bool sk_skb_is_valid_access(int off, int size,
5673                                    enum bpf_access_type type,
5674                                    const struct bpf_prog *prog,
5675                                    struct bpf_insn_access_aux *info)
5676 {
5677         switch (off) {
5678         case bpf_ctx_range(struct __sk_buff, tc_classid):
5679         case bpf_ctx_range(struct __sk_buff, data_meta):
5680                 return false;
5681         }
5682
5683         if (type == BPF_WRITE) {
5684                 switch (off) {
5685                 case bpf_ctx_range(struct __sk_buff, tc_index):
5686                 case bpf_ctx_range(struct __sk_buff, priority):
5687                         break;
5688                 default:
5689                         return false;
5690                 }
5691         }
5692
5693         switch (off) {
5694         case bpf_ctx_range(struct __sk_buff, mark):
5695                 return false;
5696         case bpf_ctx_range(struct __sk_buff, data):
5697                 info->reg_type = PTR_TO_PACKET;
5698                 break;
5699         case bpf_ctx_range(struct __sk_buff, data_end):
5700                 info->reg_type = PTR_TO_PACKET_END;
5701                 break;
5702         }
5703
5704         return bpf_skb_is_valid_access(off, size, type, prog, info);
5705 }
5706
5707 static bool sk_msg_is_valid_access(int off, int size,
5708                                    enum bpf_access_type type,
5709                                    const struct bpf_prog *prog,
5710                                    struct bpf_insn_access_aux *info)
5711 {
5712         if (type == BPF_WRITE)
5713                 return false;
5714
5715         switch (off) {
5716         case offsetof(struct sk_msg_md, data):
5717                 info->reg_type = PTR_TO_PACKET;
5718                 if (size != sizeof(__u64))
5719                         return false;
5720                 break;
5721         case offsetof(struct sk_msg_md, data_end):
5722                 info->reg_type = PTR_TO_PACKET_END;
5723                 if (size != sizeof(__u64))
5724                         return false;
5725                 break;
5726         default:
5727                 if (size != sizeof(__u32))
5728                         return false;
5729         }
5730
5731         if (off < 0 || off >= sizeof(struct sk_msg_md))
5732                 return false;
5733         if (off % size != 0)
5734                 return false;
5735
5736         return true;
5737 }
5738
5739 static u32 bpf_convert_ctx_access(enum bpf_access_type type,
5740                                   const struct bpf_insn *si,
5741                                   struct bpf_insn *insn_buf,
5742                                   struct bpf_prog *prog, u32 *target_size)
5743 {
5744         struct bpf_insn *insn = insn_buf;
5745         int off;
5746
5747         switch (si->off) {
5748         case offsetof(struct __sk_buff, len):
5749                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5750                                       bpf_target_off(struct sk_buff, len, 4,
5751                                                      target_size));
5752                 break;
5753
5754         case offsetof(struct __sk_buff, protocol):
5755                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5756                                       bpf_target_off(struct sk_buff, protocol, 2,
5757                                                      target_size));
5758                 break;
5759
5760         case offsetof(struct __sk_buff, vlan_proto):
5761                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5762                                       bpf_target_off(struct sk_buff, vlan_proto, 2,
5763                                                      target_size));
5764                 break;
5765
5766         case offsetof(struct __sk_buff, priority):
5767                 if (type == BPF_WRITE)
5768                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5769                                               bpf_target_off(struct sk_buff, priority, 4,
5770                                                              target_size));
5771                 else
5772                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5773                                               bpf_target_off(struct sk_buff, priority, 4,
5774                                                              target_size));
5775                 break;
5776
5777         case offsetof(struct __sk_buff, ingress_ifindex):
5778                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5779                                       bpf_target_off(struct sk_buff, skb_iif, 4,
5780                                                      target_size));
5781                 break;
5782
5783         case offsetof(struct __sk_buff, ifindex):
5784                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
5785                                       si->dst_reg, si->src_reg,
5786                                       offsetof(struct sk_buff, dev));
5787                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1);
5788                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5789                                       bpf_target_off(struct net_device, ifindex, 4,
5790                                                      target_size));
5791                 break;
5792
5793         case offsetof(struct __sk_buff, hash):
5794                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5795                                       bpf_target_off(struct sk_buff, hash, 4,
5796                                                      target_size));
5797                 break;
5798
5799         case offsetof(struct __sk_buff, mark):
5800                 if (type == BPF_WRITE)
5801                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
5802                                               bpf_target_off(struct sk_buff, mark, 4,
5803                                                              target_size));
5804                 else
5805                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5806                                               bpf_target_off(struct sk_buff, mark, 4,
5807                                                              target_size));
5808                 break;
5809
5810         case offsetof(struct __sk_buff, pkt_type):
5811                 *target_size = 1;
5812                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg,
5813                                       PKT_TYPE_OFFSET());
5814                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX);
5815 #ifdef __BIG_ENDIAN_BITFIELD
5816                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5);
5817 #endif
5818                 break;
5819
5820         case offsetof(struct __sk_buff, queue_mapping):
5821                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5822                                       bpf_target_off(struct sk_buff, queue_mapping, 2,
5823                                                      target_size));
5824                 break;
5825
5826         case offsetof(struct __sk_buff, vlan_present):
5827         case offsetof(struct __sk_buff, vlan_tci):
5828                 BUILD_BUG_ON(VLAN_TAG_PRESENT != 0x1000);
5829
5830                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5831                                       bpf_target_off(struct sk_buff, vlan_tci, 2,
5832                                                      target_size));
5833                 if (si->off == offsetof(struct __sk_buff, vlan_tci)) {
5834                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg,
5835                                                 ~VLAN_TAG_PRESENT);
5836                 } else {
5837                         *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 12);
5838                         *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1);
5839                 }
5840                 break;
5841
5842         case offsetof(struct __sk_buff, cb[0]) ...
