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