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