5843              offsetofend(struct __sk_buff, cb[4]) - 1:
5844                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20);
5845                 BUILD_BUG_ON((offsetof(struct sk_buff, cb) +
5846                               offsetof(struct qdisc_skb_cb, data)) %
5847                              sizeof(__u64));
5848
5849                 prog->cb_access = 1;
5850                 off  = si->off;
5851                 off -= offsetof(struct __sk_buff, cb[0]);
5852                 off += offsetof(struct sk_buff, cb);
5853                 off += offsetof(struct qdisc_skb_cb, data);
5854                 if (type == BPF_WRITE)
5855                         *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg,
5856                                               si->src_reg, off);
5857                 else
5858                         *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg,
5859                                               si->src_reg, off);
5860                 break;
5861
5862         case offsetof(struct __sk_buff, tc_classid):
5863                 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2);
5864
5865                 off  = si->off;
5866                 off -= offsetof(struct __sk_buff, tc_classid);
5867                 off += offsetof(struct sk_buff, cb);
5868                 off += offsetof(struct qdisc_skb_cb, tc_classid);
5869                 *target_size = 2;
5870                 if (type == BPF_WRITE)
5871                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg,
5872                                               si->src_reg, off);
5873                 else
5874                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg,
5875                                               si->src_reg, off);
5876                 break;
5877
5878         case offsetof(struct __sk_buff, data):
5879                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data),
5880                                       si->dst_reg, si->src_reg,
5881                                       offsetof(struct sk_buff, data));
5882                 break;
5883
5884         case offsetof(struct __sk_buff, data_meta):
5885                 off  = si->off;
5886                 off -= offsetof(struct __sk_buff, data_meta);
5887                 off += offsetof(struct sk_buff, cb);
5888                 off += offsetof(struct bpf_skb_data_end, data_meta);
5889                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5890                                       si->src_reg, off);
5891                 break;
5892
5893         case offsetof(struct __sk_buff, data_end):
5894                 off  = si->off;
5895                 off -= offsetof(struct __sk_buff, data_end);
5896                 off += offsetof(struct sk_buff, cb);
5897                 off += offsetof(struct bpf_skb_data_end, data_end);
5898                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
5899                                       si->src_reg, off);
5900                 break;
5901
5902         case offsetof(struct __sk_buff, tc_index):
5903 #ifdef CONFIG_NET_SCHED
5904                 if (type == BPF_WRITE)
5905                         *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg,
5906                                               bpf_target_off(struct sk_buff, tc_index, 2,
5907                                                              target_size));
5908                 else
5909                         *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
5910                                               bpf_target_off(struct sk_buff, tc_index, 2,
5911                                                              target_size));
5912 #else
5913                 *target_size = 2;
5914                 if (type == BPF_WRITE)
5915                         *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg);
5916                 else
5917                         *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5918 #endif
5919                 break;
5920
5921         case offsetof(struct __sk_buff, napi_id):
5922 #if defined(CONFIG_NET_RX_BUSY_POLL)
5923                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
5924                                       bpf_target_off(struct sk_buff, napi_id, 4,
5925                                                      target_size));
5926                 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1);
5927                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5928 #else
5929                 *target_size = 4;
5930                 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0);
5931 #endif
5932                 break;
5933         case offsetof(struct __sk_buff, family):
5934                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
5935
5936                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5937                                       si->dst_reg, si->src_reg,
5938                                       offsetof(struct sk_buff, sk));
5939                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
5940                                       bpf_target_off(struct sock_common,
5941                                                      skc_family,
5942                                                      2, target_size));
5943                 break;
5944         case offsetof(struct __sk_buff, remote_ip4):
5945                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
5946
5947                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5948                                       si->dst_reg, si->src_reg,
5949                                       offsetof(struct sk_buff, sk));
5950                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5951                                       bpf_target_off(struct sock_common,
5952                                                      skc_daddr,
5953                                                      4, target_size));
5954                 break;
5955         case offsetof(struct __sk_buff, local_ip4):
5956                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5957                                           skc_rcv_saddr) != 4);
5958
5959                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5960                                       si->dst_reg, si->src_reg,
5961                                       offsetof(struct sk_buff, sk));
5962                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5963                                       bpf_target_off(struct sock_common,
5964                                                      skc_rcv_saddr,
5965                                                      4, target_size));
5966                 break;
5967         case offsetof(struct __sk_buff, remote_ip6[0]) ...
5968              offsetof(struct __sk_buff, remote_ip6[3]):
5969 #if IS_ENABLED(CONFIG_IPV6)
5970                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5971                                           skc_v6_daddr.s6_addr32[0]) != 4);
5972
5973                 off = si->off;
5974                 off -= offsetof(struct __sk_buff, remote_ip6[0]);
5975
5976                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5977                                       si->dst_reg, si->src_reg,
5978                                       offsetof(struct sk_buff, sk));
5979                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
5980                                       offsetof(struct sock_common,
5981                                                skc_v6_daddr.s6_addr32[0]) +
5982                                       off);
5983 #else
5984                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
5985 #endif
5986                 break;
5987         case offsetof(struct __sk_buff, local_ip6[0]) ...
5988              offsetof(struct __sk_buff, local_ip6[3]):
5989 #if IS_ENABLED(CONFIG_IPV6)
5990                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
5991                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
5992
5993                 off = si->off;
5994                 off -= offsetof(struct __sk_buff, local_ip6[0]);
5995
5996                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
5997                                       si->dst_reg, si->src_reg,
5998                                       offsetof(struct sk_buff, sk));
5999                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6000                                       offsetof(struct sock_common,
6001                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6002                                       off);
6003 #else
6004                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6005 #endif
6006                 break;
6007
6008         case offsetof(struct __sk_buff, remote_port):
6009                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6010
6011                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6012                                       si->dst_reg, si->src_reg,
6013                                       offsetof(struct sk_buff, sk));
6014                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6015                                       bpf_target_off(struct sock_common,
6016                                                      skc_dport,
6017                                                      2, target_size));
6018 #ifndef __BIG_ENDIAN_BITFIELD
6019                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6020 #endif
6021                 break;
6022
6023         case offsetof(struct __sk_buff, local_port):
6024                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6025
6026                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk),
6027                                       si->dst_reg, si->src_reg,
6028                                       offsetof(struct sk_buff, sk));
6029                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6030                                       bpf_target_off(struct sock_common,
6031                                                      skc_num, 2, target_size));
6032                 break;
6033         }
6034
6035         return insn - insn_buf;
6036 }
6037
6038 static u32 sock_filter_convert_ctx_access(enum bpf_access_type type,
6039                                           const struct bpf_insn *si,
6040                                           struct bpf_insn *insn_buf,
6041                                           struct bpf_prog *prog, u32 *target_size)
6042 {
6043         struct bpf_insn *insn = insn_buf;
6044         int off;
6045
6046         switch (si->off) {
6047         case offsetof(struct bpf_sock, bound_dev_if):
6048                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4);
6049
6050                 if (type == BPF_WRITE)
6051                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6052                                         offsetof(struct sock, sk_bound_dev_if));
6053                 else
6054                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6055                                       offsetof(struct sock, sk_bound_dev_if));
6056                 break;
6057
6058         case offsetof(struct bpf_sock, mark):
6059                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4);
6060
6061                 if (type == BPF_WRITE)
6062                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6063                                         offsetof(struct sock, sk_mark));
6064                 else
6065                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6066                                       offsetof(struct sock, sk_mark));
6067                 break;
6068
6069         case offsetof(struct bpf_sock, priority):
6070                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4);
6071
6072                 if (type == BPF_WRITE)
6073                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6074                                         offsetof(struct sock, sk_priority));
6075                 else
6076                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6077                                       offsetof(struct sock, sk_priority));
6078                 break;
6079
6080         case offsetof(struct bpf_sock, family):
6081                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_family) != 2);
6082
6083                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg,
6084                                       offsetof(struct sock, sk_family));
6085                 break;
6086
6087         case offsetof(struct bpf_sock, type):
6088                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6089                                       offsetof(struct sock, __sk_flags_offset));
6090                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6091                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6092                 break;
6093
6094         case offsetof(struct bpf_sock, protocol):
6095                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6096                                       offsetof(struct sock, __sk_flags_offset));
6097                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6098                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT);
6099                 break;
6100
6101         case offsetof(struct bpf_sock, src_ip4):
6102                 *insn++ = BPF_LDX_MEM(
6103                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6104                         bpf_target_off(struct sock_common, skc_rcv_saddr,
6105                                        FIELD_SIZEOF(struct sock_common,
6106                                                     skc_rcv_saddr),
6107                                        target_size));
6108                 break;
6109
6110         case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]):
6111 #if IS_ENABLED(CONFIG_IPV6)
6112                 off = si->off;
6113                 off -= offsetof(struct bpf_sock, src_ip6[0]);
6114                 *insn++ = BPF_LDX_MEM(
6115                         BPF_SIZE(si->code), si->dst_reg, si->src_reg,
6116                         bpf_target_off(
6117                                 struct sock_common,
6118                                 skc_v6_rcv_saddr.s6_addr32[0],
6119                                 FIELD_SIZEOF(struct sock_common,
6120                                              skc_v6_rcv_saddr.s6_addr32[0]),
6121                                 target_size) + off);
6122 #else
6123                 (void)off;
6124                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6125 #endif
6126                 break;
6127
6128         case offsetof(struct bpf_sock, src_port):
6129                 *insn++ = BPF_LDX_MEM(
6130                         BPF_FIELD_SIZEOF(struct sock_common, skc_num),
6131                         si->dst_reg, si->src_reg,
6132                         bpf_target_off(struct sock_common, skc_num,
6133                                        FIELD_SIZEOF(struct sock_common,
6134                                                     skc_num),
6135                                        target_size));
6136                 break;
6137         }
6138
6139         return insn - insn_buf;
6140 }
6141
6142 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type,
6143                                          const struct bpf_insn *si,
6144                                          struct bpf_insn *insn_buf,
6145                                          struct bpf_prog *prog, u32 *target_size)
6146 {
6147         struct bpf_insn *insn = insn_buf;
6148
6149         switch (si->off) {
6150         case offsetof(struct __sk_buff, ifindex):
6151                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev),
6152                                       si->dst_reg, si->src_reg,
6153                                       offsetof(struct sk_buff, dev));
6154                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6155                                       bpf_target_off(struct net_device, ifindex, 4,
6156                                                      target_size));
6157                 break;
6158         default:
6159                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6160                                               target_size);
6161         }
6162
6163         return insn - insn_buf;
6164 }
6165
6166 static u32 xdp_convert_ctx_access(enum bpf_access_type type,
6167                                   const struct bpf_insn *si,
6168                                   struct bpf_insn *insn_buf,
6169                                   struct bpf_prog *prog, u32 *target_size)
6170 {
6171         struct bpf_insn *insn = insn_buf;
6172
6173         switch (si->off) {
6174         case offsetof(struct xdp_md, data):
6175                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data),
6176                                       si->dst_reg, si->src_reg,
6177                                       offsetof(struct xdp_buff, data));
6178                 break;
6179         case offsetof(struct xdp_md, data_meta):
6180                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta),
6181                                       si->dst_reg, si->src_reg,
6182                                       offsetof(struct xdp_buff, data_meta));
6183                 break;
6184         case offsetof(struct xdp_md, data_end):
6185                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end),
6186                                       si->dst_reg, si->src_reg,
6187                                       offsetof(struct xdp_buff, data_end));
6188                 break;
6189         case offsetof(struct xdp_md, ingress_ifindex):
6190                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6191                                       si->dst_reg, si->src_reg,
6192                                       offsetof(struct xdp_buff, rxq));
6193                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev),
6194                                       si->dst_reg, si->dst_reg,
6195                                       offsetof(struct xdp_rxq_info, dev));
6196                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6197                                       offsetof(struct net_device, ifindex));
6198                 break;
6199         case offsetof(struct xdp_md, rx_queue_index):
6200                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq),
6201                                       si->dst_reg, si->src_reg,
6202                                       offsetof(struct xdp_buff, rxq));
6203                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6204                                       offsetof(struct xdp_rxq_info,
6205                                                queue_index));
6206                 break;
6207         }
6208
6209         return insn - insn_buf;
6210 }
6211
6212 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of
6213  * context Structure, F is Field in context structure that contains a pointer
6214  * to Nested Structure of type NS that has the field NF.
6215  *
6216  * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make
6217  * sure that SIZE is not greater than actual size of S.F.NF.
6218  *
6219  * If offset OFF is provided, the load happens from that offset relative to
6220  * offset of NF.
6221  */
6222 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF)          \
6223         do {                                                                   \
6224                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg,     \
6225                                       si->src_reg, offsetof(S, F));            \
6226                 *insn++ = BPF_LDX_MEM(                                         \
6227                         SIZE, si->dst_reg, si->dst_reg,                        \
6228                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6229                                        target_size)                            \
6230                                 + OFF);                                        \
6231         } while (0)
6232
6233 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF)                              \
6234         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF,                     \
6235                                              BPF_FIELD_SIZEOF(NS, NF), 0)
6236
6237 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to
6238  * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation.
6239  *
6240  * It doesn't support SIZE argument though since narrow stores are not
6241  * supported for now.
6242  *
6243  * In addition it uses Temporary Field TF (member of struct S) as the 3rd
6244  * "register" since two registers available in convert_ctx_access are not
6245  * enough: we can't override neither SRC, since it contains value to store, nor
6246  * DST since it contains pointer to context that may be used by later
6247  * instructions. But we need a temporary place to save pointer to nested
6248  * structure whose field we want to store to.
6249  */
6250 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF, TF)                \
6251         do {                                                                   \
6252                 int tmp_reg = BPF_REG_9;                                       \
6253                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6254                         --tmp_reg;                                             \
6255                 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg)          \
6256                         --tmp_reg;                                             \
6257                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg,            \
6258                                       offsetof(S, TF));                        \
6259                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg,         \
6260                                       si->dst_reg, offsetof(S, F));            \
6261                 *insn++ = BPF_STX_MEM(                                         \
6262                         BPF_FIELD_SIZEOF(NS, NF), tmp_reg, si->src_reg,        \
6263                         bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF),           \
6264                                        target_size)                            \
6265                                 + OFF);                                        \
6266                 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg,            \
6267                                       offsetof(S, TF));                        \
6268         } while (0)
6269
6270 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \
6271                                                       TF)                      \
6272         do {                                                                   \
6273                 if (type == BPF_WRITE) {                                       \
6274                         SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, OFF,    \
6275                                                          TF);                  \
6276                 } else {                                                       \
6277                         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(                  \
6278                                 S, NS, F, NF, SIZE, OFF);  \
6279                 }                                                              \
6280         } while (0)
6281
6282 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF)                 \
6283         SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(                         \
6284                 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF)
6285
6286 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type,
6287                                         const struct bpf_insn *si,
6288                                         struct bpf_insn *insn_buf,
6289                                         struct bpf_prog *prog, u32 *target_size)
6290 {
6291         struct bpf_insn *insn = insn_buf;
6292         int off;
6293
6294         switch (si->off) {
6295         case offsetof(struct bpf_sock_addr, user_family):
6296                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6297                                             struct sockaddr, uaddr, sa_family);
6298                 break;
6299
6300         case offsetof(struct bpf_sock_addr, user_ip4):
6301                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6302                         struct bpf_sock_addr_kern, struct sockaddr_in, uaddr,
6303                         sin_addr, BPF_SIZE(si->code), 0, tmp_reg);
6304                 break;
6305
6306         case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]):
6307                 off = si->off;
6308                 off -= offsetof(struct bpf_sock_addr, user_ip6[0]);
6309                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6310                         struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr,
6311                         sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off,
6312                         tmp_reg);
6313                 break;
6314
6315         case offsetof(struct bpf_sock_addr, user_port):
6316                 /* To get port we need to know sa_family first and then treat
6317                  * sockaddr as either sockaddr_in or sockaddr_in6.
6318                  * Though we can simplify since port field has same offset and
6319                  * size in both structures.
6320                  * Here we check this invariant and use just one of the
6321                  * structures if it's true.
6322                  */
6323                 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) !=
6324                              offsetof(struct sockaddr_in6, sin6_port));
6325                 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) !=
6326                              FIELD_SIZEOF(struct sockaddr_in6, sin6_port));
6327                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern,
6328                                                      struct sockaddr_in6, uaddr,
6329                                                      sin6_port, tmp_reg);
6330                 break;
6331
6332         case offsetof(struct bpf_sock_addr, family):
6333                 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern,
6334                                             struct sock, sk, sk_family);
6335                 break;
6336
6337         case offsetof(struct bpf_sock_addr, type):
6338                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6339                         struct bpf_sock_addr_kern, struct sock, sk,
6340                         __sk_flags_offset, BPF_W, 0);
6341                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK);
6342                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT);
6343                 break;
6344
6345         case offsetof(struct bpf_sock_addr, protocol):
6346                 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(
6347                         struct bpf_sock_addr_kern, struct sock, sk,
6348                         __sk_flags_offset, BPF_W, 0);
6349                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
6350                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
6351                                         SK_FL_PROTO_SHIFT);
6352                 break;
6353
6354         case offsetof(struct bpf_sock_addr, msg_src_ip4):
6355                 /* Treat t_ctx as struct in_addr for msg_src_ip4. */
6356                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6357                         struct bpf_sock_addr_kern, struct in_addr, t_ctx,
6358                         s_addr, BPF_SIZE(si->code), 0, tmp_reg);
6359                 break;
6360
6361         case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0],
6362                                 msg_src_ip6[3]):
6363                 off = si->off;
6364                 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]);
6365                 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */
6366                 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(
6367                         struct bpf_sock_addr_kern, struct in6_addr, t_ctx,
6368                         s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg);
6369                 break;
6370         }
6371
6372         return insn - insn_buf;
6373 }
6374
6375 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
6376                                        const struct bpf_insn *si,
6377                                        struct bpf_insn *insn_buf,
6378                                        struct bpf_prog *prog,
6379                                        u32 *target_size)
6380 {
6381         struct bpf_insn *insn = insn_buf;
6382         int off;
6383
6384         switch (si->off) {
6385         case offsetof(struct bpf_sock_ops, op) ...
6386              offsetof(struct bpf_sock_ops, replylong[3]):
6387                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) !=
6388                              FIELD_SIZEOF(struct bpf_sock_ops_kern, op));
6389                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) !=
6390                              FIELD_SIZEOF(struct bpf_sock_ops_kern, reply));
6391                 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) !=
6392                              FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong));
6393                 off = si->off;
6394                 off -= offsetof(struct bpf_sock_ops, op);
6395                 off += offsetof(struct bpf_sock_ops_kern, op);
6396                 if (type == BPF_WRITE)
6397                         *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg,
6398                                               off);
6399                 else
6400                         *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg,
6401                                               off);
6402                 break;
6403
6404         case offsetof(struct bpf_sock_ops, family):
6405                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6406
6407                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6408                                               struct bpf_sock_ops_kern, sk),
6409                                       si->dst_reg, si->src_reg,
6410                                       offsetof(struct bpf_sock_ops_kern, sk));
6411                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6412                                       offsetof(struct sock_common, skc_family));
6413                 break;
6414
6415         case offsetof(struct bpf_sock_ops, remote_ip4):
6416                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6417
6418                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6419                                                 struct bpf_sock_ops_kern, sk),
6420                                       si->dst_reg, si->src_reg,
6421                                       offsetof(struct bpf_sock_ops_kern, sk));
6422                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6423                                       offsetof(struct sock_common, skc_daddr));
6424                 break;
6425
6426         case offsetof(struct bpf_sock_ops, local_ip4):
6427                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6428                                           skc_rcv_saddr) != 4);
6429
6430                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6431                                               struct bpf_sock_ops_kern, sk),
6432                                       si->dst_reg, si->src_reg,
6433                                       offsetof(struct bpf_sock_ops_kern, sk));
6434                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6435                                       offsetof(struct sock_common,
6436                                                skc_rcv_saddr));
6437                 break;
6438
6439         case offsetof(struct bpf_sock_ops, remote_ip6[0]) ...
6440              offsetof(struct bpf_sock_ops, remote_ip6[3]):
6441 #if IS_ENABLED(CONFIG_IPV6)
6442                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6443                                           skc_v6_daddr.s6_addr32[0]) != 4);
6444
6445                 off = si->off;
6446                 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]);
6447                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6448                                                 struct bpf_sock_ops_kern, sk),
6449                                       si->dst_reg, si->src_reg,
6450                                       offsetof(struct bpf_sock_ops_kern, sk));
6451                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6452                                       offsetof(struct sock_common,
6453                                                skc_v6_daddr.s6_addr32[0]) +
6454                                       off);
6455 #else
6456                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6457 #endif
6458                 break;
6459
6460         case offsetof(struct bpf_sock_ops, local_ip6[0]) ...
6461              offsetof(struct bpf_sock_ops, local_ip6[3]):
6462 #if IS_ENABLED(CONFIG_IPV6)
6463                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6464                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6465
6466                 off = si->off;
6467                 off -= offsetof(struct bpf_sock_ops, local_ip6[0]);
6468                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6469                                                 struct bpf_sock_ops_kern, sk),
6470                                       si->dst_reg, si->src_reg,
6471                                       offsetof(struct bpf_sock_ops_kern, sk));
6472                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6473                                       offsetof(struct sock_common,
6474                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6475                                       off);
6476 #else
6477                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6478 #endif
6479                 break;
6480
6481         case offsetof(struct bpf_sock_ops, remote_port):
6482                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6483
6484                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6485                                                 struct bpf_sock_ops_kern, sk),
6486                                       si->dst_reg, si->src_reg,
6487                                       offsetof(struct bpf_sock_ops_kern, sk));
6488                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6489                                       offsetof(struct sock_common, skc_dport));
6490 #ifndef __BIG_ENDIAN_BITFIELD
6491                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6492 #endif
6493                 break;
6494
6495         case offsetof(struct bpf_sock_ops, local_port):
6496                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6497
6498                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6499                                                 struct bpf_sock_ops_kern, sk),
6500                                       si->dst_reg, si->src_reg,
6501                                       offsetof(struct bpf_sock_ops_kern, sk));
6502                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6503                                       offsetof(struct sock_common, skc_num));
6504                 break;
6505
6506         case offsetof(struct bpf_sock_ops, is_fullsock):
6507                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6508                                                 struct bpf_sock_ops_kern,
6509                                                 is_fullsock),
6510                                       si->dst_reg, si->src_reg,
6511                                       offsetof(struct bpf_sock_ops_kern,
6512                                                is_fullsock));
6513                 break;
6514
6515         case offsetof(struct bpf_sock_ops, state):
6516                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1);
6517
6518                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6519                                                 struct bpf_sock_ops_kern, sk),
6520                                       si->dst_reg, si->src_reg,
6521                                       offsetof(struct bpf_sock_ops_kern, sk));
6522                 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg,
6523                                       offsetof(struct sock_common, skc_state));
6524                 break;
6525
6526         case offsetof(struct bpf_sock_ops, rtt_min):
6527                 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) !=
6528                              sizeof(struct minmax));
6529                 BUILD_BUG_ON(sizeof(struct minmax) <
6530                              sizeof(struct minmax_sample));
6531
6532                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6533                                                 struct bpf_sock_ops_kern, sk),
6534                                       si->dst_reg, si->src_reg,
6535                                       offsetof(struct bpf_sock_ops_kern, sk));
6536                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6537                                       offsetof(struct tcp_sock, rtt_min) +
6538                                       FIELD_SIZEOF(struct minmax_sample, t));
6539                 break;
6540
6541 /* Helper macro for adding read access to tcp_sock or sock fields. */
6542 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6543         do {                                                                  \
6544                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6545                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6546                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6547                                                 struct bpf_sock_ops_kern,     \
6548                                                 is_fullsock),                 \
6549                                       si->dst_reg, si->src_reg,               \
6550                                       offsetof(struct bpf_sock_ops_kern,      \
6551                                                is_fullsock));                 \
6552                 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2);            \
6553                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6554                                                 struct bpf_sock_ops_kern, sk),\
6555                                       si->dst_reg, si->src_reg,               \
6556                                       offsetof(struct bpf_sock_ops_kern, sk));\
6557                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ,                   \
6558                                                        OBJ_FIELD),            \
6559                                       si->dst_reg, si->dst_reg,               \
6560                                       offsetof(OBJ, OBJ_FIELD));              \
6561         } while (0)
6562
6563 /* Helper macro for adding write access to tcp_sock or sock fields.
6564  * The macro is called with two registers, dst_reg which contains a pointer
6565  * to ctx (context) and src_reg which contains the value that should be
6566  * stored. However, we need an additional register since we cannot overwrite
6567  * dst_reg because it may be used later in the program.
6568  * Instead we "borrow" one of the other register. We first save its value
6569  * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore
6570  * it at the end of the macro.
6571  */
6572 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ)                         \
6573         do {                                                                  \
6574                 int reg = BPF_REG_9;                                          \
6575                 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) >                   \
6576                              FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD));   \
6577                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6578                         reg--;                                                \
6579                 if (si->dst_reg == reg || si->src_reg == reg)                 \
6580                         reg--;                                                \
6581                 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg,               \
6582                                       offsetof(struct bpf_sock_ops_kern,      \
6583                                                temp));                        \
6584                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6585                                                 struct bpf_sock_ops_kern,     \
6586                                                 is_fullsock),                 \
6587                                       reg, si->dst_reg,                       \
6588                                       offsetof(struct bpf_sock_ops_kern,      \
6589                                                is_fullsock));                 \
6590                 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2);                    \
6591                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(                       \
6592                                                 struct bpf_sock_ops_kern, sk),\
6593                                       reg, si->dst_reg,                       \
6594                                       offsetof(struct bpf_sock_ops_kern, sk));\
6595                 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD),       \
6596                                       reg, si->src_reg,                       \
6597                                       offsetof(OBJ, OBJ_FIELD));              \
6598                 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg,               \
6599                                       offsetof(struct bpf_sock_ops_kern,      \
6600                                                temp));                        \
6601         } while (0)
6602
6603 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE)            \
6604         do {                                                                  \
6605                 if (TYPE == BPF_WRITE)                                        \
6606                         SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6607                 else                                                          \
6608                         SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ);        \
6609         } while (0)
6610
6611         case offsetof(struct bpf_sock_ops, snd_cwnd):
6612                 SOCK_OPS_GET_FIELD(snd_cwnd, snd_cwnd, struct tcp_sock);
6613                 break;
6614
6615         case offsetof(struct bpf_sock_ops, srtt_us):
6616                 SOCK_OPS_GET_FIELD(srtt_us, srtt_us, struct tcp_sock);
6617                 break;
6618
6619         case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags):
6620                 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags,
6621                                    struct tcp_sock);
6622                 break;
6623
6624         case offsetof(struct bpf_sock_ops, snd_ssthresh):
6625                 SOCK_OPS_GET_FIELD(snd_ssthresh, snd_ssthresh, struct tcp_sock);
6626                 break;
6627
6628         case offsetof(struct bpf_sock_ops, rcv_nxt):
6629                 SOCK_OPS_GET_FIELD(rcv_nxt, rcv_nxt, struct tcp_sock);
6630                 break;
6631
6632         case offsetof(struct bpf_sock_ops, snd_nxt):
6633                 SOCK_OPS_GET_FIELD(snd_nxt, snd_nxt, struct tcp_sock);
6634                 break;
6635
6636         case offsetof(struct bpf_sock_ops, snd_una):
6637                 SOCK_OPS_GET_FIELD(snd_una, snd_una, struct tcp_sock);
6638                 break;
6639
6640         case offsetof(struct bpf_sock_ops, mss_cache):
6641                 SOCK_OPS_GET_FIELD(mss_cache, mss_cache, struct tcp_sock);
6642                 break;
6643
6644         case offsetof(struct bpf_sock_ops, ecn_flags):
6645                 SOCK_OPS_GET_FIELD(ecn_flags, ecn_flags, struct tcp_sock);
6646                 break;
6647
6648         case offsetof(struct bpf_sock_ops, rate_delivered):
6649                 SOCK_OPS_GET_FIELD(rate_delivered, rate_delivered,
6650                                    struct tcp_sock);
6651                 break;
6652
6653         case offsetof(struct bpf_sock_ops, rate_interval_us):
6654                 SOCK_OPS_GET_FIELD(rate_interval_us, rate_interval_us,
6655                                    struct tcp_sock);
6656                 break;
6657
6658         case offsetof(struct bpf_sock_ops, packets_out):
6659                 SOCK_OPS_GET_FIELD(packets_out, packets_out, struct tcp_sock);
6660                 break;
6661
6662         case offsetof(struct bpf_sock_ops, retrans_out):
6663                 SOCK_OPS_GET_FIELD(retrans_out, retrans_out, struct tcp_sock);
6664                 break;
6665
6666         case offsetof(struct bpf_sock_ops, total_retrans):
6667                 SOCK_OPS_GET_FIELD(total_retrans, total_retrans,
6668                                    struct tcp_sock);
6669                 break;
6670
6671         case offsetof(struct bpf_sock_ops, segs_in):
6672                 SOCK_OPS_GET_FIELD(segs_in, segs_in, struct tcp_sock);
6673                 break;
6674
6675         case offsetof(struct bpf_sock_ops, data_segs_in):
6676                 SOCK_OPS_GET_FIELD(data_segs_in, data_segs_in, struct tcp_sock);
6677                 break;
6678
6679         case offsetof(struct bpf_sock_ops, segs_out):
6680                 SOCK_OPS_GET_FIELD(segs_out, segs_out, struct tcp_sock);
6681                 break;
6682
6683         case offsetof(struct bpf_sock_ops, data_segs_out):
6684                 SOCK_OPS_GET_FIELD(data_segs_out, data_segs_out,
6685                                    struct tcp_sock);
6686                 break;
6687
6688         case offsetof(struct bpf_sock_ops, lost_out):
6689                 SOCK_OPS_GET_FIELD(lost_out, lost_out, struct tcp_sock);
6690                 break;
6691
6692         case offsetof(struct bpf_sock_ops, sacked_out):
6693                 SOCK_OPS_GET_FIELD(sacked_out, sacked_out, struct tcp_sock);
6694                 break;
6695
6696         case offsetof(struct bpf_sock_ops, sk_txhash):
6697                 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash,
6698                                           struct sock, type);
6699                 break;
6700
6701         case offsetof(struct bpf_sock_ops, bytes_received):
6702                 SOCK_OPS_GET_FIELD(bytes_received, bytes_received,
6703                                    struct tcp_sock);
6704                 break;
6705
6706         case offsetof(struct bpf_sock_ops, bytes_acked):
6707                 SOCK_OPS_GET_FIELD(bytes_acked, bytes_acked, struct tcp_sock);
6708                 break;
6709
6710         }
6711         return insn - insn_buf;
6712 }
6713
6714 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type,
6715                                      const struct bpf_insn *si,
6716                                      struct bpf_insn *insn_buf,
6717                                      struct bpf_prog *prog, u32 *target_size)
6718 {
6719         struct bpf_insn *insn = insn_buf;
6720         int off;
6721
6722         switch (si->off) {
6723         case offsetof(struct __sk_buff, data_end):
6724                 off  = si->off;
6725                 off -= offsetof(struct __sk_buff, data_end);
6726                 off += offsetof(struct sk_buff, cb);
6727                 off += offsetof(struct tcp_skb_cb, bpf.data_end);
6728                 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg,
6729                                       si->src_reg, off);
6730                 break;
6731         default:
6732                 return bpf_convert_ctx_access(type, si, insn_buf, prog,
6733                                               target_size);
6734         }
6735
6736         return insn - insn_buf;
6737 }
6738
6739 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type,
6740                                      const struct bpf_insn *si,
6741                                      struct bpf_insn *insn_buf,
6742                                      struct bpf_prog *prog, u32 *target_size)
6743 {
6744         struct bpf_insn *insn = insn_buf;
6745 #if IS_ENABLED(CONFIG_IPV6)
6746         int off;
6747 #endif
6748
6749         switch (si->off) {
6750         case offsetof(struct sk_msg_md, data):
6751                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data),
6752                                       si->dst_reg, si->src_reg,
6753                                       offsetof(struct sk_msg_buff, data));
6754                 break;
6755         case offsetof(struct sk_msg_md, data_end):
6756                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_buff, data_end),
6757                                       si->dst_reg, si->src_reg,
6758                                       offsetof(struct sk_msg_buff, data_end));
6759                 break;
6760         case offsetof(struct sk_msg_md, family):
6761                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2);
6762
6763                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6764                                               struct sk_msg_buff, sk),
6765                                       si->dst_reg, si->src_reg,
6766                                       offsetof(struct sk_msg_buff, sk));
6767                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6768                                       offsetof(struct sock_common, skc_family));
6769                 break;
6770
6771         case offsetof(struct sk_msg_md, remote_ip4):
6772                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4);
6773
6774                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6775                                                 struct sk_msg_buff, sk),
6776                                       si->dst_reg, si->src_reg,
6777                                       offsetof(struct sk_msg_buff, sk));
6778                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6779                                       offsetof(struct sock_common, skc_daddr));
6780                 break;
6781
6782         case offsetof(struct sk_msg_md, local_ip4):
6783                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6784                                           skc_rcv_saddr) != 4);
6785
6786                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6787                                               struct sk_msg_buff, sk),
6788                                       si->dst_reg, si->src_reg,
6789                                       offsetof(struct sk_msg_buff, sk));
6790                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6791                                       offsetof(struct sock_common,
6792                                                skc_rcv_saddr));
6793                 break;
6794
6795         case offsetof(struct sk_msg_md, remote_ip6[0]) ...
6796              offsetof(struct sk_msg_md, remote_ip6[3]):
6797 #if IS_ENABLED(CONFIG_IPV6)
6798                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6799                                           skc_v6_daddr.s6_addr32[0]) != 4);
6800
6801                 off = si->off;
6802                 off -= offsetof(struct sk_msg_md, remote_ip6[0]);
6803                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6804                                                 struct sk_msg_buff, sk),
6805                                       si->dst_reg, si->src_reg,
6806                                       offsetof(struct sk_msg_buff, sk));
6807                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6808                                       offsetof(struct sock_common,
6809                                                skc_v6_daddr.s6_addr32[0]) +
6810                                       off);
6811 #else
6812                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6813 #endif
6814                 break;
6815
6816         case offsetof(struct sk_msg_md, local_ip6[0]) ...
6817              offsetof(struct sk_msg_md, local_ip6[3]):
6818 #if IS_ENABLED(CONFIG_IPV6)
6819                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common,
6820                                           skc_v6_rcv_saddr.s6_addr32[0]) != 4);
6821
6822                 off = si->off;
6823                 off -= offsetof(struct sk_msg_md, local_ip6[0]);
6824                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6825                                                 struct sk_msg_buff, sk),
6826                                       si->dst_reg, si->src_reg,
6827                                       offsetof(struct sk_msg_buff, sk));
6828                 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
6829                                       offsetof(struct sock_common,
6830                                                skc_v6_rcv_saddr.s6_addr32[0]) +
6831                                       off);
6832 #else
6833                 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
6834 #endif
6835                 break;
6836
6837         case offsetof(struct sk_msg_md, remote_port):
6838                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2);
6839
6840                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6841                                                 struct sk_msg_buff, sk),
6842                                       si->dst_reg, si->src_reg,
6843                                       offsetof(struct sk_msg_buff, sk));
6844                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6845                                       offsetof(struct sock_common, skc_dport));
6846 #ifndef __BIG_ENDIAN_BITFIELD
6847                 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16);
6848 #endif
6849                 break;
6850
6851         case offsetof(struct sk_msg_md, local_port):
6852                 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2);
6853
6854                 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
6855                                                 struct sk_msg_buff, sk),
6856                                       si->dst_reg, si->src_reg,
6857                                       offsetof(struct sk_msg_buff, sk));
6858                 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg,
6859                                       offsetof(struct sock_common, skc_num));
6860                 break;
6861         }
6862
6863         return insn - insn_buf;
6864 }
6865
6866 const struct bpf_verifier_ops sk_filter_verifier_ops = {
6867         .get_func_proto         = sk_filter_func_proto,
6868         .is_valid_access        = sk_filter_is_valid_access,
6869         .convert_ctx_access     = bpf_convert_ctx_access,
6870         .gen_ld_abs             = bpf_gen_ld_abs,
6871 };
6872
6873 const struct bpf_prog_ops sk_filter_prog_ops = {
6874         .test_run               = bpf_prog_test_run_skb,
6875 };
6876
6877 const struct bpf_verifier_ops tc_cls_act_verifier_ops = {
6878         .get_func_proto         = tc_cls_act_func_proto,
6879         .is_valid_access        = tc_cls_act_is_valid_access,
6880         .convert_ctx_access     = tc_cls_act_convert_ctx_access,
6881         .gen_prologue           = tc_cls_act_prologue,
6882         .gen_ld_abs             = bpf_gen_ld_abs,
6883 };
6884
6885 const struct bpf_prog_ops tc_cls_act_prog_ops = {
6886         .test_run               = bpf_prog_test_run_skb,
6887 };
6888
6889 const struct bpf_verifier_ops xdp_verifier_ops = {
6890         .get_func_proto         = xdp_func_proto,
6891         .is_valid_access        = xdp_is_valid_access,
6892         .convert_ctx_access     = xdp_convert_ctx_access,
6893 };
6894
6895 const struct bpf_prog_ops xdp_prog_ops = {
6896         .test_run               = bpf_prog_test_run_xdp,
6897 };
6898
6899 const struct bpf_verifier_ops cg_skb_verifier_ops = {
6900         .get_func_proto         = cg_skb_func_proto,
6901         .is_valid_access        = sk_filter_is_valid_access,
6902         .convert_ctx_access     = bpf_convert_ctx_access,
6903 };
6904
6905 const struct bpf_prog_ops cg_skb_prog_ops = {
6906         .test_run               = bpf_prog_test_run_skb,
6907 };
6908
6909 const struct bpf_verifier_ops lwt_in_verifier_ops = {
6910         .get_func_proto         = lwt_in_func_proto,
6911         .is_valid_access        = lwt_is_valid_access,
6912         .convert_ctx_access     = bpf_convert_ctx_access,
6913 };
6914
6915 const struct bpf_prog_ops lwt_in_prog_ops = {
6916         .test_run               = bpf_prog_test_run_skb,
6917 };
6918
6919 const struct bpf_verifier_ops lwt_out_verifier_ops = {
6920         .get_func_proto         = lwt_out_func_proto,
6921         .is_valid_access        = lwt_is_valid_access,
6922         .convert_ctx_access     = bpf_convert_ctx_access,
6923 };
6924
6925 const struct bpf_prog_ops lwt_out_prog_ops = {
6926         .test_run               = bpf_prog_test_run_skb,
6927 };
6928
6929 const struct bpf_verifier_ops lwt_xmit_verifier_ops = {
6930         .get_func_proto         = lwt_xmit_func_proto,
6931         .is_valid_access        = lwt_is_valid_access,
6932         .convert_ctx_access     = bpf_convert_ctx_access,
6933         .gen_prologue           = tc_cls_act_prologue,
6934 };
6935
6936 const struct bpf_prog_ops lwt_xmit_prog_ops = {
6937         .test_run               = bpf_prog_test_run_skb,
6938 };
6939
6940 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = {
6941         .get_func_proto         = lwt_seg6local_func_proto,
6942         .is_valid_access        = lwt_is_valid_access,
6943         .convert_ctx_access     = bpf_convert_ctx_access,
6944 };
6945
6946 const struct bpf_prog_ops lwt_seg6local_prog_ops = {
6947         .test_run               = bpf_prog_test_run_skb,
6948 };
6949
6950 const struct bpf_verifier_ops cg_sock_verifier_ops = {
6951         .get_func_proto         = sock_filter_func_proto,
6952         .is_valid_access        = sock_filter_is_valid_access,
6953         .convert_ctx_access     = sock_filter_convert_ctx_access,
6954 };
6955
6956 const struct bpf_prog_ops cg_sock_prog_ops = {
6957 };
6958
6959 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = {
6960         .get_func_proto         = sock_addr_func_proto,
6961         .is_valid_access        = sock_addr_is_valid_access,
6962         .convert_ctx_access     = sock_addr_convert_ctx_access,
6963 };
6964
6965 const struct bpf_prog_ops cg_sock_addr_prog_ops = {
6966 };
6967
6968 const struct bpf_verifier_ops sock_ops_verifier_ops = {
6969         .get_func_proto         = sock_ops_func_proto,
6970         .is_valid_access        = sock_ops_is_valid_access,
6971         .convert_ctx_access     = sock_ops_convert_ctx_access,
6972 };
6973
6974 const struct bpf_prog_ops sock_ops_prog_ops = {
6975 };
6976
6977 const struct bpf_verifier_ops sk_skb_verifier_ops = {
6978         .get_func_proto         = sk_skb_func_proto,
6979         .is_valid_access        = sk_skb_is_valid_access,
6980         .convert_ctx_access     = sk_skb_convert_ctx_access,
6981         .gen_prologue           = sk_skb_prologue,
6982 };
6983
6984 const struct bpf_prog_ops sk_skb_prog_ops = {
6985 };
6986
6987 const struct bpf_verifier_ops sk_msg_verifier_ops = {
6988         .get_func_proto         = sk_msg_func_proto,
6989         .is_valid_access        = sk_msg_is_valid_access,
6990         .convert_ctx_access     = sk_msg_convert_ctx_access,
6991 };
6992
6993 const struct bpf_prog_ops sk_msg_prog_ops = {
6994 };
6995
6996 int sk_detach_filter(struct sock *sk)
6997 {
6998         int ret = -ENOENT;
6999         struct sk_filter *filter;
7000
7001         if (sock_flag(sk, SOCK_FILTER_LOCKED))
7002                 return -EPERM;
7003
7004         filter = rcu_dereference_protected(sk->sk_filter,
7005                                            lockdep_sock_is_held(sk));
7006         if (filter) {
7007                 RCU_INIT_POINTER(sk->sk_filter, NULL);
7008                 sk_filter_uncharge(sk, filter);
7009                 ret = 0;
7010         }
7011
7012         return ret;
7013 }
7014 EXPORT_SYMBOL_GPL(sk_detach_filter);
7015
7016 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf,
7017                   unsigned int len)
7018 {
7019         struct sock_fprog_kern *fprog;
7020         struct sk_filter *filter;
7021         int ret = 0;
7022
7023         lock_sock(sk);
7024         filter = rcu_dereference_protected(sk->sk_filter,
7025                                            lockdep_sock_is_held(sk));
7026         if (!filter)
7027                 goto out;
7028
7029         /* We're copying the filter that has been originally attached,
7030          * so no conversion/decode needed anymore. eBPF programs that
7031          * have no original program cannot be dumped through this.
7032          */
7033         ret = -EACCES;
7034         fprog = filter->prog->orig_prog;
7035         if (!fprog)
7036                 goto out;
7037
7038         ret = fprog->len;
7039         if (!len)
7040                 /* User space only enquires number of filter blocks. */
7041                 goto out;
7042
7043         ret = -EINVAL;
7044         if (len < fprog->len)
7045                 goto out;
7046
7047         ret = -EFAULT;
7048         if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog)))
7049                 goto out;
7050
7051         /* Instead of bytes, the API requests to return the number
7052          * of filter blocks.
7053          */
7054         ret = fprog->len;
7055 out:
7056         release_sock(sk);
7057         return ret;
7058 }
7059
7060 #ifdef CONFIG_INET
7061 struct sk_reuseport_kern {
7062         struct sk_buff *skb;
7063         struct sock *sk;
7064         struct sock *selected_sk;
7065         void *data_end;
7066         u32 hash;
7067         u32 reuseport_id;
7068         bool bind_inany;
7069 };
7070
7071 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern,
7072                                     struct sock_reuseport *reuse,
7073                                     struct sock *sk, struct sk_buff *skb,
7074                                     u32 hash)
7075 {
7076         reuse_kern->skb = skb;
7077         reuse_kern->sk = sk;
7078         reuse_kern->selected_sk = NULL;
7079         reuse_kern->data_end = skb->data + skb_headlen(skb);
7080         reuse_kern->hash = hash;
7081         reuse_kern->reuseport_id = reuse->reuseport_id;
7082         reuse_kern->bind_inany = reuse->bind_inany;
7083 }
7084
7085 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk,
7086                                   struct bpf_prog *prog, struct sk_buff *skb,
7087                                   u32 hash)
7088 {
7089         struct sk_reuseport_kern reuse_kern;
7090         enum sk_action action;
7091
7092         bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash);
7093         action = BPF_PROG_RUN(prog, &reuse_kern);
7094
7095         if (action == SK_PASS)
7096                 return reuse_kern.selected_sk;
7097         else
7098                 return ERR_PTR(-ECONNREFUSED);
7099 }
7100
7101 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern,
7102            struct bpf_map *, map, void *, key, u32, flags)
7103 {
7104         struct sock_reuseport *reuse;
7105         struct sock *selected_sk;
7106
7107         selected_sk = map->ops->map_lookup_elem(map, key);
7108         if (!selected_sk)
7109                 return -ENOENT;
7110
7111         reuse = rcu_dereference(selected_sk->sk_reuseport_cb);
7112         if (!reuse)
7113                 /* selected_sk is unhashed (e.g. by close()) after the
7114                  * above map_lookup_elem().  Treat selected_sk has already
7115                  * been removed from the map.
7116                  */
7117                 return -ENOENT;
7118
7119         if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) {
7120                 struct sock *sk;
7121
7122                 if (unlikely(!reuse_kern->reuseport_id))
7123                         /* There is a small race between adding the
7124                          * sk to the map and setting the
7125                          * reuse_kern->reuseport_id.
7126                          * Treat it as the sk has not been added to
7127                          * the bpf map yet.
7128                          */
7129                         return -ENOENT;
7130
7131                 sk = reuse_kern->sk;
7132                 if (sk->sk_protocol != selected_sk->sk_protocol)
7133                         return -EPROTOTYPE;
7134                 else if (sk->sk_family != selected_sk->sk_family)
7135                         return -EAFNOSUPPORT;
7136
7137                 /* Catch all. Likely bound to a different sockaddr. */
7138                 return -EBADFD;
7139         }
7140
7141         reuse_kern->selected_sk = selected_sk;
7142
7143         return 0;
7144 }
7145
7146 static const struct bpf_func_proto sk_select_reuseport_proto = {
7147         .func           = sk_select_reuseport,
7148         .gpl_only       = false,
7149         .ret_type       = RET_INTEGER,
7150         .arg1_type      = ARG_PTR_TO_CTX,
7151         .arg2_type      = ARG_CONST_MAP_PTR,
7152         .arg3_type      = ARG_PTR_TO_MAP_KEY,
7153         .arg4_type      = ARG_ANYTHING,
7154 };
7155
7156 BPF_CALL_4(sk_reuseport_load_bytes,
7157            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7158            void *, to, u32, len)
7159 {
7160         return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len);
7161 }
7162
7163 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = {
7164         .func           = sk_reuseport_load_bytes,
7165         .gpl_only       = false,
7166         .ret_type       = RET_INTEGER,
7167         .arg1_type      = ARG_PTR_TO_CTX,
7168         .arg2_type      = ARG_ANYTHING,
7169         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7170         .arg4_type      = ARG_CONST_SIZE,
7171 };
7172
7173 BPF_CALL_5(sk_reuseport_load_bytes_relative,
7174            const struct sk_reuseport_kern *, reuse_kern, u32, offset,
7175            void *, to, u32, len, u32, start_header)
7176 {
7177         return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to,
7178                                                len, start_header);
7179 }
7180
7181 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = {
7182         .func           = sk_reuseport_load_bytes_relative,
7183         .gpl_only       = false,
7184         .ret_type       = RET_INTEGER,
7185         .arg1_type      = ARG_PTR_TO_CTX,
7186         .arg2_type      = ARG_ANYTHING,
7187         .arg3_type      = ARG_PTR_TO_UNINIT_MEM,
7188         .arg4_type      = ARG_CONST_SIZE,
7189         .arg5_type      = ARG_ANYTHING,
7190 };
7191
7192 static const struct bpf_func_proto *
7193 sk_reuseport_func_proto(enum bpf_func_id func_id,
7194                         const struct bpf_prog *prog)
7195 {
7196         switch (func_id) {
7197         case BPF_FUNC_sk_select_reuseport:
7198                 return &sk_select_reuseport_proto;
7199         case BPF_FUNC_skb_load_bytes:
7200                 return &sk_reuseport_load_bytes_proto;
7201         case BPF_FUNC_skb_load_bytes_relative:
7202                 return &sk_reuseport_load_bytes_relative_proto;
7203         default:
7204                 return bpf_base_func_proto(func_id);
7205         }
7206 }
7207
7208 static bool
7209 sk_reuseport_is_valid_access(int off, int size,
7210                              enum bpf_access_type type,
7211                              const struct bpf_prog *prog,
7212                              struct bpf_insn_access_aux *info)
7213 {
7214         const u32 size_default = sizeof(__u32);
7215
7216         if (off < 0 || off >= sizeof(struct sk_reuseport_md) ||
7217             off % size || type != BPF_READ)
7218                 return false;
7219
7220         switch (off) {
7221         case offsetof(struct sk_reuseport_md, data):
7222                 info->reg_type = PTR_TO_PACKET;
7223                 return size == sizeof(__u64);
7224
7225         case offsetof(struct sk_reuseport_md, data_end):
7226                 info->reg_type = PTR_TO_PACKET_END;
7227                 return size == sizeof(__u64);
7228
7229         case offsetof(struct sk_reuseport_md, hash):
7230                 return size == size_default;
7231
7232         /* Fields that allow narrowing */
7233         case offsetof(struct sk_reuseport_md, eth_protocol):
7234                 if (size < FIELD_SIZEOF(struct sk_buff, protocol))
7235                         return false;
7236                 /* fall through */
7237         case offsetof(struct sk_reuseport_md, ip_protocol):
7238         case offsetof(struct sk_reuseport_md, bind_inany):
7239         case offsetof(struct sk_reuseport_md, len):
7240                 bpf_ctx_record_field_size(info, size_default);
7241                 return bpf_ctx_narrow_access_ok(off, size, size_default);
7242
7243         default:
7244                 return false;
7245         }
7246 }
7247
7248 #define SK_REUSEPORT_LOAD_FIELD(F) ({                                   \
7249         *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7250                               si->dst_reg, si->src_reg,                 \
7251                               bpf_target_off(struct sk_reuseport_kern, F, \
7252                                              FIELD_SIZEOF(struct sk_reuseport_kern, F), \
7253                                              target_size));             \
7254         })
7255
7256 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD)                          \
7257         SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern,           \
7258                                     struct sk_buff,                     \
7259                                     skb,                                \
7260                                     SKB_FIELD)
7261
7262 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \
7263         SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern,  \
7264                                              struct sock,               \
7265                                              sk,                        \
7266                                              SK_FIELD, BPF_SIZE, EXTRA_OFF)
7267
7268 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type,
7269                                            const struct bpf_insn *si,
7270                                            struct bpf_insn *insn_buf,
7271                                            struct bpf_prog *prog,
7272                                            u32 *target_size)
7273 {
7274         struct bpf_insn *insn = insn_buf;
7275
7276         switch (si->off) {
7277         case offsetof(struct sk_reuseport_md, data):
7278                 SK_REUSEPORT_LOAD_SKB_FIELD(data);
7279                 break;
7280
7281         case offsetof(struct sk_reuseport_md, len):
7282                 SK_REUSEPORT_LOAD_SKB_FIELD(len);
7283                 break;
7284
7285         case offsetof(struct sk_reuseport_md, eth_protocol):
7286                 SK_REUSEPORT_LOAD_SKB_FIELD(protocol);
7287                 break;
7288
7289         case offsetof(struct sk_reuseport_md, ip_protocol):
7290                 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE);
7291                 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset,
7292                                                     BPF_W, 0);
7293                 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK);
7294                 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg,
7295                                         SK_FL_PROTO_SHIFT);
7296                 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian
7297                  * aware.  No further narrowing or masking is needed.
7298                  */
7299                 *target_size = 1;
7300                 break;
7301
7302         case offsetof(struct sk_reuseport_md, data_end):
7303                 SK_REUSEPORT_LOAD_FIELD(data_end);
7304                 break;
7305
7306         case offsetof(struct sk_reuseport_md, hash):
7307                 SK_REUSEPORT_LOAD_FIELD(hash);
7308                 break;
7309
7310         case offsetof(struct sk_reuseport_md, bind_inany):
7311                 SK_REUSEPORT_LOAD_FIELD(bind_inany);
7312                 break;
7313         }
7314
7315         return insn - insn_buf;
7316 }
7317
7318 const struct bpf_verifier_ops sk_reuseport_verifier_ops = {
7319         .get_func_proto         = sk_reuseport_func_proto,
7320         .is_valid_access        = sk_reuseport_is_valid_access,
7321         .convert_ctx_access     = sk_reuseport_convert_ctx_access,
7322 };
7323
7324 const struct bpf_prog_ops sk_reuseport_prog_ops = {
7325 };
7326 #endif /* CONFIG_INET */